Warning

This document is for an in-development version of Galaxy. You can alternatively view this page in the latest release if it exists or view the top of the latest release's documentation.

Source code for galaxy.model

"""
Galaxy data model classes

Naming: try to use class names that have a distinct plural form so that
the relationship cardinalities are obvious (e.g. prefer Dataset to Data)
"""

import abc
import base64
import errno
import json
import logging
import numbers
import operator
import os
import pwd
import random
import string
from collections import defaultdict
from collections.abc import Callable
from dataclasses import dataclass
from datetime import (
    datetime,
    timedelta,
)
from decimal import Decimal
from enum import Enum
from secrets import token_hex
from string import Template
from typing import (
    Any,
    cast,
    ClassVar,
    Dict,
    Generic,
    Iterable,
    List,
    NamedTuple,
    Optional,
    overload,
    Set,
    Tuple,
    Type,
    TYPE_CHECKING,
    TypeVar,
    Union,
)
from uuid import (
    UUID,
    uuid4,
)

import sqlalchemy
from boltons.iterutils import remap
from pydantic import BaseModel
from social_core.storage import (
    AssociationMixin,
    CodeMixin,
    NonceMixin,
    PartialMixin,
    UserMixin,
)
from sqlalchemy import (
    alias,
    and_,
    asc,
    BigInteger,
    bindparam,
    Boolean,
    case,
    Column,
    column,
    DateTime,
    delete,
    desc,
    event,
    false,
    ForeignKey,
    func,
    Index,
    inspect,
    Integer,
    join,
    MetaData,
    not_,
    Numeric,
    or_,
    PrimaryKeyConstraint,
    select,
    String,
    Table,
    TEXT,
    Text,
    text,
    true,
    tuple_,
    type_coerce,
    Unicode,
    UniqueConstraint,
    update,
    VARCHAR,
)
from sqlalchemy.exc import (
    CompileError,
    OperationalError,
)
from sqlalchemy.ext import hybrid
from sqlalchemy.ext.associationproxy import (
    association_proxy,
    AssociationProxy,
)
from sqlalchemy.ext.orderinglist import ordering_list
from sqlalchemy.orm import (
    aliased,
    column_property,
    DeclarativeBase,
    deferred,
    joinedload,
    Mapped,
    mapped_column,
    object_session,
    Query,
    reconstructor,
    registry,
    relationship,
)
from sqlalchemy.orm.attributes import flag_modified
from sqlalchemy.orm.collections import attribute_keyed_dict
from sqlalchemy.sql import exists
from sqlalchemy.sql.expression import FromClause
from typing_extensions import (
    Literal,
    Protocol,
    TypeAlias,
    TypedDict,
)

import galaxy.exceptions
import galaxy.model.metadata
import galaxy.model.tags
import galaxy.security.passwords
import galaxy.util
from galaxy.files.templates import (
    FileSourceConfiguration,
    FileSourceTemplate,
    template_to_configuration as file_source_template_to_configuration,
)
from galaxy.model.base import (
    ensure_object_added_to_session,
    transaction,
)
from galaxy.model.custom_types import (
    DoubleEncodedJsonType,
    JSONType,
    MetadataType,
    MutableJSONType,
    TrimmedString,
    UUIDType,
)
from galaxy.model.database_object_names import NAMING_CONVENTION
from galaxy.model.item_attrs import (
    get_item_annotation_str,
    UsesAnnotations,
)
from galaxy.model.orm.now import now
from galaxy.model.orm.util import add_object_to_object_session
from galaxy.objectstore.templates import (
    ObjectStoreConfiguration,
    ObjectStoreTemplate,
    template_to_configuration as object_store_template_to_configuration,
)
from galaxy.schema.invocation import (
    InvocationCancellationUserRequest,
    InvocationState,
    InvocationStepState,
)
from galaxy.schema.schema import (
    DatasetCollectionPopulatedState,
    DatasetState,
    DatasetValidatedState,
    InvocationsStateCounts,
    JobState,
    ToolRequestState,
)
from galaxy.schema.workflow.comments import WorkflowCommentModel
from galaxy.security import get_permitted_actions
from galaxy.security.idencoding import IdEncodingHelper
from galaxy.security.validate_user_input import validate_password_str
from galaxy.util import (
    directory_hash_id,
    enum_values,
    hex_to_lowercase_alphanum,
    listify,
    ready_name_for_url,
    unicodify,
    unique_id,
)
from galaxy.util.config_templates import (
    EnvironmentDict,
    ImplicitConfigurationParameters,
    SecretsDict,
    Template as ConfigTemplate,
    TemplateEnvironment,
)
from galaxy.util.dictifiable import (
    dict_for,
    Dictifiable,
)
from galaxy.util.form_builder import (
    AddressField,
    CheckboxField,
    HistoryField,
    PasswordField,
    SelectField,
    TextArea,
    TextField,
    WorkflowField,
    WorkflowMappingField,
)
from galaxy.util.hash_util import (
    HashFunctionNameEnum,
    md5_hash_str,
    new_insecure_hash,
)
from galaxy.util.json import safe_loads
from galaxy.util.sanitize_html import sanitize_html

if TYPE_CHECKING:
    from galaxy.objectstore import (
        BaseObjectStore,
        ObjectStorePopulator,
    )
    from galaxy.schema.invocation import InvocationMessageUnion

log = logging.getLogger(__name__)

_datatypes_registry = None

STR_TO_STR_DICT = Dict[str, str]


[docs]class ConfigurationTemplateEnvironmentSecret(TypedDict): name: str type: Literal["secret"] vault_key: str
[docs]class ConfigurationTemplateEnvironmentVariable(TypedDict): name: str type: Literal["variable"] variable: str
CONFIGURATION_TEMPLATE_ENVIRONMENT_ENTRY = Union[ ConfigurationTemplateEnvironmentSecret, ConfigurationTemplateEnvironmentVariable ] CONFIGURATION_TEMPLATE_ENVIRONMENT = List[CONFIGURATION_TEMPLATE_ENVIRONMENT_ENTRY] CONFIGURATION_TEMPLATE_CONFIGURATION_VALUE_TYPE = Union[str, bool, int] CONFIGURATION_TEMPLATE_CONFIGURATION_VARIABLES_TYPE = Dict[str, CONFIGURATION_TEMPLATE_CONFIGURATION_VALUE_TYPE] CONFIGURATION_TEMPLATE_CONFIGURATION_SECRET_NAMES_TYPE = List[str] CONFIGURATION_TEMPLATE_DEFINITION_TYPE = Dict[str, Any]
[docs]class TransformAction(TypedDict): action: str
TRANSFORM_ACTIONS = List[TransformAction] mapper_registry = registry( type_annotation_map={ Optional[STR_TO_STR_DICT]: JSONType, Optional[TRANSFORM_ACTIONS]: MutableJSONType, Optional[CONFIGURATION_TEMPLATE_CONFIGURATION_VARIABLES_TYPE]: JSONType, Optional[CONFIGURATION_TEMPLATE_CONFIGURATION_SECRET_NAMES_TYPE]: JSONType, Optional[CONFIGURATION_TEMPLATE_DEFINITION_TYPE]: JSONType, Optional[CONFIGURATION_TEMPLATE_ENVIRONMENT]: JSONType, }, ) # When constructing filters with in for a fixed set of ids, maximum # number of items to place in the IN statement. Different databases # are going to have different limits so it is likely best to not let # this be unlimited - filter in Python if over this limit. MAX_IN_FILTER_LENGTH = 100 # The column sizes for job metrics. Note: changing these values does not change the column sizes, a migration must be # performed to do that. JOB_METRIC_MAX_LENGTH = 1023 JOB_METRIC_PRECISION = 26 JOB_METRIC_SCALE = 7 # Tags that get automatically propagated from inputs to outputs when running jobs. AUTO_PROPAGATED_TAGS = ["name"] YIELD_PER_ROWS = 100 CANNOT_SHARE_PRIVATE_DATASET_MESSAGE = "Attempting to share a non-shareable dataset." if TYPE_CHECKING: from galaxy.datatypes.data import Data from galaxy.tools import DefaultToolState from galaxy.workflow.modules import WorkflowModule class _HasTable: table: FromClause __table__: FromClause else: _HasTable = object
[docs]def get_uuid(uuid: Optional[Union[UUID, str]] = None) -> UUID: if isinstance(uuid, UUID): return uuid if not uuid: return uuid4() return UUID(str(uuid))
[docs]class Base(DeclarativeBase, _HasTable): __abstract__ = True metadata = MetaData(naming_convention=NAMING_CONVENTION) mapper_registry.metadata = metadata registry = mapper_registry @classmethod def __declare_last__(cls): cls.table = cls.__table__
[docs]class RepresentById: id: Mapped[int] def __repr__(self): try: r = f"<galaxy.model.{self.__class__.__name__}({cached_id(self)}) at {hex(id(self))}>" except Exception: r = object.__repr__(self) log.exception("Caught exception attempting to generate repr for: %s", r) return r
[docs]class NoConverterException(Exception):
[docs] def __init__(self, value): self.value = value
def __str__(self): return repr(self.value)
[docs]class ConverterDependencyException(Exception):
[docs] def __init__(self, value): self.value = value
def __str__(self): return repr(self.value)
def _get_datatypes_registry(): if _datatypes_registry is None: raise Exception( "galaxy.model.set_datatypes_registry must be called before performing certain DatasetInstance operations." ) return _datatypes_registry
[docs]def set_datatypes_registry(d_registry): """ Set up datatypes_registry """ global _datatypes_registry _datatypes_registry = d_registry
[docs]class HasTags: dict_collection_visible_keys = ["tags"] dict_element_visible_keys = ["tags"]
[docs] def to_dict(self, *args, **kwargs): rval = super().to_dict(*args, **kwargs) rval["tags"] = self.make_tag_string_list() return rval
[docs] def make_tag_string_list(self): # add tags string list tags_str_list = [] for tag in self.tags: tag_str = tag.user_tname if tag.value is not None: tag_str += f":{tag.user_value}" tags_str_list.append(tag_str) return tags_str_list
[docs] def copy_tags_from(self, target_user, source): for source_tag_assoc in source.tags: new_tag_assoc = source_tag_assoc.copy() new_tag_assoc.user = target_user self.tags.append(new_tag_assoc)
@property def auto_propagated_tags(self): return [t for t in self.tags if t.user_tname in AUTO_PROPAGATED_TAGS]
[docs]class SerializeFilesHandler(Protocol):
[docs] def serialize_files(self, dataset: "DatasetInstance", as_dict: Dict[str, Any]) -> None: pass
[docs]class SerializationOptions:
[docs] def __init__( self, for_edit: bool, serialize_dataset_objects: Optional[bool] = None, serialize_files_handler: Optional[SerializeFilesHandler] = None, strip_metadata_files: Optional[bool] = None, ) -> None: self.for_edit = for_edit if serialize_dataset_objects is None: serialize_dataset_objects = for_edit self.serialize_dataset_objects = serialize_dataset_objects self.serialize_files_handler = serialize_files_handler if strip_metadata_files is None: # If we're editing datasets - keep MetadataFile(s) in tact. For pure export # expect metadata tool to be rerun. strip_metadata_files = not for_edit self.strip_metadata_files = strip_metadata_files
[docs] def attach_identifier(self, id_encoder, obj, ret_val): if self.for_edit and obj.id: ret_val["id"] = obj.id elif obj.id: ret_val["encoded_id"] = id_encoder.encode_id(obj.id, kind="model_export") else: if not hasattr(obj, "temp_id"): obj.temp_id = uuid4().hex ret_val["encoded_id"] = obj.temp_id
[docs] def get_identifier(self, id_encoder, obj): if self.for_edit and obj.id: return obj.id elif obj.id: return id_encoder.encode_id(obj.id, kind="model_export") else: if not hasattr(obj, "temp_id"): obj.temp_id = uuid4().hex return obj.temp_id
[docs] def get_identifier_for_id(self, id_encoder, obj_id): if self.for_edit and obj_id: return obj_id elif obj_id: return id_encoder.encode_id(obj_id, kind="model_export") else: raise NotImplementedError()
[docs] def serialize_files(self, dataset, as_dict): if self.serialize_files_handler is not None: self.serialize_files_handler.serialize_files(dataset, as_dict)
[docs]class Serializable(RepresentById):
[docs] def serialize( self, id_encoder: IdEncodingHelper, serialization_options: SerializationOptions, for_link: bool = False ) -> Dict[str, Any]: """Serialize model for a re-population in (potentially) another Galaxy instance.""" if for_link: rval = dict_for(self) serialization_options.attach_identifier(id_encoder, self, rval) return rval return self._serialize(id_encoder, serialization_options)
@abc.abstractmethod def _serialize(self, id_encoder: IdEncodingHelper, serialization_options: SerializationOptions) -> Dict[str, Any]: """Serialize model for a re-population in (potentially) another Galaxy instance."""
[docs]class HasName:
[docs] def get_display_name(self): """ These objects have a name attribute can be either a string or a unicode object. If string, convert to unicode object assuming 'utf-8' format. """ name = self.name name = unicodify(name, "utf-8") return name
[docs]class UsesCreateAndUpdateTime: update_time: Mapped[Optional[datetime]] @property def seconds_since_updated(self): update_time = self.update_time or now() # In case not yet flushed return (now() - update_time).total_seconds() @property def seconds_since_created(self): create_time = self.create_time or now() # In case not yet flushed return (now() - create_time).total_seconds()
[docs] def update(self): self.update_time = now()
[docs]class WorkerProcess(Base, UsesCreateAndUpdateTime): __tablename__ = "worker_process" __table_args__ = (UniqueConstraint("server_name", "hostname"),) id: Mapped[int] = mapped_column(primary_key=True) server_name: Mapped[Optional[str]] = mapped_column(String(255), index=True) hostname: Mapped[Optional[str]] = mapped_column(String(255)) pid: Mapped[Optional[int]] update_time: Mapped[Optional[datetime]] = mapped_column(default=now, onupdate=now)
[docs]def cached_id(galaxy_model_object): """Get model object id attribute without a firing a database query. Useful to fetching the id of a typical Galaxy model object after a flush, where SA is going to mark the id attribute as unloaded but we know the id is immutable and so we can use the database identity to fetch. With Galaxy's default SA initialization - any flush marks all attributes as unloaded - even objects completely unrelated to the flushed changes and even attributes we know to be immutable like id. See test_galaxy_mapping.py for verification of this behavior. This method is a workaround that uses the fact that we know all Galaxy objects use the id attribute as identity and SA internals (_sa_instance_state) to infer the previously loaded ID value. I tried digging into the SA internals extensively and couldn't find a way to get the previously loaded values after a flush to allow a generalization of this for other attributes. """ if hasattr(galaxy_model_object, "_sa_instance_state"): identity = galaxy_model_object._sa_instance_state.identity if identity: assert len(identity) == 1 return identity[0] return galaxy_model_object.id
[docs]class JobLike: MAX_NUMERIC = 10 ** (JOB_METRIC_PRECISION - JOB_METRIC_SCALE) - 1 def _init_metrics(self): self.text_metrics = [] self.numeric_metrics = []
[docs] def add_metric(self, plugin, metric_name, metric_value): plugin = unicodify(plugin, "utf-8") metric_name = unicodify(metric_name, "utf-8") number = isinstance(metric_value, numbers.Number) if number and int(metric_value) <= JobLike.MAX_NUMERIC: metric = self._numeric_metric(plugin, metric_name, metric_value) self.numeric_metrics.append(metric) elif number: log.warning( "Cannot store metric due to database column overflow (max: %s): %s: %s", JobLike.MAX_NUMERIC, metric_name, metric_value, ) else: metric_value = unicodify(metric_value, "utf-8") if len(metric_value) > (JOB_METRIC_MAX_LENGTH - 1): # Truncate these values - not needed with sqlite # but other backends must need it. metric_value = metric_value[: (JOB_METRIC_MAX_LENGTH - 1)] metric = self._text_metric(plugin, metric_name, metric_value) self.text_metrics.append(metric)
@property def metrics(self): # TODO: Make iterable, concatenate with chain return self.text_metrics + self.numeric_metrics
[docs] def set_streams(self, tool_stdout, tool_stderr, job_stdout=None, job_stderr=None, job_messages=None): def shrink_and_unicodify(what, stream): if stream and len(stream) > galaxy.util.DATABASE_MAX_STRING_SIZE: log.info( "%s for %s %d is greater than %s, only a portion will be logged to database", what, type(self), self.id, galaxy.util.DATABASE_MAX_STRING_SIZE_PRETTY, ) return galaxy.util.shrink_and_unicodify(stream) self.tool_stdout = shrink_and_unicodify("tool_stdout", tool_stdout) self.tool_stderr = shrink_and_unicodify("tool_stderr", tool_stderr) if job_stdout is not None: self.job_stdout = shrink_and_unicodify("job_stdout", job_stdout) else: self.job_stdout = None if job_stderr is not None: self.job_stderr = shrink_and_unicodify("job_stderr", job_stderr) else: self.job_stderr = None if job_messages is not None: self.job_messages = job_messages
[docs] def log_str(self): extra = "" if (safe_id := getattr(self, "id", None)) is not None: extra += f"id={safe_id}" else: extra += "unflushed" return f"{self.__class__.__name__}[{extra},tool_id={self.tool_id}]"
@property def stdout(self): stdout = self.tool_stdout or "" if self.job_stdout: stdout += f"\n{self.job_stdout}" return stdout @stdout.setter def stdout(self, stdout): raise NotImplementedError("Attempt to set stdout, must set tool_stdout or job_stdout") @property def stderr(self): stderr = self.tool_stderr or "" if self.job_stderr: stderr += f"\n{self.job_stderr}" return stderr @stderr.setter def stderr(self, stderr): raise NotImplementedError("Attempt to set stdout, must set tool_stderr or job_stderr")
UNIQUE_DATASET_USER_USAGE = """ WITH per_user_histories AS ( SELECT id FROM history WHERE user_id = :id AND NOT purged ), per_hist_hdas AS ( SELECT DISTINCT dataset_id FROM history_dataset_association WHERE NOT purged AND history_id IN (SELECT id FROM per_user_histories) ) SELECT COALESCE(SUM(COALESCE(dataset.total_size, dataset.file_size, 0)), 0) FROM dataset LEFT OUTER JOIN library_dataset_dataset_association ON dataset.id = library_dataset_dataset_association.dataset_id WHERE dataset.id IN (SELECT dataset_id FROM per_hist_hdas) AND library_dataset_dataset_association.id IS NULL {and_dataset_condition} """
[docs]def calculate_user_disk_usage_statements(user_id, quota_source_map, for_sqlite=False): """Standalone function so can be reused for postgres directly in pgcleanup.py.""" statements = [] default_quota_enabled = quota_source_map.default_quota_enabled default_exclude_ids = quota_source_map.default_usage_excluded_ids() default_cond = "dataset.object_store_id IS NULL" if default_quota_enabled and default_exclude_ids else "" exclude_cond = "dataset.object_store_id NOT IN :exclude_object_store_ids" if default_exclude_ids else "" use_or = " OR " if (default_cond != "" and exclude_cond != "") else "" default_usage_dataset_condition = f"{default_cond} {use_or} {exclude_cond}" if default_usage_dataset_condition.strip(): default_usage_dataset_condition = f"AND ( {default_usage_dataset_condition} )" default_usage = UNIQUE_DATASET_USER_USAGE.format(and_dataset_condition=default_usage_dataset_condition) default_usage = f""" UPDATE galaxy_user SET disk_usage = ({default_usage}) WHERE id = :id """ params = {"id": user_id} if default_exclude_ids: params["exclude_object_store_ids"] = default_exclude_ids statements.append((default_usage, params)) source = quota_source_map.ids_per_quota_source() # TODO: Merge a lot of these settings together by generating a temp table for # the object_store_id to quota_source_label into a temp table of values for quota_source_label, object_store_ids in source.items(): label_usage = UNIQUE_DATASET_USER_USAGE.format( and_dataset_condition="AND ( dataset.object_store_id IN :include_object_store_ids )" ) if for_sqlite: # hacky alternative for older sqlite statement = f""" WITH new (user_id, quota_source_label, disk_usage) AS ( VALUES(:id, :label, ({label_usage})) ) INSERT OR REPLACE INTO user_quota_source_usage (id, user_id, quota_source_label, disk_usage) SELECT old.id, new.user_id, new.quota_source_label, new.disk_usage FROM new LEFT JOIN user_quota_source_usage AS old ON new.user_id = old.user_id AND new.quota_source_label = old.quota_source_label """ else: statement = f""" INSERT INTO user_quota_source_usage(user_id, quota_source_label, disk_usage) VALUES(:id, :label, ({label_usage})) ON CONFLICT ON constraint uqsu_unique_label_per_user DO UPDATE SET disk_usage = excluded.disk_usage """ statements.append( (statement, {"id": user_id, "label": quota_source_label, "include_object_store_ids": object_store_ids}) ) params = {"id": user_id} source_labels = list(source.keys()) if len(source_labels) > 0: clean_old_statement = """ DELETE FROM user_quota_source_usage WHERE user_id = :id AND quota_source_label NOT IN :labels """ params["labels"] = source_labels else: clean_old_statement = """ DELETE FROM user_quota_source_usage WHERE user_id = :id AND quota_source_label IS NOT NULL """ statements.append((clean_old_statement, params)) return statements
UNIQUE_DATASET_USER_USAGE_PER_OBJECTSTORE = """ WITH per_user_histories AS ( SELECT id FROM history WHERE user_id = :id AND NOT purged ), per_hist_hdas AS ( SELECT DISTINCT dataset_id FROM history_dataset_association WHERE NOT purged AND history_id IN (SELECT id FROM per_user_histories) ) SELECT SUM(COALESCE(dataset.total_size, dataset.file_size, 0)) as usage, dataset.object_store_id FROM dataset LEFT OUTER JOIN library_dataset_dataset_association ON dataset.id = library_dataset_dataset_association.dataset_id WHERE dataset.id IN (SELECT dataset_id FROM per_hist_hdas) AND library_dataset_dataset_association.id IS NULL GROUP BY dataset.object_store_id """
[docs]def calculate_disk_usage_per_objectstore(sa_session, user_id: int): statement = UNIQUE_DATASET_USER_USAGE_PER_OBJECTSTORE params = {"id": user_id} return sa_session.execute(text(statement), params).all()
# move these to galaxy.schema.schema once galaxy-data depends on # galaxy-schema.
[docs]class UserQuotaBasicUsage(BaseModel): quota_source_label: Optional[str] = None total_disk_usage: float
[docs]class UserQuotaUsage(UserQuotaBasicUsage): quota_percent: Optional[float] = None quota_bytes: Optional[int] = None quota: Optional[str] = None
[docs]class UserObjectstoreUsage(BaseModel): object_store_id: str total_disk_usage: float
[docs]class User(Base, Dictifiable, RepresentById): """ Data for a Galaxy user or admin and relations to their histories, credentials, and roles. """ use_pbkdf2 = True bootstrap_admin_user = False # api_keys: 'List[APIKeys]' already declared as relationship() __tablename__ = "galaxy_user" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) email: Mapped[str] = mapped_column(TrimmedString(255), index=True, unique=True) username: Mapped[Optional[str]] = mapped_column(TrimmedString(255), index=True, unique=True) password: Mapped[str] = mapped_column(TrimmedString(255)) last_password_change: Mapped[Optional[datetime]] = mapped_column(default=now) external: Mapped[Optional[bool]] = mapped_column(default=False) form_values_id: Mapped[Optional[int]] = mapped_column(ForeignKey("form_values.id"), index=True) preferred_object_store_id: Mapped[str] = mapped_column(String(255), nullable=True) deleted: Mapped[Optional[bool]] = mapped_column(index=True, default=False) purged: Mapped[Optional[bool]] = mapped_column(index=True, default=False) disk_usage: Mapped[Optional[Decimal]] = mapped_column(Numeric(15, 0), index=True) # Column("person_metadata", JSONType), # TODO: add persistent, configurable metadata rep for workflow creator active: Mapped[bool] = mapped_column(index=True, default=True) activation_token: Mapped[Optional[str]] = mapped_column(TrimmedString(64), index=True) addresses: Mapped[List["UserAddress"]] = relationship( back_populates="user", order_by=lambda: desc(UserAddress.update_time), cascade_backrefs=False ) custos_auth: Mapped[List["CustosAuthnzToken"]] = relationship(back_populates="user") default_permissions: Mapped[List["DefaultUserPermissions"]] = relationship(back_populates="user") groups: Mapped[List["UserGroupAssociation"]] = relationship(back_populates="user") histories: Mapped[List["History"]] = relationship( back_populates="user", order_by=lambda: desc(History.update_time), cascade_backrefs=False # type: ignore[has-type] ) active_histories: Mapped[List["History"]] = relationship( primaryjoin=(lambda: (History.user_id == User.id) & (not_(History.deleted)) & (not_(History.archived))), viewonly=True, order_by=lambda: desc(History.update_time), # type: ignore[has-type] ) galaxy_sessions: Mapped[List["GalaxySession"]] = relationship( back_populates="user", order_by=lambda: desc(GalaxySession.update_time), cascade_backrefs=False ) object_stores: Mapped[List["UserObjectStore"]] = relationship(back_populates="user") file_sources: Mapped[List["UserFileSource"]] = relationship(back_populates="user") quotas: Mapped[List["UserQuotaAssociation"]] = relationship(back_populates="user") quota_source_usages: Mapped[List["UserQuotaSourceUsage"]] = relationship(back_populates="user") social_auth: Mapped[List["UserAuthnzToken"]] = relationship(back_populates="user") stored_workflow_menu_entries: Mapped[List["StoredWorkflowMenuEntry"]] = relationship( primaryjoin=( lambda: (StoredWorkflowMenuEntry.user_id == User.id) & (StoredWorkflowMenuEntry.stored_workflow_id == StoredWorkflow.id) & not_(StoredWorkflow.deleted) ), back_populates="user", cascade="all, delete-orphan", collection_class=ordering_list("order_index"), ) _preferences: Mapped[Dict[str, "UserPreference"]] = relationship(collection_class=attribute_keyed_dict("name")) values: Mapped[List["FormValues"]] = relationship(primaryjoin=(lambda: User.form_values_id == FormValues.id)) # Add type hint (will this work w/SA?) api_keys: Mapped[List["APIKeys"]] = relationship( back_populates="user", order_by=lambda: desc(APIKeys.create_time), primaryjoin=( lambda: and_( User.id == APIKeys.user_id, not_(APIKeys.deleted == true()), ) ), ) data_manager_histories: Mapped[List["DataManagerHistoryAssociation"]] = relationship(back_populates="user") roles: Mapped[List["UserRoleAssociation"]] = relationship(back_populates="user") stored_workflows: Mapped[List["StoredWorkflow"]] = relationship( back_populates="user", primaryjoin=(lambda: User.id == StoredWorkflow.user_id), cascade_backrefs=False, ) all_notifications: Mapped[List["UserNotificationAssociation"]] = relationship( back_populates="user", cascade_backrefs=False ) preferences: AssociationProxy[Any] # attributes that will be accessed and returned when calling to_dict( view='collection' ) dict_collection_visible_keys = ["id", "email", "username", "deleted", "active", "last_password_change"] # attributes that will be accessed and returned when calling to_dict( view='element' ) dict_element_visible_keys = [ "id", "email", "username", "total_disk_usage", "nice_total_disk_usage", "deleted", "active", "last_password_change", "preferred_object_store_id", ] def __init__(self, email=None, password=None, username=None): self.email = email self.password = password self.external = False self.deleted = False self.purged = False self.active = False self.username = username
[docs] def get_user_data_tables(self, data_table: str): session = object_session(self) assert session metadata_select = ( select(HistoryDatasetAssociation) .join(Dataset) .join(History) .where( HistoryDatasetAssociation.deleted == false(), HistoryDatasetAssociation.extension == "data_manager_json", History.user_id == self.id, Dataset.state == "ok", # excludes data manager runs that actually populated tables. # maybe track this formally by creating a different datatype for bundles ? Dataset.total_size != Dataset.file_size, HistoryDatasetAssociation._metadata.contains(data_table), ) .order_by(HistoryDatasetAssociation.id) ) return session.execute(metadata_select).scalars().all()
@property def extra_preferences(self): data = defaultdict(lambda: None) if extra_user_preferences := self.preferences.get("extra_user_preferences"): try: data.update(json.loads(extra_user_preferences)) except Exception: pass return data
[docs] def set_password_cleartext(self, cleartext): """ Set user password to the digest of `cleartext`. """ if message := validate_password_str(cleartext): raise Exception(f"Invalid password: {message}") if User.use_pbkdf2: self.password = galaxy.security.passwords.hash_password(cleartext) else: self.password = new_insecure_hash(text_type=cleartext) self.last_password_change = now()
[docs] def set_random_password(self, length=16): """ Sets user password to a random string of the given length. :return: void """ self.set_password_cleartext( "".join(random.SystemRandom().choice(string.ascii_letters + string.digits) for _ in range(length)) )
[docs] def check_password(self, cleartext): """ Check if `cleartext` matches user password when hashed. """ return galaxy.security.passwords.check_password(cleartext, self.password)
[docs] def system_user_pwent(self, real_system_username): """ Gives the system user pwent entry based on e-mail or username depending on the value in real_system_username """ if real_system_username == "user_email": username = self.email.split("@")[0] elif real_system_username == "username": username = self.username else: username = real_system_username try: return pwd.getpwnam(username) except Exception: log.exception(f"Error getting the password database entry for user {username}") raise
[docs] def all_roles(self): """ Return a unique list of Roles associated with this user or any of their groups. """ try: db_session = object_session(self) user = ( db_session.query(User) .filter_by(id=self.id) # don't use get, it will use session variant. .options( joinedload(User.roles), joinedload(User.roles.role), joinedload(User.groups), joinedload(User.groups.group), joinedload(User.groups.group.roles), joinedload(User.groups.group.roles.role), ) .one() ) except Exception: # If not persistent user, just use models normaly and # skip optimizations... user = self roles = [ura.role for ura in user.roles] for group in [uga.group for uga in user.groups]: for role in [gra.role for gra in group.roles]: if role not in roles: roles.append(role) return roles
[docs] def all_roles_exploiting_cache(self): """ """ roles = [ura.role for ura in self.roles] for group in [uga.group for uga in self.groups]: for role in [gra.role for gra in group.roles]: if role not in roles: roles.append(role) return roles
[docs] def get_disk_usage(self, nice_size=False, quota_source_label=None): """ Return byte count of disk space used by user or a human-readable string if `nice_size` is `True`. """ if quota_source_label is None: rval = 0 if self.disk_usage is not None: rval = self.disk_usage else: statement = """ SELECT DISK_USAGE FROM user_quota_source_usage WHERE user_id = :user_id and quota_source_label = :label """ sa_session = object_session(self) params = { "user_id": self.id, "label": quota_source_label, } row = sa_session.execute(text(statement), params).fetchone() if row is not None: rval = row[0] else: rval = 0 if nice_size: rval = galaxy.util.nice_size(rval) return rval
[docs] def set_disk_usage(self, bytes): """ Manually set the disk space used by a user to `bytes`. """ self.disk_usage = bytes
total_disk_usage = property(get_disk_usage, set_disk_usage)
[docs] def adjust_total_disk_usage(self, amount, quota_source_label): assert amount is not None if amount != 0: if quota_source_label is None: self.disk_usage = (self.disk_usage or 0) + amount else: # else would work on newer sqlite - 3.24.0 engine = object_session(self).bind if "sqlite" in engine.dialect.name: # hacky alternative for older sqlite statement = """ WITH new (user_id, quota_source_label) AS ( VALUES(:user_id, :label) ) INSERT OR REPLACE INTO user_quota_source_usage (id, user_id, quota_source_label, disk_usage) SELECT old.id, new.user_id, new.quota_source_label, COALESCE(old.disk_usage + :amount, :amount) FROM new LEFT JOIN user_quota_source_usage AS old ON new.user_id = old.user_id AND NEW.quota_source_label = old.quota_source_label; """ else: statement = """ INSERT INTO user_quota_source_usage(user_id, disk_usage, quota_source_label) VALUES(:user_id, :amount, :label) ON CONFLICT ON constraint uqsu_unique_label_per_user DO UPDATE SET disk_usage = user_quota_source_usage.disk_usage + :amount """ statement = text(statement) params = { "user_id": self.id, "amount": int(amount), "label": quota_source_label, } with engine.connect() as conn, conn.begin(): conn.execute(statement, params)
def _get_social_auth(self, provider_backend): if not self.social_auth: return None for auth in self.social_auth: if auth.provider == provider_backend and auth.extra_data: return auth return None def _get_custos_auth(self, provider_backend): if not self.custos_auth: return None for auth in self.custos_auth: if auth.provider == provider_backend and auth.refresh_token: return auth return None
[docs] def get_oidc_tokens(self, provider_backend): tokens = {"id": None, "access": None, "refresh": None} auth = self._get_social_auth(provider_backend) if auth: tokens["access"] = auth.extra_data.get("access_token", None) tokens["refresh"] = auth.extra_data.get("refresh_token", None) tokens["id"] = auth.extra_data.get("id_token", None) return tokens # no social auth found, check custos auth auth = self._get_custos_auth(provider_backend) if auth: tokens["access"] = auth.access_token tokens["refresh"] = auth.refresh_token tokens["id"] = auth.id_token return tokens
@property def nice_total_disk_usage(self): """ Return byte count of disk space used in a human-readable string. """ return self.get_disk_usage(nice_size=True)
[docs] def calculate_disk_usage_default_source(self, object_store): """ Return byte count total of disk space used by all non-purged, non-library HDAs in non-purged histories assigned to default quota source. """ # only used in set_user_disk_usage.py assert object_store is not None quota_source_map = object_store.get_quota_source_map() default_quota_enabled = quota_source_map.default_quota_enabled exclude_objectstore_ids = quota_source_map.default_usage_excluded_ids() default_cond = "dataset.object_store_id IS NULL OR" if default_quota_enabled and exclude_objectstore_ids else "" default_usage_dataset_condition = ( f"AND ( {default_cond} dataset.object_store_id NOT IN :exclude_object_store_ids )" if exclude_objectstore_ids else "" ) default_usage = UNIQUE_DATASET_USER_USAGE.format(and_dataset_condition=default_usage_dataset_condition) sql_calc = text(default_usage) params = {"id": self.id} bindparams = [bindparam("id")] if exclude_objectstore_ids: params["exclude_object_store_ids"] = exclude_objectstore_ids bindparams.append(bindparam("exclude_object_store_ids", expanding=True)) sql_calc = sql_calc.bindparams(*bindparams) sa_session = object_session(self) usage = sa_session.scalar(sql_calc, params) return usage
[docs] def calculate_and_set_disk_usage(self, object_store): """ Calculates and sets user disk usage. """ self._calculate_or_set_disk_usage(object_store=object_store)
def _calculate_or_set_disk_usage(self, object_store): """ Utility to calculate and return the disk usage. If dryrun is False, the new value is set immediately. """ assert object_store is not None quota_source_map = object_store.get_quota_source_map() sa_session = object_session(self) for_sqlite = "sqlite" in sa_session.bind.dialect.name statements = calculate_user_disk_usage_statements(self.id, quota_source_map, for_sqlite) for sql, args in statements: statement = text(sql) binds = [] for key, _ in args.items(): expand_binding = key.endswith("s") binds.append(bindparam(key, expanding=expand_binding)) statement = statement.bindparams(*binds) sa_session.execute(statement, args) # expire user.disk_usage so sqlalchemy knows to ignore # the existing value - we're setting it in raw SQL for # performance reasons and bypassing object properties. sa_session.expire(self, ["disk_usage"]) with transaction(sa_session): sa_session.commit()
[docs] @staticmethod def user_template_environment(user): """ >>> env = User.user_template_environment(None) >>> env['__user_email__'] 'Anonymous' >>> env['__user_id__'] 'Anonymous' >>> user = User('foo@example.com') >>> user.id = 6 >>> user.username = 'foo2' >>> env = User.user_template_environment(user) >>> env['__user_id__'] '6' >>> env['__user_name__'] 'foo2' """ if user: user_id = "%d" % user.id user_email = str(user.email) user_name = str(user.username) else: user = None user_id = "Anonymous" user_email = "Anonymous" user_name = "Anonymous" environment = {} environment["__user__"] = user environment["__user_id__"] = environment["userId"] = user_id environment["__user_email__"] = environment["userEmail"] = user_email environment["__user_name__"] = user_name return environment
[docs] @staticmethod def expand_user_properties(user, in_string): """ """ environment = User.user_template_environment(user) return Template(in_string).safe_substitute(environment)
# above templating is for Cheetah in tools where we discouraged user details from being exposed. # the following templating if user details in Jinja for object stores and file sources where user # details are critical and documented.
[docs] def config_template_details(self) -> Dict[str, Any]: return { "username": self.username, "email": self.email, "id": self.id, }
[docs] def is_active(self): return self.active
[docs] def is_authenticated(self): # TODO: is required for python social auth (PSA); however, a user authentication is relative to the backend. # For instance, a user who is authenticated with Google, is not necessarily authenticated # with Amazon. Therefore, this function should also receive the backend and check if this # user is already authenticated on that backend or not. For now, returning always True # seems reasonable. Besides, this is also how a PSA example is implemented: # https://github.com/python-social-auth/social-examples/blob/master/example-cherrypy/example/db/user.py return True
[docs] def attempt_create_private_role(self): session = object_session(self) role_name = self.email role_desc = f"Private Role for {self.email}" role_type = Role.types.PRIVATE role = Role(name=role_name, description=role_desc, type=role_type) assoc = UserRoleAssociation(self, role) session.add(assoc) with transaction(session): session.commit()
[docs] def dictify_objectstore_usage(self) -> List[UserObjectstoreUsage]: session = object_session(self) rows = calculate_disk_usage_per_objectstore(session, self.id) return [ UserObjectstoreUsage(object_store_id=r.object_store_id, total_disk_usage=r.usage) for r in rows if r.object_store_id ]
[docs] def dictify_usage(self, object_store=None) -> List[UserQuotaBasicUsage]: """Include object_store to include empty/unused usage info.""" used_labels: Set[Union[str, None]] = set() rval: List[UserQuotaBasicUsage] = [ UserQuotaBasicUsage( quota_source_label=None, total_disk_usage=float(self.disk_usage or 0), ) ] used_labels.add(None) for quota_source_usage in self.quota_source_usages: label = quota_source_usage.quota_source_label rval.append( UserQuotaBasicUsage( quota_source_label=label, total_disk_usage=float(quota_source_usage.disk_usage), ) ) used_labels.add(label) if object_store is not None: for label in object_store.get_quota_source_map().ids_per_quota_source().keys(): if label not in used_labels: rval.append( UserQuotaBasicUsage( quota_source_label=label, total_disk_usage=0.0, ) ) return rval
[docs] def dictify_usage_for(self, quota_source_label: Optional[str]) -> UserQuotaBasicUsage: rval: UserQuotaBasicUsage if quota_source_label is None: rval = UserQuotaBasicUsage( quota_source_label=None, total_disk_usage=float(self.disk_usage or 0), ) else: quota_source_usage = self.quota_source_usage_for(quota_source_label) if quota_source_usage is None: rval = UserQuotaBasicUsage( quota_source_label=quota_source_label, total_disk_usage=0.0, ) else: rval = UserQuotaBasicUsage( quota_source_label=quota_source_label, total_disk_usage=float(quota_source_usage.disk_usage), ) return rval
[docs] def quota_source_usage_for(self, quota_source_label: Optional[str]) -> Optional["UserQuotaSourceUsage"]: for quota_source_usage in self.quota_source_usages: if quota_source_usage.quota_source_label == quota_source_label: return quota_source_usage return None
[docs] def count_stored_workflow_user_assocs(self, stored_workflow) -> int: sq = select(StoredWorkflowUserShareAssociation).filter_by(user=self, stored_workflow=stored_workflow).subquery() stmt = select(func.count()).select_from(sq) session = object_session(self) assert session return session.scalar(stmt) or 0
[docs]class PasswordResetToken(Base): __tablename__ = "password_reset_token" token: Mapped[str] = mapped_column(String(32), primary_key=True, unique=True, index=True) expiration_time: Mapped[Optional[datetime]] user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) user: Mapped[Optional["User"]] = relationship() def __init__(self, user, token=None): if token: self.token = token else: self.token = unique_id() self.user = user self.expiration_time = now() + timedelta(hours=24)
[docs]class ToolSource(Base, Dictifiable, RepresentById): __tablename__ = "tool_source" id: Mapped[int] = mapped_column(primary_key=True) hash: Mapped[Optional[str]] = mapped_column(Unicode(255)) source: Mapped[dict] = mapped_column(JSONType)
[docs]class ToolRequest(Base, Dictifiable, RepresentById): __tablename__ = "tool_request" states: TypeAlias = ToolRequestState id: Mapped[int] = mapped_column(primary_key=True) tool_source_id: Mapped[int] = mapped_column(ForeignKey("tool_source.id"), index=True) history_id: Mapped[Optional[int]] = mapped_column(ForeignKey("history.id"), index=True) request: Mapped[dict] = mapped_column(JSONType) state: Mapped[Optional[str]] = mapped_column(TrimmedString(32), index=True) state_message: Mapped[Optional[str]] = mapped_column(JSONType, index=True) tool_source: Mapped["ToolSource"] = relationship() history: Mapped[Optional["History"]] = relationship(back_populates="tool_requests")
[docs]class DynamicTool(Base, Dictifiable, RepresentById): __tablename__ = "dynamic_tool" id: Mapped[int] = mapped_column(primary_key=True) uuid: Mapped[Optional[Union[UUID, str]]] = mapped_column(UUIDType()) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(index=True, default=now, onupdate=now, nullable=True) tool_id: Mapped[Optional[str]] = mapped_column(Unicode(255)) tool_version: Mapped[Optional[str]] = mapped_column(Unicode(255)) tool_format: Mapped[Optional[str]] = mapped_column(Unicode(255)) tool_path: Mapped[Optional[str]] = mapped_column(Unicode(255)) tool_directory: Mapped[Optional[str]] = mapped_column(Unicode(255)) hidden: Mapped[Optional[bool]] = mapped_column(default=True) active: Mapped[Optional[bool]] = mapped_column(default=True) value: Mapped[Optional[bytes]] = mapped_column(MutableJSONType) dict_collection_visible_keys = ("id", "tool_id", "tool_format", "tool_version", "uuid", "active", "hidden") dict_element_visible_keys = ("id", "tool_id", "tool_format", "tool_version", "uuid", "active", "hidden") def __init__(self, active=True, hidden=True, **kwd): super().__init__(**kwd) self.active = active self.hidden = hidden _uuid = kwd.get("uuid") self.uuid = get_uuid(_uuid)
[docs]class BaseJobMetric(Base): __abstract__ = True
[docs] def __init__(self, plugin, metric_name, metric_value): super().__init__() self.plugin = plugin self.metric_name = metric_name self.metric_value = metric_value
[docs]class JobMetricText(BaseJobMetric, RepresentById): __tablename__ = "job_metric_text" id: Mapped[int] = mapped_column(primary_key=True) job_id: Mapped[Optional[int]] = mapped_column(ForeignKey("job.id"), index=True) plugin: Mapped[Optional[str]] = mapped_column(Unicode(255)) metric_name: Mapped[Optional[str]] = mapped_column(Unicode(255)) metric_value: Mapped[Optional[str]] = mapped_column(Unicode(JOB_METRIC_MAX_LENGTH))
[docs]class JobMetricNumeric(BaseJobMetric, RepresentById): __tablename__ = "job_metric_numeric" id: Mapped[int] = mapped_column(primary_key=True) job_id: Mapped[Optional[int]] = mapped_column(ForeignKey("job.id"), index=True) plugin: Mapped[Optional[str]] = mapped_column(Unicode(255)) metric_name: Mapped[Optional[str]] = mapped_column(Unicode(255)) metric_value: Mapped[Optional[Decimal]] = mapped_column(Numeric(JOB_METRIC_PRECISION, JOB_METRIC_SCALE))
[docs]class TaskMetricText(BaseJobMetric, RepresentById): __tablename__ = "task_metric_text" id: Mapped[int] = mapped_column(primary_key=True) task_id: Mapped[Optional[int]] = mapped_column(ForeignKey("task.id"), index=True) plugin: Mapped[Optional[str]] = mapped_column(Unicode(255)) metric_name: Mapped[Optional[str]] = mapped_column(Unicode(255)) metric_value: Mapped[Optional[str]] = mapped_column(Unicode(JOB_METRIC_MAX_LENGTH))
[docs]class TaskMetricNumeric(BaseJobMetric, RepresentById): __tablename__ = "task_metric_numeric" id: Mapped[int] = mapped_column(primary_key=True) task_id: Mapped[Optional[int]] = mapped_column(ForeignKey("task.id"), index=True) plugin: Mapped[Optional[str]] = mapped_column(Unicode(255)) metric_name: Mapped[Optional[str]] = mapped_column(Unicode(255)) metric_value: Mapped[Optional[Decimal]] = mapped_column(Numeric(JOB_METRIC_PRECISION, JOB_METRIC_SCALE))
[docs]class IoDicts(NamedTuple): inp_data: Dict[str, Optional["DatasetInstance"]] out_data: Dict[str, "DatasetInstance"] out_collections: Dict[str, Union["DatasetCollectionInstance", "DatasetCollection"]]
[docs]class Job(Base, JobLike, UsesCreateAndUpdateTime, Dictifiable, Serializable): """ A job represents a request to run a tool given input datasets, tool parameters, and output datasets. """ __tablename__ = "job" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, index=True, nullable=True) history_id: Mapped[Optional[int]] = mapped_column(ForeignKey("history.id"), index=True) library_folder_id: Mapped[Optional[int]] = mapped_column(ForeignKey("library_folder.id"), index=True) tool_id: Mapped[Optional[str]] = mapped_column(String(255)) tool_version: Mapped[Optional[str]] = mapped_column(TEXT, default="1.0.0") galaxy_version: Mapped[Optional[str]] = mapped_column(String(64), default=None) dynamic_tool_id: Mapped[Optional[int]] = mapped_column(ForeignKey("dynamic_tool.id"), index=True) state: Mapped[str] = mapped_column(String(64), index=True, nullable=True) info: Mapped[Optional[str]] = mapped_column(TrimmedString(255)) copied_from_job_id: Mapped[Optional[int]] command_line: Mapped[Optional[str]] = mapped_column(TEXT) dependencies: Mapped[Optional[bytes]] = mapped_column(MutableJSONType) job_messages: Mapped[Optional[bytes]] = mapped_column(MutableJSONType) param_filename: Mapped[Optional[str]] = mapped_column(String(1024)) runner_name: Mapped[Optional[str]] = mapped_column(String(255)) job_stdout: Mapped[Optional[str]] = mapped_column(TEXT) job_stderr: Mapped[Optional[str]] = mapped_column(TEXT) tool_stdout: Mapped[Optional[str]] = mapped_column(TEXT) tool_stderr: Mapped[Optional[str]] = mapped_column(TEXT) exit_code: Mapped[Optional[int]] traceback: Mapped[Optional[str]] = mapped_column(TEXT) session_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_session.id"), index=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) job_runner_name: Mapped[Optional[str]] = mapped_column(String(255)) job_runner_external_id: Mapped[Optional[str]] = mapped_column(String(255), index=True) destination_id: Mapped[Optional[str]] = mapped_column(String(255)) destination_params: Mapped[Optional[bytes]] = mapped_column(MutableJSONType) object_store_id: Mapped[Optional[str]] = mapped_column(TrimmedString(255), index=True) imported: Mapped[Optional[bool]] = mapped_column(default=False, index=True) params: Mapped[Optional[str]] = mapped_column(TrimmedString(255), index=True) handler: Mapped[Optional[str]] = mapped_column(TrimmedString(255), index=True) preferred_object_store_id: Mapped[Optional[str]] = mapped_column(String(255)) object_store_id_overrides: Mapped[Optional[STR_TO_STR_DICT]] = mapped_column(JSONType) tool_request_id: Mapped[Optional[int]] = mapped_column(ForeignKey("tool_request.id"), index=True) tool_request: Mapped[Optional["ToolRequest"]] = relationship() user: Mapped[Optional["User"]] = relationship() galaxy_session: Mapped[Optional["GalaxySession"]] = relationship() history: Mapped[Optional["History"]] = relationship(back_populates="jobs") library_folder: Mapped[Optional["LibraryFolder"]] = relationship() parameters = relationship("JobParameter") input_datasets: Mapped[List["JobToInputDatasetAssociation"]] = relationship( "JobToInputDatasetAssociation", back_populates="job" ) input_dataset_collections: Mapped[List["JobToInputDatasetCollectionAssociation"]] = relationship( back_populates="job" ) input_dataset_collection_elements: Mapped[List["JobToInputDatasetCollectionElementAssociation"]] = relationship( back_populates="job" ) output_dataset_collection_instances: Mapped[List["JobToOutputDatasetCollectionAssociation"]] = relationship( back_populates="job" ) output_dataset_collections: Mapped[List["JobToImplicitOutputDatasetCollectionAssociation"]] = relationship( back_populates="job" ) post_job_actions: Mapped[List["PostJobActionAssociation"]] = relationship( back_populates="job", cascade_backrefs=False ) input_library_datasets: Mapped[List["JobToInputLibraryDatasetAssociation"]] = relationship(back_populates="job") output_library_datasets: Mapped[List["JobToOutputLibraryDatasetAssociation"]] = relationship(back_populates="job") external_output_metadata: Mapped[List["JobExternalOutputMetadata"]] = relationship(back_populates="job") tasks: Mapped[List["Task"]] = relationship(back_populates="job") output_datasets: Mapped[List["JobToOutputDatasetAssociation"]] = relationship(back_populates="job") state_history: Mapped[List["JobStateHistory"]] = relationship() text_metrics: Mapped[List["JobMetricText"]] = relationship() numeric_metrics: Mapped[List["JobMetricNumeric"]] = relationship() interactivetool_entry_points: Mapped[List["InteractiveToolEntryPoint"]] = relationship( back_populates="job", uselist=True ) implicit_collection_jobs_association: Mapped[List["ImplicitCollectionJobsJobAssociation"]] = relationship( back_populates="job", uselist=False, cascade_backrefs=False ) container: Mapped[Optional["JobContainerAssociation"]] = relationship(back_populates="job", uselist=False) data_manager_association: Mapped[Optional["DataManagerJobAssociation"]] = relationship( back_populates="job", uselist=False, cascade_backrefs=False ) history_dataset_collection_associations: Mapped[List["HistoryDatasetCollectionAssociation"]] = relationship( back_populates="job" ) workflow_invocation_step: Mapped[Optional["WorkflowInvocationStep"]] = relationship( back_populates="job", uselist=False, cascade_backrefs=False ) any_output_dataset_collection_instances_deleted = None any_output_dataset_deleted = None dict_collection_visible_keys = ["id", "state", "exit_code", "update_time", "create_time", "galaxy_version"] dict_element_visible_keys = [ "id", "state", "exit_code", "update_time", "create_time", "galaxy_version", "command_version", "copied_from_job_id", "user_id", ] _numeric_metric = JobMetricNumeric _text_metric = JobMetricText states: TypeAlias = JobState # states that are not expected to change, except through admin action or re-scheduling terminal_states = [states.OK, states.ERROR, states.DELETED] # deleting state should not turn back into any of the non-ready states finished_states = terminal_states + [states.DELETING] #: job states where the job hasn't finished and the model may still change non_ready_states = [ states.NEW, states.RESUBMITTED, states.UPLOAD, states.WAITING, states.QUEUED, states.RUNNING, ] # Please include an accessor (get/set pair) for any new columns/members. def __init__(self): self.dependencies = [] self.state = Job.states.NEW self.imported = False self._init_metrics() self.state_history.append(JobStateHistory(self)) @property def running(self): return self.state == Job.states.RUNNING @property def finished(self): return self.state in self.finished_states
[docs] def io_dicts(self, exclude_implicit_outputs=False) -> IoDicts: inp_data: Dict[str, Optional[DatasetInstance]] = {da.name: da.dataset for da in self.input_datasets} out_data: Dict[str, DatasetInstance] = {da.name: da.dataset for da in self.output_datasets} inp_data.update([(da.name, da.dataset) for da in self.input_library_datasets]) out_data.update([(da.name, da.dataset) for da in self.output_library_datasets]) out_collections: Dict[str, Union[DatasetCollectionInstance, DatasetCollection]] if not exclude_implicit_outputs: out_collections = { obj.name: obj.dataset_collection_instance for obj in self.output_dataset_collection_instances } else: out_collections = {} for obj in self.output_dataset_collection_instances: if obj.name not in out_data: out_collections[obj.name] = obj.dataset_collection_instance # else this is a mapped over output out_collections.update([(obj.name, obj.dataset_collection) for obj in self.output_dataset_collections]) return IoDicts(inp_data, out_data, out_collections)
# TODO: Add accessors for members defined in SQL Alchemy for the Job table and # for the mapper defined to the Job table.
[docs] def get_external_output_metadata(self): """ The external_output_metadata is currently a reference from Job to JobExternalOutputMetadata. It exists for a job but not a task. """ return self.external_output_metadata
[docs] def get_session_id(self): return self.session_id
[docs] def get_user_id(self): return self.user_id
[docs] def get_tool_id(self): return self.tool_id
[docs] def get_tool_version(self): return self.tool_version
[docs] def get_command_line(self): return self.command_line
[docs] def get_dependencies(self): return self.dependencies
[docs] def get_param_filename(self): return self.param_filename
[docs] def get_parameters(self): return self.parameters
[docs] def get_copied_from_job_id(self): return self.copied_from_job_id
[docs] def get_input_datasets(self): return self.input_datasets
[docs] def get_output_datasets(self): return self.output_datasets
[docs] def get_input_library_datasets(self): return self.input_library_datasets
[docs] def get_output_library_datasets(self): return self.output_library_datasets
[docs] def get_state(self): return self.state
[docs] def get_info(self): return self.info
[docs] def get_job_runner_name(self): # This differs from the Task class in that job_runner_name is # accessed instead of task_runner_name. Note that the field # runner_name is not the same thing. return self.job_runner_name
[docs] def get_job_runner_external_id(self): # This is different from the Task just in the member accessed: return self.job_runner_external_id
[docs] def get_post_job_actions(self): return self.post_job_actions
[docs] def get_imported(self): return self.imported
[docs] def get_handler(self): return self.handler
[docs] def get_params(self): return self.params
[docs] def get_user(self): # This is defined in the SQL Alchemy mapper as a relation to the User. return self.user
[docs] def get_tasks(self): # The tasks member is pert of a reference in the SQL Alchemy schema: return self.tasks
[docs] def get_id_tag(self): """ Return a tag that can be useful in identifying a Job. This returns the Job's get_id """ return f"{self.id}"
[docs] def set_session_id(self, session_id): self.session_id = session_id
[docs] def set_user_id(self, user_id): self.user_id = user_id
[docs] def set_tool_id(self, tool_id): self.tool_id = tool_id
[docs] def get_user_email(self): if self.user is not None: return self.user.email elif self.galaxy_session is not None and self.galaxy_session.user is not None: return self.galaxy_session.user.email elif self.history is not None and self.history.user is not None: return self.history.user.email return None
[docs] def set_tool_version(self, tool_version): self.tool_version = tool_version
[docs] def set_command_line(self, command_line): self.command_line = command_line
[docs] def set_dependencies(self, dependencies): self.dependencies = dependencies
[docs] def set_param_filename(self, param_filename): self.param_filename = param_filename
[docs] def set_parameters(self, parameters): self.parameters = parameters
[docs] def set_copied_from_job_id(self, job_id): self.copied_from_job_id = job_id
[docs] def set_input_datasets(self, input_datasets): self.input_datasets = input_datasets
[docs] def set_output_datasets(self, output_datasets): self.output_datasets = output_datasets
[docs] def set_input_library_datasets(self, input_library_datasets): self.input_library_datasets = input_library_datasets
[docs] def set_output_library_datasets(self, output_library_datasets): self.output_library_datasets = output_library_datasets
[docs] def set_info(self, info): self.info = info
[docs] def set_runner_name(self, job_runner_name): self.job_runner_name = job_runner_name
[docs] def get_job(self): # Added so job and task have same interface (.get_job() ) to get at # underlying job object. return self
[docs] def set_runner_external_id(self, job_runner_external_id): self.job_runner_external_id = job_runner_external_id
[docs] def set_post_job_actions(self, post_job_actions): self.post_job_actions = post_job_actions
[docs] def set_imported(self, imported): self.imported = imported
[docs] def set_handler(self, handler): self.handler = handler
[docs] def set_params(self, params): self.params = params
[docs] def add_parameter(self, name, value): self.parameters.append(JobParameter(name, value))
[docs] def add_input_dataset(self, name, dataset=None, dataset_id=None): assoc = JobToInputDatasetAssociation(name, dataset) if dataset is None and dataset_id is not None: assoc.dataset_id = dataset_id add_object_to_object_session(self, assoc) self.input_datasets.append(assoc)
[docs] def add_output_dataset(self, name, dataset): joda = JobToOutputDatasetAssociation(name, dataset) if dataset.dataset.job is None: # Only set job if dataset doesn't already have associated job. # database operation tools that make copies should not modify the job here. dataset.dataset.job = self add_object_to_object_session(self, joda) self.output_datasets.append(joda)
[docs] def add_input_dataset_collection(self, name, dataset_collection): self.input_dataset_collections.append(JobToInputDatasetCollectionAssociation(name, dataset_collection))
[docs] def add_input_dataset_collection_element(self, name, dataset_collection_element): self.input_dataset_collection_elements.append( JobToInputDatasetCollectionElementAssociation(name, dataset_collection_element) )
[docs] def add_output_dataset_collection(self, name, dataset_collection_instance): self.output_dataset_collection_instances.append( JobToOutputDatasetCollectionAssociation(name, dataset_collection_instance) )
[docs] def add_implicit_output_dataset_collection(self, name, dataset_collection): self.output_dataset_collections.append( JobToImplicitOutputDatasetCollectionAssociation(name, dataset_collection) )
[docs] def add_input_library_dataset(self, name, dataset): self.input_library_datasets.append(JobToInputLibraryDatasetAssociation(name, dataset))
[docs] def add_output_library_dataset(self, name, dataset): self.output_library_datasets.append(JobToOutputLibraryDatasetAssociation(name, dataset))
[docs] def add_post_job_action(self, pja): self.post_job_actions.append(PostJobActionAssociation(pja, self))
[docs] def get_change_datatype_actions(self): return { pja.post_job_action.output_name: pja.post_job_action.action_arguments["newtype"] for pja in self.post_job_actions if pja.post_job_action.action_type == "ChangeDatatypeAction" }
@property def all_entry_points_configured(self): # consider an actual DB attribute for this. all_configured = True for ep in self.interactivetool_entry_points: all_configured = ep.configured and all_configured return all_configured
[docs] def set_state(self, state: JobState) -> bool: """ Save state history. Returns True if state has changed, else False. """ if self.state == state: # Nothing changed, no action needed return False session = object_session(self) if session and self.id and state not in Job.finished_states: # Do not update if job is in a terminal state rval = session.execute( update(Job).where(Job.id == self.id, ~Job.state.in_((state, *Job.finished_states))).values(state=state) ) if rval.rowcount == 1: # Need to expire state since we just updated it, but ORM doesn't know about it. session.expire(self, ["state"]) self.state_history.append(JobStateHistory(self)) return True else: return False else: self.state = state self.state_history.append(JobStateHistory(self)) return True
[docs] def get_param_values(self, app, ignore_errors=False): """ Read encoded parameter values from the database and turn back into a dict of tool parameter values. """ param_dict = self.raw_param_dict() tool = app.toolbox.get_tool(self.tool_id, tool_version=self.tool_version) param_dict = tool.params_from_strings(param_dict, app, ignore_errors=ignore_errors) return param_dict
[docs] def raw_param_dict(self): param_dict = {p.name: p.value for p in self.parameters} return param_dict
[docs] def check_if_output_datasets_deleted(self): """ Return true if all of the output datasets associated with this job are in the deleted state """ for dataset_assoc in self.output_datasets: dataset = dataset_assoc.dataset # only the originator of the job can delete a dataset to cause # cancellation of the job, no need to loop through history_associations if not dataset.deleted: return False return True
[docs] def mark_stopped(self, track_jobs_in_database=False): """ Mark this job as stopped """ if self.finished: # Do not modify the state/outputs of jobs that are already terminal return if track_jobs_in_database: self.state = Job.states.STOPPING else: self.state = Job.states.STOPPED
[docs] def mark_deleted(self, track_jobs_in_database=False): """ Mark this job as deleted, and mark any output datasets as discarded. """ if self.finished: # Do not modify the state/outputs of jobs that are already terminal return if track_jobs_in_database: self.state = Job.states.DELETING else: self.state = Job.states.DELETED self.info = "Job output deleted by user before job completed." for jtoda in self.output_datasets: output_hda = jtoda.dataset output_hda.deleted = True output_hda.state = output_hda.states.DISCARDED for shared_hda in output_hda.dataset.history_associations: # propagate info across shared datasets shared_hda.deleted = True shared_hda.blurb = "deleted" shared_hda.peek = "Job deleted" shared_hda.info = "Job output deleted by user before job completed"
[docs] def mark_failed(self, info="Job execution failed", blurb=None, peek=None): """ Mark this job as failed, and mark any output datasets as errored. """ self.state = self.states.FAILED self.info = info for jtod in self.output_datasets: jtod.dataset.state = jtod.dataset.states.ERROR for hda in jtod.dataset.dataset.history_associations: hda.state = hda.states.ERROR if blurb: hda.blurb = blurb if peek: hda.peek = peek hda.info = info
[docs] def resume(self, flush=True): if self.state == self.states.PAUSED: self.set_state(self.states.NEW) object_session(self).add(self) jobs_to_resume = set() for jtod in self.output_datasets: jobs_to_resume.update(jtod.dataset.unpause_dependent_jobs(jobs_to_resume)) for job in jobs_to_resume: job.resume(flush=False) if flush: session = object_session(self) with transaction(session): session.commit()
def _serialize(self, id_encoder, serialization_options): job_attrs = dict_for(self) serialization_options.attach_identifier(id_encoder, self, job_attrs) job_attrs["tool_id"] = self.tool_id job_attrs["tool_version"] = self.tool_version job_attrs["galaxy_version"] = self.galaxy_version job_attrs["state"] = self.state job_attrs["info"] = self.info job_attrs["traceback"] = self.traceback job_attrs["command_line"] = self.command_line job_attrs["tool_stderr"] = self.tool_stderr job_attrs["job_stderr"] = self.job_stderr job_attrs["tool_stdout"] = self.tool_stdout job_attrs["job_stdout"] = self.job_stdout job_attrs["exit_code"] = self.exit_code job_attrs["create_time"] = self.create_time.isoformat() job_attrs["update_time"] = self.update_time.isoformat() job_attrs["job_messages"] = self.job_messages # Get the job's parameters param_dict = self.raw_param_dict() params_objects = {} for key in param_dict: params_objects[key] = safe_loads(param_dict[key]) def remap_objects(p, k, obj): if isinstance(obj, dict) and "src" in obj and obj["src"] in ["hda", "hdca", "dce"]: new_id = serialization_options.get_identifier_for_id(id_encoder, obj["id"]) new_obj = obj.copy() new_obj["id"] = new_id return (k, new_obj) return (k, obj) params_objects = remap(params_objects, remap_objects) params_dict = {} for name, value in params_objects.items(): params_dict[name] = value job_attrs["params"] = params_dict return job_attrs
[docs] def requires_shareable_storage(self, security_agent): # An easy optimization would be to calculate this in galaxy.tools.actions when the # job is created and all the output permissions are already known. Having to reload # these permissions in the job code shouldn't strictly be needed. requires_sharing = False for dataset_assoc in self.output_datasets + self.output_library_datasets: if not security_agent.dataset_is_private_to_a_user(dataset_assoc.dataset.dataset): requires_sharing = True break return requires_sharing
[docs] def to_dict(self, view="collection", system_details=False): if view == "admin_job_list": rval = super().to_dict(view="collection") else: rval = super().to_dict(view=view) rval["tool_id"] = self.tool_id rval["tool_version"] = self.tool_version rval["history_id"] = self.history_id if system_details or view == "admin_job_list": # System level details that only admins should have. rval["external_id"] = self.job_runner_external_id rval["command_line"] = self.command_line rval["traceback"] = self.traceback if view == "admin_job_list": rval["user_email"] = self.user.email if self.user else None rval["handler"] = self.handler rval["job_runner_name"] = self.job_runner_name rval["info"] = self.info rval["session_id"] = self.session_id if self.galaxy_session and self.galaxy_session.remote_host: rval["remote_host"] = self.galaxy_session.remote_host if view == "element": param_dict = {p.name: p.value for p in self.parameters} rval["params"] = param_dict input_dict = {} for i in self.input_datasets: if i.dataset is not None: input_dict[i.name] = { "id": i.dataset.id, "src": "hda", "uuid": str(i.dataset.dataset.uuid) if i.dataset.dataset.uuid is not None else None, } for i in self.input_library_datasets: if i.dataset is not None: input_dict[i.name] = { "id": i.dataset.id, "src": "ldda", "uuid": str(i.dataset.dataset.uuid) if i.dataset.dataset.uuid is not None else None, } for k in input_dict: if k in param_dict: del param_dict[k] rval["inputs"] = input_dict output_dict = {} for i in self.output_datasets: if i.dataset is not None: output_dict[i.name] = { "id": i.dataset.id, "src": "hda", "uuid": str(i.dataset.dataset.uuid) if i.dataset.dataset.uuid is not None else None, } for i in self.output_library_datasets: if i.dataset is not None: output_dict[i.name] = { "id": i.dataset.id, "src": "ldda", "uuid": str(i.dataset.dataset.uuid) if i.dataset.dataset.uuid is not None else None, } rval["outputs"] = output_dict rval["output_collections"] = { jtodca.name: {"id": jtodca.dataset_collection_instance.id, "src": "hdca"} for jtodca in self.output_dataset_collection_instances } return rval
[docs] def update_hdca_update_time_for_job(self, update_time, sa_session, supports_skip_locked): subq = ( sa_session.query(HistoryDatasetCollectionAssociation.id) .join(ImplicitCollectionJobs) .join(ImplicitCollectionJobsJobAssociation) .filter(ImplicitCollectionJobsJobAssociation.job_id == self.id) ) if supports_skip_locked: subq = subq.with_for_update(skip_locked=True).subquery() implicit_statement = ( HistoryDatasetCollectionAssociation.table.update() .where(HistoryDatasetCollectionAssociation.id.in_(select(subq))) .values(update_time=update_time) ) explicit_statement = ( HistoryDatasetCollectionAssociation.table.update() .where(HistoryDatasetCollectionAssociation.job_id == self.id) .values(update_time=update_time) ) sa_session.execute(explicit_statement) if supports_skip_locked: sa_session.execute(implicit_statement) else: conn = sa_session.connection(execution_options={"isolation_level": "SERIALIZABLE"}) with conn.begin() as trans: try: conn.execute(implicit_statement) trans.commit() except OperationalError as e: # If this is a serialization failure on PostgreSQL, then e.orig is a psycopg2 TransactionRollbackError # and should have attribute `code`. Other engines should just report the message and move on. if int(getattr(e.orig, "pgcode", -1)) != 40001: log.debug( f"Updating implicit collection uptime_time for job {self.id} failed (this is expected for large collections and not a problem): {unicodify(e)}" ) trans.rollback()
[docs] def set_final_state(self, final_state, supports_skip_locked): self.set_state(final_state) # TODO: migrate to where-in subqueries? statement = text( """ UPDATE workflow_invocation_step SET update_time = :update_time WHERE job_id = :job_id; """ ) sa_session = object_session(self) update_time = now() self.update_hdca_update_time_for_job( update_time=update_time, sa_session=sa_session, supports_skip_locked=supports_skip_locked ) params = {"job_id": self.id, "update_time": update_time} sa_session.execute(statement, params)
[docs] def get_destination_configuration(self, dest_params, config, key, default=None): """Get a destination parameter that can be defaulted back in specified config if it needs to be applied globally. """ param_unspecified = object() config_value = (self.destination_params or {}).get(key, param_unspecified) if config_value is param_unspecified: config_value = dest_params.get(key, param_unspecified) if config_value is param_unspecified: config_value = getattr(config, key, param_unspecified) if config_value is param_unspecified: config_value = default return config_value
@property def command_version(self): # TODO: make actual database property and track properly - we should be recording this on the job and not on the datasets for dataset_assoc in self.output_datasets: return dataset_assoc.dataset.tool_version
[docs] def update_output_states(self, supports_skip_locked): # TODO: migrate to where-in subqueries? statements = [ text( """ UPDATE dataset SET state = :state, update_time = :update_time WHERE dataset.job_id = :job_id """ ), text( """ UPDATE history_dataset_association SET info = :info, update_time = :update_time FROM dataset WHERE history_dataset_association.dataset_id = dataset.id AND dataset.job_id = :job_id; """ ), text( """ UPDATE library_dataset_dataset_association SET info = :info, update_time = :update_time FROM dataset WHERE library_dataset_dataset_association.dataset_id = dataset.id AND dataset.job_id = :job_id; """ ), ] sa_session = object_session(self) update_time = now() self.update_hdca_update_time_for_job( update_time=update_time, sa_session=sa_session, supports_skip_locked=supports_skip_locked ) params = {"job_id": self.id, "state": self.state, "info": self.info, "update_time": update_time} for statement in statements: sa_session.execute(statement, params)
[docs] def remappable(self): """ Check whether job is remappable when rerun """ if self.state == self.states.ERROR: try: for jtod in self.output_datasets: if jtod.dataset.dependent_jobs: return True if self.output_dataset_collection_instances: # We'll want to replace this item return "job_produced_collection_elements" except Exception: log.exception(f"Error trying to determine if job {self.id} is remappable") return False
[docs] def hide_outputs(self, flush=True): for output_association in self.output_datasets + self.output_dataset_collection_instances: output_association.item.visible = False if flush: session = object_session(self) with transaction(session): session.commit()
[docs]class Task(Base, JobLike, RepresentById): """ A task represents a single component of a job. """ __tablename__ = "task" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) execution_time: Mapped[Optional[datetime]] update_time: Mapped[Optional[datetime]] = mapped_column(default=now, onupdate=now) state: Mapped[Optional[str]] = mapped_column(String(64), index=True) command_line: Mapped[Optional[str]] = mapped_column(TEXT) param_filename: Mapped[Optional[str]] = mapped_column(String(1024)) runner_name: Mapped[Optional[str]] = mapped_column(String(255)) job_stdout: Mapped[Optional[str]] = mapped_column( TEXT ) # job_stdout makes sense here because it is short for job script standard out job_stderr: Mapped[Optional[str]] = mapped_column(TEXT) tool_stdout: Mapped[Optional[str]] = mapped_column(TEXT) tool_stderr: Mapped[Optional[str]] = mapped_column(TEXT) exit_code: Mapped[Optional[int]] job_messages: Mapped[Optional[bytes]] = mapped_column(MutableJSONType) info: Mapped[Optional[str]] = mapped_column(TrimmedString(255)) traceback: Mapped[Optional[str]] = mapped_column(TEXT) job_id: Mapped[int] = mapped_column(ForeignKey("job.id"), index=True) working_directory: Mapped[Optional[str]] = mapped_column(String(1024)) task_runner_name: Mapped[Optional[str]] = mapped_column(String(255)) task_runner_external_id: Mapped[Optional[str]] = mapped_column(String(255)) prepare_input_files_cmd: Mapped[Optional[str]] = mapped_column(TEXT) job: Mapped["Job"] = relationship(back_populates="tasks") text_metrics: Mapped[List["TaskMetricText"]] = relationship() numeric_metrics: Mapped[List["TaskMetricNumeric"]] = relationship() _numeric_metric = TaskMetricNumeric _text_metric = TaskMetricText
[docs] class states(str, Enum): NEW = "new" WAITING = "waiting" QUEUED = "queued" RUNNING = "running" OK = "ok" ERROR = "error" DELETED = "deleted"
# Please include an accessor (get/set pair) for any new columns/members. def __init__(self, job, working_directory, prepare_files_cmd): self.parameters = [] self.state = Task.states.NEW self.working_directory = working_directory add_object_to_object_session(self, job) self.job = job self.prepare_input_files_cmd = prepare_files_cmd self._init_metrics()
[docs] def get_param_values(self, app): """ Read encoded parameter values from the database and turn back into a dict of tool parameter values. """ param_dict = {p.name: p.value for p in self.job.parameters} tool = app.toolbox.get_tool(self.job.tool_id, tool_version=self.job.tool_version) param_dict = tool.params_from_strings(param_dict, app) return param_dict
[docs] def get_id_tag(self): """ Return an id tag suitable for identifying the task. This combines the task's job id and the task's own id. """ return f"{self.job.id}_{self.id}"
[docs] def get_command_line(self): return self.command_line
[docs] def get_parameters(self): return self.parameters
[docs] def get_state(self): return self.state
[docs] def get_info(self): return self.info
[docs] def get_working_directory(self): return self.working_directory
[docs] def get_task_runner_name(self): return self.task_runner_name
[docs] def get_task_runner_external_id(self): return self.task_runner_external_id
[docs] def get_job(self): return self.job
[docs] def get_prepare_input_files_cmd(self): return self.prepare_input_files_cmd
# The following accessors are for members that are in the Job class but # not in the Task class. So they can either refer to the parent Job # or return None, depending on whether Tasks need to point to the parent # (e.g., for a session) or never use the member (e.g., external output # metdata). These can be filled in as needed.
[docs] def get_external_output_metadata(self): """ The external_output_metadata is currently a backref to JobExternalOutputMetadata. It exists for a job but not a task, and when a task is cancelled its corresponding parent Job will be cancelled. So None is returned now, but that could be changed to self.get_job().get_external_output_metadata(). """ return None
[docs] def get_job_runner_name(self): """ Since runners currently access Tasks the same way they access Jobs, this method just refers to *this* instance's runner. """ return self.task_runner_name
[docs] def get_job_runner_external_id(self): """ Runners will use the same methods to get information about the Task class as they will about the Job class, so this method just returns the task's external id. """ # TODO: Merge into get_runner_external_id. return self.task_runner_external_id
[docs] def get_session_id(self): # The Job's galaxy session is equal to the Job's session, so the # Job's session is the same as the Task's session. return self.get_job().get_session_id()
[docs] def set_id(self, id): # This is defined in the SQL Alchemy's mapper and not here. # This should never be called. self.id = id
[docs] def set_command_line(self, command_line): self.command_line = command_line
[docs] def set_parameters(self, parameters): self.parameters = parameters
[docs] def set_state(self, state): self.state = state
[docs] def set_info(self, info): self.info = info
[docs] def set_working_directory(self, working_directory): self.working_directory = working_directory
[docs] def set_task_runner_name(self, task_runner_name): self.task_runner_name = task_runner_name
[docs] def set_job_runner_external_id(self, task_runner_external_id): # This method is available for runners that do not want/need to # differentiate between the kinds of Runnable things (Jobs and Tasks) # that they're using. log.debug("Task %d: Set external id to %s" % (self.id, task_runner_external_id)) self.task_runner_external_id = task_runner_external_id
[docs] def set_task_runner_external_id(self, task_runner_external_id): self.task_runner_external_id = task_runner_external_id
[docs] def set_job(self, job): self.job = job
[docs] def set_prepare_input_files_cmd(self, prepare_input_files_cmd): self.prepare_input_files_cmd = prepare_input_files_cmd
[docs]class JobParameter(Base, RepresentById): __tablename__ = "job_parameter" id: Mapped[int] = mapped_column(primary_key=True) job_id: Mapped[Optional[int]] = mapped_column(ForeignKey("job.id"), index=True) name: Mapped[Optional[str]] = mapped_column(String(255)) value: Mapped[Optional[str]] = mapped_column(TEXT) def __init__(self, name, value): self.name = name self.value = value
[docs] def copy(self): return JobParameter(name=self.name, value=self.value)
[docs]class JobToInputDatasetAssociation(Base, RepresentById): __tablename__ = "job_to_input_dataset" id: Mapped[int] = mapped_column(primary_key=True) job_id: Mapped[int] = mapped_column(ForeignKey("job.id"), index=True, nullable=True) dataset_id: Mapped[int] = mapped_column(ForeignKey("history_dataset_association.id"), index=True, nullable=True) dataset_version: Mapped[Optional[int]] name: Mapped[str] = mapped_column(String(255), nullable=True) dataset: Mapped["HistoryDatasetAssociation"] = relationship(lazy="joined", back_populates="dependent_jobs") job: Mapped["Job"] = relationship(back_populates="input_datasets") def __init__(self, name, dataset): self.name = name add_object_to_object_session(self, dataset) self.dataset = dataset self.dataset_version = 0 # We start with version 0 and update once the job is ready
[docs]class JobToOutputDatasetAssociation(Base, RepresentById): __tablename__ = "job_to_output_dataset" id: Mapped[int] = mapped_column(primary_key=True) job_id: Mapped[int] = mapped_column(ForeignKey("job.id"), index=True, nullable=True) dataset_id: Mapped[int] = mapped_column(ForeignKey("history_dataset_association.id"), index=True, nullable=True) name: Mapped[str] = mapped_column(String(255), nullable=True) dataset: Mapped["HistoryDatasetAssociation"] = relationship( lazy="joined", back_populates="creating_job_associations" ) job: Mapped["Job"] = relationship(back_populates="output_datasets") def __init__(self, name, dataset): self.name = name add_object_to_object_session(self, dataset) self.dataset = dataset @property def item(self): return self.dataset
[docs]class JobToInputDatasetCollectionAssociation(Base, RepresentById): __tablename__ = "job_to_input_dataset_collection" id: Mapped[int] = mapped_column(primary_key=True) job_id: Mapped[int] = mapped_column(ForeignKey("job.id"), index=True, nullable=True) dataset_collection_id: Mapped[int] = mapped_column( ForeignKey("history_dataset_collection_association.id"), index=True, nullable=True ) name: Mapped[str] = mapped_column(String(255), nullable=True) dataset_collection: Mapped["HistoryDatasetCollectionAssociation"] = relationship(lazy="joined") job: Mapped["Job"] = relationship(back_populates="input_dataset_collections") def __init__(self, name, dataset_collection): self.name = name self.dataset_collection = dataset_collection
[docs]class JobToInputDatasetCollectionElementAssociation(Base, RepresentById): __tablename__ = "job_to_input_dataset_collection_element" id: Mapped[int] = mapped_column(primary_key=True) job_id: Mapped[int] = mapped_column(ForeignKey("job.id"), index=True, nullable=True) dataset_collection_element_id: Mapped[int] = mapped_column( ForeignKey("dataset_collection_element.id"), index=True, nullable=True ) name: Mapped[str] = mapped_column(Unicode(255), nullable=True) dataset_collection_element: Mapped["DatasetCollectionElement"] = relationship(lazy="joined") job: Mapped["Job"] = relationship(back_populates="input_dataset_collection_elements") def __init__(self, name, dataset_collection_element): self.name = name self.dataset_collection_element = dataset_collection_element
# Many jobs may map to one HistoryDatasetCollection using these for a given # tool output (if mapping over an input collection).
[docs]class JobToOutputDatasetCollectionAssociation(Base, RepresentById): __tablename__ = "job_to_output_dataset_collection" id: Mapped[int] = mapped_column(primary_key=True) job_id: Mapped[int] = mapped_column(ForeignKey("job.id"), index=True, nullable=True) dataset_collection_id: Mapped[int] = mapped_column( ForeignKey("history_dataset_collection_association.id"), index=True, nullable=True ) name: Mapped[str] = mapped_column(Unicode(255), nullable=True) dataset_collection_instance: Mapped["HistoryDatasetCollectionAssociation"] = relationship(lazy="joined") job: Mapped["Job"] = relationship(back_populates="output_dataset_collection_instances") def __init__(self, name, dataset_collection_instance): self.name = name self.dataset_collection_instance = dataset_collection_instance @property def item(self): return self.dataset_collection_instance
# A DatasetCollection will be mapped to at most one job per tool output # using these. (You can think of many of these models as going into the # creation of a JobToOutputDatasetCollectionAssociation.)
[docs]class JobToImplicitOutputDatasetCollectionAssociation(Base, RepresentById): __tablename__ = "job_to_implicit_output_dataset_collection" id: Mapped[int] = mapped_column(primary_key=True) job_id: Mapped[int] = mapped_column(ForeignKey("job.id"), index=True, nullable=True) dataset_collection_id: Mapped[int] = mapped_column(ForeignKey("dataset_collection.id"), index=True, nullable=True) name: Mapped[str] = mapped_column(Unicode(255), nullable=True) dataset_collection: Mapped["DatasetCollection"] = relationship() job: Mapped["Job"] = relationship(back_populates="output_dataset_collections") def __init__(self, name, dataset_collection): self.name = name self.dataset_collection = dataset_collection
[docs]class JobToInputLibraryDatasetAssociation(Base, RepresentById): __tablename__ = "job_to_input_library_dataset" id: Mapped[int] = mapped_column(primary_key=True) job_id: Mapped[int] = mapped_column(ForeignKey("job.id"), index=True, nullable=True) ldda_id: Mapped[int] = mapped_column( ForeignKey("library_dataset_dataset_association.id"), index=True, nullable=True ) name: Mapped[str] = mapped_column(Unicode(255), nullable=True) job: Mapped["Job"] = relationship(back_populates="input_library_datasets") dataset: Mapped["LibraryDatasetDatasetAssociation"] = relationship(lazy="joined", back_populates="dependent_jobs") def __init__(self, name, dataset): self.name = name add_object_to_object_session(self, dataset) self.dataset = dataset
[docs]class JobToOutputLibraryDatasetAssociation(Base, RepresentById): __tablename__ = "job_to_output_library_dataset" id: Mapped[int] = mapped_column(primary_key=True) job_id: Mapped[int] = mapped_column(ForeignKey("job.id"), index=True, nullable=True) ldda_id: Mapped[int] = mapped_column( ForeignKey("library_dataset_dataset_association.id"), index=True, nullable=True ) name: Mapped[str] = mapped_column(Unicode(255), nullable=True) job: Mapped["Job"] = relationship(back_populates="output_library_datasets") dataset: Mapped["LibraryDatasetDatasetAssociation"] = relationship( lazy="joined", back_populates="creating_job_associations" ) def __init__(self, name, dataset): self.name = name add_object_to_object_session(self, dataset) self.dataset = dataset
[docs]class JobStateHistory(Base, RepresentById): __tablename__ = "job_state_history" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) job_id: Mapped[Optional[int]] = mapped_column(ForeignKey("job.id"), index=True) state: Mapped[Optional[str]] = mapped_column(String(64), index=True) info: Mapped[Optional[str]] = mapped_column(TrimmedString(255)) def __init__(self, job): self.job_id = job.id self.state = job.state self.info = job.info
[docs]class ImplicitlyCreatedDatasetCollectionInput(Base, RepresentById): __tablename__ = "implicitly_created_dataset_collection_inputs" id: Mapped[int] = mapped_column(primary_key=True) dataset_collection_id: Mapped[Optional[int]] = mapped_column( ForeignKey("history_dataset_collection_association.id"), index=True ) input_dataset_collection_id: Mapped[Optional[int]] = mapped_column( ForeignKey("history_dataset_collection_association.id"), index=True ) name: Mapped[Optional[str]] = mapped_column(Unicode(255)) input_dataset_collection: Mapped[Optional["HistoryDatasetCollectionAssociation"]] = relationship( primaryjoin=( lambda: HistoryDatasetCollectionAssociation.id == ImplicitlyCreatedDatasetCollectionInput.input_dataset_collection_id ), ) def __init__(self, name, input_dataset_collection): self.name = name self.input_dataset_collection = input_dataset_collection
[docs]class ImplicitCollectionJobs(Base, Serializable): __tablename__ = "implicit_collection_jobs" id: Mapped[int] = mapped_column(primary_key=True) populated_state: Mapped[str] = mapped_column(TrimmedString(64), default="new") jobs: Mapped[List["ImplicitCollectionJobsJobAssociation"]] = relationship( back_populates="implicit_collection_jobs", cascade_backrefs=False )
[docs] class populated_states(str, Enum): NEW = "new" # New implicit jobs object, unpopulated job associations OK = "ok" # Job associations are set and fixed. FAILED = "failed" # There were issues populating job associations, object is in error.
def __init__(self, populated_state=None): self.populated_state = populated_state or ImplicitCollectionJobs.populated_states.NEW @property def job_list(self): return [icjja.job for icjja in self.jobs] def _serialize(self, id_encoder, serialization_options): rval = dict_for( self, populated_state=self.populated_state, jobs=[serialization_options.get_identifier(id_encoder, j_a.job) for j_a in self.jobs], ) serialization_options.attach_identifier(id_encoder, self, rval) return rval
[docs]class ImplicitCollectionJobsJobAssociation(Base, RepresentById): __tablename__ = "implicit_collection_jobs_job_association" id: Mapped[int] = mapped_column(primary_key=True) implicit_collection_jobs_id: Mapped[int] = mapped_column( ForeignKey("implicit_collection_jobs.id"), index=True, nullable=True ) job_id: Mapped[int] = mapped_column( ForeignKey("job.id"), index=True, nullable=True ) # Consider making this nullable... order_index: Mapped[int] implicit_collection_jobs = relationship("ImplicitCollectionJobs", back_populates="jobs") job: Mapped["Job"] = relationship(back_populates="implicit_collection_jobs_association")
[docs]class PostJobAction(Base, RepresentById): __tablename__ = "post_job_action" id: Mapped[int] = mapped_column(primary_key=True) workflow_step_id: Mapped[Optional[int]] = mapped_column(ForeignKey("workflow_step.id"), index=True) action_type: Mapped[str] = mapped_column(String(255)) output_name: Mapped[Optional[str]] = mapped_column(String(255)) action_arguments: Mapped[Optional[bytes]] = mapped_column(MutableJSONType) workflow_step: Mapped[Optional["WorkflowStep"]] = relationship( back_populates="post_job_actions", primaryjoin=(lambda: WorkflowStep.id == PostJobAction.workflow_step_id), ) def __init__(self, action_type, workflow_step=None, output_name=None, action_arguments=None): self.action_type = action_type self.output_name = output_name self.action_arguments = action_arguments self.workflow_step = workflow_step ensure_object_added_to_session(self, object_in_session=workflow_step)
[docs]class PostJobActionAssociation(Base, RepresentById): __tablename__ = "post_job_action_association" id: Mapped[int] = mapped_column(primary_key=True) job_id: Mapped[int] = mapped_column(ForeignKey("job.id"), index=True) post_job_action_id: Mapped[int] = mapped_column(ForeignKey("post_job_action.id"), index=True) post_job_action: Mapped["PostJobAction"] = relationship() job: Mapped["Job"] = relationship(back_populates="post_job_actions") def __init__(self, pja, job=None, job_id=None): if job is not None: self.job = job ensure_object_added_to_session(self, object_in_session=job) elif job_id is not None: self.job_id = job_id else: raise Exception("PostJobActionAssociation must be created with a job or a job_id.") self.post_job_action = pja
[docs]class JobExternalOutputMetadata(Base, RepresentById): __tablename__ = "job_external_output_metadata" id: Mapped[int] = mapped_column(primary_key=True) job_id: Mapped[Optional[int]] = mapped_column(ForeignKey("job.id"), index=True) history_dataset_association_id: Mapped[Optional[int]] = mapped_column( ForeignKey("history_dataset_association.id"), index=True ) library_dataset_dataset_association_id: Mapped[Optional[int]] = mapped_column( ForeignKey("library_dataset_dataset_association.id"), index=True ) is_valid: Mapped[Optional[bool]] = mapped_column(default=True) filename_in: Mapped[Optional[str]] = mapped_column(String(255)) filename_out: Mapped[Optional[str]] = mapped_column(String(255)) filename_results_code: Mapped[Optional[str]] = mapped_column(String(255)) filename_kwds: Mapped[Optional[str]] = mapped_column(String(255)) filename_override_metadata: Mapped[Optional[str]] = mapped_column(String(255)) job_runner_external_pid: Mapped[Optional[str]] = mapped_column(String(255)) history_dataset_association: Mapped[Optional["HistoryDatasetAssociation"]] = relationship(lazy="joined") library_dataset_dataset_association: Mapped[Optional["LibraryDatasetDatasetAssociation"]] = relationship( lazy="joined" ) job: Mapped[Optional["Job"]] = relationship(back_populates="external_output_metadata") def __init__(self, job=None, dataset=None): add_object_to_object_session(self, job) self.job = job if isinstance(dataset, galaxy.model.HistoryDatasetAssociation): self.history_dataset_association = dataset elif isinstance(dataset, galaxy.model.LibraryDatasetDatasetAssociation): self.library_dataset_dataset_association = dataset @property def dataset(self): if self.history_dataset_association: return self.history_dataset_association elif self.library_dataset_dataset_association: return self.library_dataset_dataset_association return None
# Set up output dataset association for export history jobs. Because job # uses a Dataset rather than an HDA or LDA, it's necessary to set up a # fake dataset association that provides the needed attributes for # preparing a job.
[docs]class FakeDatasetAssociation: fake_dataset_association = True
[docs] def __init__(self, dataset: Optional["Dataset"] = None) -> None: self.dataset = dataset self.metadata: Dict = {} self.has_deferred_data = False
[docs] def get_file_name(self, sync_cache: bool = True) -> str: assert self.dataset return self.dataset.get_file_name(sync_cache)
def __eq__(self, other): return isinstance(other, FakeDatasetAssociation) and self.dataset == other.dataset
[docs]class JobExportHistoryArchive(Base, RepresentById): __tablename__ = "job_export_history_archive" id: Mapped[int] = mapped_column(primary_key=True) job_id: Mapped[Optional[int]] = mapped_column(ForeignKey("job.id"), index=True) history_id: Mapped[Optional[int]] = mapped_column(ForeignKey("history.id"), index=True) dataset_id: Mapped[Optional[int]] = mapped_column(ForeignKey("dataset.id"), index=True) compressed: Mapped[Optional[bool]] = mapped_column(index=True, default=False) history_attrs_filename: Mapped[Optional[str]] = mapped_column(TEXT) job: Mapped[Optional["Job"]] = relationship() dataset: Mapped[Optional["Dataset"]] = relationship() history: Mapped[Optional["History"]] = relationship(back_populates="exports") ATTRS_FILENAME_HISTORY = "history_attrs.txt" def __init__(self, compressed=False, **kwd): if "history" in kwd: add_object_to_object_session(self, kwd["history"]) super().__init__(**kwd) self.compressed = compressed @property def fda(self): return FakeDatasetAssociation(self.dataset) @property def temp_directory(self): return os.path.split(self.history_attrs_filename)[0] @property def up_to_date(self): """Return False, if a new export should be generated for corresponding history. """ job = self.job return job.state not in [Job.states.ERROR, Job.states.DELETED] and job.update_time > self.history.update_time @property def ready(self): return self.job.state == Job.states.OK @property def preparing(self): return self.job.state in [Job.states.RUNNING, Job.states.QUEUED, Job.states.WAITING] @property def export_name(self): # Stream archive. hname = ready_name_for_url(self.history.name) hname = f"Galaxy-History-{hname}.tar" if self.compressed: hname += ".gz" return hname
[docs] @staticmethod def create_for_history(history, job, sa_session, object_store, compressed): # Create dataset that will serve as archive. archive_dataset = Dataset() sa_session.add(archive_dataset) with transaction(sa_session): sa_session.commit() # ensure job.id and archive_dataset.id are available object_store.create(archive_dataset) # set the object store id, create dataset (if applicable) # Add association for keeping track of job, history, archive relationship. jeha = JobExportHistoryArchive(job=job, history=history, dataset=archive_dataset, compressed=compressed) sa_session.add(jeha) # # Create attributes/metadata files for export. # jeha.dataset.create_extra_files_path() temp_output_dir = jeha.dataset.extra_files_path history_attrs_filename = os.path.join(temp_output_dir, jeha.ATTRS_FILENAME_HISTORY) jeha.history_attrs_filename = history_attrs_filename return jeha
[docs] def to_dict(self): return { "id": self.id, "job_id": self.job.id, "ready": self.ready, "preparing": self.preparing, "up_to_date": self.up_to_date, }
[docs]class JobImportHistoryArchive(Base, RepresentById): __tablename__ = "job_import_history_archive" id: Mapped[int] = mapped_column(primary_key=True) job_id: Mapped[Optional[int]] = mapped_column(ForeignKey("job.id"), index=True) history_id: Mapped[Optional[int]] = mapped_column(ForeignKey("history.id"), index=True) archive_dir: Mapped[Optional[str]] = mapped_column(TEXT) job: Mapped[Optional["Job"]] = relationship() history: Mapped[Optional["History"]] = relationship()
[docs]class StoreExportAssociation(Base, RepresentById): __tablename__ = "store_export_association" __table_args__ = (Index("ix_store_export_object", "object_id", "object_type"),) id: Mapped[int] = mapped_column(primary_key=True) task_uuid: Mapped[Optional[Union[UUID, str]]] = mapped_column(UUIDType(), index=True, unique=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) object_type: Mapped[Optional[str]] = mapped_column(TrimmedString(32)) object_id: Mapped[Optional[int]] export_metadata: Mapped[Optional[bytes]] = mapped_column(JSONType)
[docs]class JobContainerAssociation(Base, RepresentById): __tablename__ = "job_container_association" id: Mapped[int] = mapped_column(primary_key=True) job_id: Mapped[int] = mapped_column(ForeignKey("job.id"), index=True, nullable=True) container_type: Mapped[Optional[str]] = mapped_column(TEXT) container_name: Mapped[Optional[str]] = mapped_column(TEXT) container_info: Mapped[Optional[bytes]] = mapped_column(MutableJSONType) created_time: Mapped[Optional[datetime]] = mapped_column(default=now) modified_time: Mapped[Optional[datetime]] = mapped_column(default=now, onupdate=now) job: Mapped["Job"] = relationship(back_populates="container") def __init__(self, **kwd): if "job" in kwd: add_object_to_object_session(self, kwd["job"]) super().__init__(**kwd) self.container_info = self.container_info or {}
[docs]class InteractiveToolEntryPoint(Base, Dictifiable, RepresentById): __tablename__ = "interactivetool_entry_point" id: Mapped[int] = mapped_column(primary_key=True) job_id: Mapped[Optional[int]] = mapped_column(ForeignKey("job.id"), index=True) name: Mapped[Optional[str]] = mapped_column(TEXT) token: Mapped[Optional[str]] = mapped_column(TEXT) tool_port: Mapped[Optional[int]] host: Mapped[Optional[str]] = mapped_column(TEXT) port: Mapped[Optional[int]] protocol: Mapped[Optional[str]] = mapped_column(TEXT) entry_url: Mapped[Optional[str]] = mapped_column(TEXT) requires_domain: Mapped[Optional[bool]] = mapped_column(default=True) requires_path_in_url: Mapped[Optional[bool]] = mapped_column(default=False) requires_path_in_header_named: Mapped[Optional[str]] = mapped_column(TEXT) info: Mapped[Optional[bytes]] = mapped_column(MutableJSONType) configured: Mapped[Optional[bool]] = mapped_column(default=False) deleted: Mapped[Optional[bool]] = mapped_column(default=False) created_time: Mapped[Optional[datetime]] = mapped_column(default=now) modified_time: Mapped[Optional[datetime]] = mapped_column(default=now, onupdate=now) label: Mapped[Optional[str]] = mapped_column(TEXT) job: Mapped[Optional["Job"]] = relationship(back_populates="interactivetool_entry_points", uselist=False) dict_collection_visible_keys = [ "id", "job_id", "name", "label", "active", "created_time", "modified_time", "output_datasets_ids", ] dict_element_visible_keys = [ "id", "job_id", "name", "label", "active", "created_time", "modified_time", "output_datasets_ids", ] def __init__( self, requires_domain=True, requires_path_in_url=False, configured=False, deleted=False, **kwd, ): super().__init__(**kwd) self.requires_domain = requires_domain self.requires_path_in_url = requires_path_in_url self.configured = configured self.deleted = deleted self.token = self.token or hex_to_lowercase_alphanum(token_hex(8)) self.info = self.info or {} @property def active(self): if self.configured and not self.deleted: return self.job.running return False @property def class_id(self): return "ep" @property def output_datasets_ids(self): return [da.dataset.id for da in self.job.output_datasets]
[docs]class GenomeIndexToolData(Base, RepresentById): # TODO: params arg is lost __tablename__ = "genome_index_tool_data" id: Mapped[int] = mapped_column(primary_key=True) job_id: Mapped[Optional[int]] = mapped_column(ForeignKey("job.id"), index=True) dataset_id: Mapped[Optional[int]] = mapped_column(ForeignKey("dataset.id"), index=True) fasta_path: Mapped[Optional[str]] = mapped_column(String(255)) created_time: Mapped[Optional[datetime]] = mapped_column(default=now) modified_time: Mapped[Optional[datetime]] = mapped_column(default=now, onupdate=now) indexer: Mapped[Optional[str]] = mapped_column(String(64)) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) job: Mapped[Optional["Job"]] = relationship() dataset: Mapped[Optional["Dataset"]] = relationship() user: Mapped[Optional["User"]] = relationship()
[docs]class Group(Base, Dictifiable, RepresentById): __tablename__ = "galaxy_group" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) name: Mapped[Optional[str]] = mapped_column(String(255), index=True, unique=True) deleted: Mapped[Optional[bool]] = mapped_column(index=True, default=False) quotas: Mapped[List["GroupQuotaAssociation"]] = relationship(back_populates="group") roles: Mapped[List["GroupRoleAssociation"]] = relationship(back_populates="group", cascade_backrefs=False) users: Mapped[List["UserGroupAssociation"]] = relationship("UserGroupAssociation", back_populates="group") dict_collection_visible_keys = ["id", "name"] dict_element_visible_keys = ["id", "name"] def __init__(self, name=None): self.name = name self.deleted = False
[docs]class UserGroupAssociation(Base, RepresentById): __tablename__ = "user_group_association" __table_args__ = (UniqueConstraint("user_id", "group_id"),) id: Mapped[int] = mapped_column(primary_key=True) user_id: Mapped[int] = mapped_column(ForeignKey("galaxy_user.id"), index=True) group_id: Mapped[int] = mapped_column(ForeignKey("galaxy_group.id"), index=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) user: Mapped["User"] = relationship(back_populates="groups") group: Mapped["Group"] = relationship(back_populates="users") def __init__(self, user, group): add_object_to_object_session(self, user) self.user = user self.group = group
[docs]class Notification(Base, Dictifiable, RepresentById): __tablename__ = "notification" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) publication_time: Mapped[Optional[datetime]] = mapped_column( default=now ) # The date of publication, can be a future date to allow scheduling expiration_time: Mapped[Optional[datetime]] = mapped_column( default=now() + timedelta(days=30 * 6) ) # The expiration date, expired notifications will be permanently removed from DB regularly source: Mapped[str] = mapped_column( String(32), index=True, nullable=True ) # Who (or what) generated the notification category: Mapped[str] = mapped_column( String(64), index=True, nullable=True ) # Category of the notification, defines its contents. Used for filtering, un/subscribing, etc variant: Mapped[str] = mapped_column( String(16), index=True, nullable=True ) # Defines the 'importance' of the notification ('info', 'warning', 'urgent', etc.). Used for filtering, highlight rendering, etc dispatched: Mapped[bool] = mapped_column( Boolean, index=True, default=False ) # Whether the notification has been dispatched to users via other channels galaxy_url: Mapped[Optional[str]] = mapped_column( String(255) ) # The URL to the Galaxy instance, used for generating links in the notification # A bug in early 23.1 led to values being stored as json string, so we use this special type to process the result value twice. # content should always be a dict content: Mapped[Optional[bytes]] = mapped_column(DoubleEncodedJsonType) user_notification_associations: Mapped[List["UserNotificationAssociation"]] = relationship( back_populates="notification" ) def __init__(self, source: str, category: str, variant: str, content): self.source = source self.category = category self.variant = variant self.content = content
[docs]class UserNotificationAssociation(Base, RepresentById): __tablename__ = "user_notification_association" id: Mapped[int] = mapped_column(primary_key=True) user_id: Mapped[int] = mapped_column(ForeignKey("galaxy_user.id"), index=True, nullable=True) notification_id: Mapped[int] = mapped_column(ForeignKey("notification.id"), index=True, nullable=True) seen_time: Mapped[Optional[datetime]] deleted: Mapped[Optional[bool]] = mapped_column(index=True, default=False) update_time: Mapped[Optional[datetime]] = mapped_column(default=now, onupdate=now) user: Mapped["User"] = relationship(back_populates="all_notifications") notification: Mapped["Notification"] = relationship(back_populates="user_notification_associations") def __init__(self, user, notification): self.user = user ensure_object_added_to_session(self, object_in_session=user) self.notification = notification
[docs]def is_hda(d): return isinstance(d, HistoryDatasetAssociation)
[docs]class HistoryAudit(Base): __tablename__ = "history_audit" __table_args__ = (PrimaryKeyConstraint(sqlite_on_conflict="IGNORE"),) history_id: Mapped[int] = mapped_column(ForeignKey("history.id"), primary_key=True) update_time: Mapped[datetime] = mapped_column(default=now, primary_key=True) # This class should never be instantiated. # See https://github.com/galaxyproject/galaxy/pull/11914 for details. __init__ = None # type: ignore[assignment] def __repr__(self): try: r = f"<galaxy.model.{self.__class__.__name__}({cached_id(self)}) at {hex(id(self))}>" except Exception: r = object.__repr__(self) log.exception("Caught exception attempting to generate repr for: %s", r) return r
[docs] @classmethod def prune(cls, sa_session): latest_subq = ( select(cls.history_id, func.max(cls.update_time).label("max_update_time")) .group_by(cls.history_id) .subquery() ) not_latest_query = ( select(cls.history_id, cls.update_time) .select_from(latest_subq) .join( cls, and_( cls.update_time < latest_subq.columns.max_update_time, cls.history_id == latest_subq.columns.history_id, ), ) .subquery() ) q = cls.__table__.delete().where(tuple_(cls.history_id, cls.update_time).in_(select(not_latest_query))) with sa_session() as session, session.begin(): session.execute(q)
[docs]class History(Base, HasTags, Dictifiable, UsesAnnotations, HasName, Serializable, UsesCreateAndUpdateTime): __tablename__ = "history" __table_args__ = (Index("ix_history_slug", "slug", mysql_length=200),) id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) _update_time: Mapped[datetime] = mapped_column( "update_time", DateTime, index=True, default=now, onupdate=now, nullable=True ) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) name: Mapped[Optional[str]] = mapped_column(TrimmedString(255)) hid_counter: Mapped[Optional[int]] = mapped_column(default=1) deleted: Mapped[Optional[bool]] = mapped_column(index=True, default=False) purged: Mapped[Optional[bool]] = mapped_column(index=True, default=False) importing: Mapped[Optional[bool]] = mapped_column(index=True, default=False) genome_build: Mapped[Optional[str]] = mapped_column(TrimmedString(40)) importable: Mapped[Optional[bool]] = mapped_column(default=False) slug: Mapped[Optional[str]] = mapped_column(TEXT) published: Mapped[Optional[bool]] = mapped_column(index=True, default=False) preferred_object_store_id: Mapped[Optional[str]] = mapped_column(String(255)) archived: Mapped[Optional[bool]] = mapped_column(index=True, default=False, server_default=false()) archive_export_id: Mapped[Optional[int]] = mapped_column(ForeignKey("store_export_association.id"), default=None) datasets: Mapped[List["HistoryDatasetAssociation"]] = relationship( back_populates="history", cascade_backrefs=False, order_by=lambda: asc(HistoryDatasetAssociation.hid) # type: ignore[has-type] ) exports: Mapped[List["JobExportHistoryArchive"]] = relationship( back_populates="history", primaryjoin=lambda: JobExportHistoryArchive.history_id == History.id, order_by=lambda: desc(JobExportHistoryArchive.id), ) active_datasets: Mapped[List["HistoryDatasetAssociation"]] = relationship( primaryjoin=( lambda: and_( HistoryDatasetAssociation.history_id == History.id, # type: ignore[arg-type] not_(HistoryDatasetAssociation.deleted), # type: ignore[has-type] ) ), order_by=lambda: asc(HistoryDatasetAssociation.hid), # type: ignore[has-type] viewonly=True, ) dataset_collections: Mapped[List["HistoryDatasetCollectionAssociation"]] = relationship(back_populates="history") active_dataset_collections: Mapped[List["HistoryDatasetCollectionAssociation"]] = relationship( primaryjoin=( lambda: ( and_( HistoryDatasetCollectionAssociation.history_id == History.id, not_(HistoryDatasetCollectionAssociation.deleted), # type: ignore[arg-type] ) ) ), order_by=lambda: asc(HistoryDatasetCollectionAssociation.hid), viewonly=True, ) visible_datasets: Mapped[List["HistoryDatasetAssociation"]] = relationship( primaryjoin=( lambda: and_( HistoryDatasetAssociation.history_id == History.id, # type: ignore[arg-type] not_(HistoryDatasetAssociation.deleted), # type: ignore[has-type] HistoryDatasetAssociation.visible, # type: ignore[has-type] ) ), order_by=lambda: asc(HistoryDatasetAssociation.hid), # type: ignore[has-type] viewonly=True, ) visible_dataset_collections: Mapped[List["HistoryDatasetCollectionAssociation"]] = relationship( primaryjoin=( lambda: and_( HistoryDatasetCollectionAssociation.history_id == History.id, not_(HistoryDatasetCollectionAssociation.deleted), # type: ignore[arg-type] HistoryDatasetCollectionAssociation.visible, # type: ignore[arg-type] ) ), order_by=lambda: asc(HistoryDatasetCollectionAssociation.hid), viewonly=True, ) tags: Mapped[List["HistoryTagAssociation"]] = relationship( order_by=lambda: HistoryTagAssociation.id, back_populates="history" ) annotations: Mapped[List["HistoryAnnotationAssociation"]] = relationship( order_by=lambda: HistoryAnnotationAssociation.id, back_populates="history" ) ratings: Mapped[List["HistoryRatingAssociation"]] = relationship( order_by=lambda: HistoryRatingAssociation.id, back_populates="history", ) default_permissions: Mapped[List["DefaultHistoryPermissions"]] = relationship(back_populates="history") users_shared_with: Mapped[List["HistoryUserShareAssociation"]] = relationship(back_populates="history") galaxy_sessions = relationship("GalaxySessionToHistoryAssociation", back_populates="history") workflow_invocations: Mapped[List["WorkflowInvocation"]] = relationship( back_populates="history", cascade_backrefs=False ) user: Mapped[Optional["User"]] = relationship(back_populates="histories") jobs: Mapped[List["Job"]] = relationship(back_populates="history", cascade_backrefs=False) tool_requests: Mapped[List["ToolRequest"]] = relationship(back_populates="history") update_time = column_property( select(func.max(HistoryAudit.update_time)).where(HistoryAudit.history_id == id).scalar_subquery(), ) users_shared_with_count = None average_rating = None # Set up proxy so that # History.users_shared_with # returns a list of users that history is shared with. users_shared_with_dot_users = association_proxy("users_shared_with", "user") dict_collection_visible_keys = ["id", "name", "published", "deleted"] dict_element_visible_keys = [ "id", "name", "archived", "create_time", "deleted", "empty", "genome_build", "hid_counter", "importable", "preferred_object_store_id", "purged", "published", "slug", "tags", "update_time", "username", ] default_name = "Unnamed history" def __init__(self, id=None, name=None, user=None): self.id = id self.name = name or History.default_name self.deleted = False self.purged = False self.importing = False self.published = False add_object_to_object_session(self, user) self.user = user # Objects to eventually add to history self._pending_additions = [] self._item_by_hid_cache = None
[docs] @reconstructor def init_on_load(self): # Restores properties that are not tracked in the database self._pending_additions = []
[docs] def stage_addition(self, items): history_id = self.id for item in listify(items): item.history = self if history_id: item.history_id = history_id self._pending_additions.append(item)
@property def empty(self): return self.hid_counter is None or self.hid_counter == 1 @property def username(self): if user := self.user: return user.username return None @property def count(self): return self.hid_counter - 1
[docs] def update(self): self._update_time = now()
[docs] def add_pending_items(self, set_output_hid=True): # These are assumed to be either copies of existing datasets or new, empty datasets, # so we don't need to set the quota. self.add_datasets( object_session(self), self._pending_additions, set_hid=set_output_hid, quota=False, flush=False ) self._pending_additions = []
def _next_hid(self, n=1): """ Generate next_hid from the database in a concurrency safe way: 1. Retrieve hid_counter from database 2. Increment hid_counter by n and store in database 3. Return retrieved hid_counter. Handle with SQLAlchemy Core to keep this independent from current session state, except: expire hid_counter attribute, since its value in the session is no longer valid. """ session = object_session(self) engine = session.bind table = self.__table__ history_id = cached_id(self) update_stmt = update(table).where(table.c.id == history_id).values(hid_counter=table.c.hid_counter + n) update_returning_stmnt = update_stmt.returning(table.c.hid_counter) if engine.name != "sqlite": with engine.begin() as conn: hid = conn.execute(update_returning_stmnt).scalar() - n else: try: hid = session.execute(update_returning_stmnt).scalar() - n except (CompileError, OperationalError): # The RETURNING clause was added to SQLite in version 3.35.0. # Not using FOR UPDATE either, since that was added in 3.45.0, # and anyway does a whole-table lock select_stmt = select(table.c.hid_counter).where(table.c.id == history_id) hid = session.execute(select_stmt).scalar() session.execute(update_stmt) session.expire(self, ["hid_counter"]) return hid
[docs] def add_galaxy_session(self, galaxy_session, association=None): if association is None: self.galaxy_sessions.append(GalaxySessionToHistoryAssociation(galaxy_session, self)) else: self.galaxy_sessions.append(association)
[docs] def add_dataset(self, dataset, parent_id=None, genome_build=None, set_hid=True, quota=True): if isinstance(dataset, Dataset): dataset = HistoryDatasetAssociation(dataset=dataset) object_session(self).add(dataset) session = object_session(self) with transaction(session): session.commit() elif not isinstance(dataset, (HistoryDatasetAssociation, HistoryDatasetCollectionAssociation)): raise TypeError( "You can only add Dataset and HistoryDatasetAssociation instances to a history" f" ( you tried to add {str(dataset)} )." ) is_dataset = is_hda(dataset) if parent_id: for data in self.datasets: if data.id == parent_id: dataset.hid = data.hid break else: if set_hid: dataset.hid = self._next_hid() else: if set_hid: dataset.hid = self._next_hid() add_object_to_object_session(dataset, self) if quota and is_dataset and self.user: quota_source_info = dataset.dataset.quota_source_info if quota_source_info.use: self.user.adjust_total_disk_usage(dataset.quota_amount(self.user), quota_source_info.label) dataset.history = self if is_dataset and genome_build not in [None, "?"]: self.genome_build = genome_build dataset.history_id = self.id return dataset
[docs] def add_datasets( self, sa_session, datasets, parent_id=None, genome_build=None, set_hid=True, quota=True, flush=False ): """Optimized version of add_dataset above that minimizes database interactions when adding many datasets and collections to history at once. """ optimize = len(datasets) > 1 and parent_id is None and set_hid if optimize: self.__add_datasets_optimized(datasets, genome_build=genome_build) if quota and self.user: disk_usage = sum(d.get_total_size() for d in datasets if is_hda(d)) if disk_usage: quota_source_info = datasets[0].dataset.quota_source_info if quota_source_info.use: self.user.adjust_total_disk_usage(disk_usage, quota_source_info.label) sa_session.add_all(datasets) if flush: with transaction(sa_session): sa_session.commit() else: for dataset in datasets: self.add_dataset(dataset, parent_id=parent_id, genome_build=genome_build, set_hid=set_hid, quota=quota) sa_session.add(dataset) if flush: with transaction(sa_session): sa_session.commit()
def __add_datasets_optimized(self, datasets, genome_build=None): """Optimized version of add_dataset above that minimizes database interactions when adding many datasets to history at once under certain circumstances. """ n = len(datasets) base_hid = self._next_hid(n=n) set_genome = genome_build not in [None, "?"] for i, dataset in enumerate(datasets): dataset.hid = base_hid + i dataset.history = self dataset.history_id = cached_id(self) if set_genome and is_hda(dataset): self.genome_build = genome_build return datasets
[docs] def add_dataset_collection(self, history_dataset_collection, set_hid=True): if set_hid: history_dataset_collection.hid = self._next_hid() add_object_to_object_session(history_dataset_collection, self) history_dataset_collection.history = self # TODO: quota? self.dataset_collections.append(history_dataset_collection) return history_dataset_collection
[docs] def copy(self, name=None, target_user=None, activatable=False, all_datasets=False): """ Return a copy of this history using the given `name` and `target_user`. If `activatable`, copy only non-deleted datasets. If `all_datasets`, copy non-deleted, deleted, and purged datasets. """ name = name or self.name applies_to_quota = target_user != self.user # Create new history. new_history = History(name=name, user=target_user) db_session = object_session(self) db_session.add(new_history) db_session.flush([new_history]) # copy history tags and annotations (if copying user is not anonymous) if target_user: self.copy_item_annotation(db_session, self.user, self, target_user, new_history) new_history.copy_tags_from(target_user=target_user, source=self) # Copy HDAs. if activatable: hdas = self.activatable_datasets elif all_datasets: hdas = self.datasets else: hdas = self.active_datasets for hda in hdas: # Copy HDA. new_hda = hda.copy(flush=False) new_history.add_dataset(new_hda, set_hid=False, quota=applies_to_quota) if target_user: new_hda.copy_item_annotation(db_session, self.user, hda, target_user, new_hda) new_hda.copy_tags_from(target_user, hda) # Copy history dataset collections if all_datasets: hdcas = self.dataset_collections else: hdcas = self.active_dataset_collections for hdca in hdcas: new_hdca = hdca.copy(flush=False, element_destination=new_history, set_hid=False, minimize_copies=True) new_history.add_dataset_collection(new_hdca, set_hid=False) db_session.add(new_hdca) if target_user: new_hdca.copy_item_annotation(db_session, self.user, hdca, target_user, new_hdca) new_hdca.copy_tags_from(target_user, hdca) new_history.hid_counter = self.hid_counter with transaction(db_session): db_session.commit() return new_history
[docs] def get_dataset_by_hid(self, hid): if self._item_by_hid_cache is None: self._item_by_hid_cache = {dataset.hid: dataset for dataset in self.datasets} return self._item_by_hid_cache.get(hid)
@property def has_possible_members(self): return True @property def activatable_datasets(self): # This needs to be a list return [hda for hda in self.datasets if not hda.dataset.deleted] def _serialize(self, id_encoder, serialization_options): history_attrs = dict_for( self, create_time=self.create_time.__str__(), update_time=self.update_time.__str__(), name=unicodify(self.name), hid_counter=self.hid_counter, genome_build=self.genome_build, annotation=unicodify(get_item_annotation_str(object_session(self), self.user, self)), tags=self.make_tag_string_list(), ) serialization_options.attach_identifier(id_encoder, self, history_attrs) return history_attrs
[docs] def to_dict(self, view="collection", value_mapper=None): # Get basic value. rval = super().to_dict(view=view, value_mapper=value_mapper) if view == "element": rval["size"] = int(self.disk_size) return rval
@property def latest_export(self): exports = self.exports return exports and exports[0]
[docs] def unhide_datasets(self): for dataset in self.datasets: dataset.mark_unhidden()
[docs] def resume_paused_jobs(self): job = None for job in self.paused_jobs: job.resume(flush=False) if job is not None: # We'll flush once if there was a paused job session = object_session(job) with transaction(session): session.commit()
@property def paused_jobs(self): stmt = select(Job).where(Job.history_id == self.id, Job.state == Job.states.PAUSED) return object_session(self).scalars(stmt).all() @hybrid.hybrid_property def disk_size(self): """ Return the size in bytes of this history by summing the 'total_size's of all non-purged, unique datasets within it. """ # non-.expression part of hybrid.hybrid_property: called when an instance is the namespace (not the class) subq = ( select(HistoryDatasetAssociation.dataset_id, Dataset.total_size) .join(Dataset) .where(HistoryDatasetAssociation.history_id == self.id) .where(HistoryDatasetAssociation.purged != true()) .where(Dataset.purged != true()) .distinct() # unique datasets only .subquery() ) stmt = select(func.sum(subq.c.total_size)) return object_session(self).scalar(stmt) or 0 @disk_size.expression # type: ignore[no-redef] def disk_size(cls): """ Return a query scalar that will get any history's size in bytes by summing the 'total_size's of all non-purged, unique datasets within it. """ # .expression acts as a column_property and should return a scalar # first, get the distinct datasets within a history that are not purged hda_to_dataset_join = join( HistoryDatasetAssociation, Dataset, HistoryDatasetAssociation.dataset_id == Dataset.id ) distinct_datasets = ( select( # use labels here to better access from the query above HistoryDatasetAssociation.history_id.label("history_id"), Dataset.total_size.label("dataset_size"), Dataset.id.label("dataset_id"), ) .where(HistoryDatasetAssociation.purged != true()) .where(Dataset.purged != true()) .select_from(hda_to_dataset_join) # TODO: slow (in general) but most probably here - index total_size for easier sorting/distinct? .distinct() ) # postgres needs an alias on FROM distinct_datasets_alias = aliased(distinct_datasets.subquery(), name="datasets") # then, bind as property of history using the cls.id size_query = ( select(func.coalesce(func.sum(distinct_datasets_alias.c.dataset_size), 0)) .select_from(distinct_datasets_alias) .where(distinct_datasets_alias.c.history_id == cls.id) ) # label creates a scalar return size_query.label("disk_size") @property def disk_nice_size(self): """Returns human readable size of history on disk.""" return galaxy.util.nice_size(self.disk_size) def _active_dataset_and_roles_query(self): return ( select(HistoryDatasetAssociation) .where(HistoryDatasetAssociation.history_id == self.id) .where(not_(HistoryDatasetAssociation.deleted)) .order_by(HistoryDatasetAssociation.hid.asc()) .options( joinedload(HistoryDatasetAssociation.dataset) .joinedload(Dataset.actions) .joinedload(DatasetPermissions.role), joinedload(HistoryDatasetAssociation.tags), ) ) @property def active_datasets_and_roles(self): if not hasattr(self, "_active_datasets_and_roles"): stmt = self._active_dataset_and_roles_query() self._active_datasets_and_roles = object_session(self).scalars(stmt).unique().all() return self._active_datasets_and_roles @property def active_visible_datasets_and_roles(self): if not hasattr(self, "_active_visible_datasets_and_roles"): stmt = self._active_dataset_and_roles_query().where(HistoryDatasetAssociation.visible) self._active_visible_datasets_and_roles = object_session(self).scalars(stmt).unique().all() return self._active_visible_datasets_and_roles @property def active_visible_dataset_collections(self): if not hasattr(self, "_active_visible_dataset_collections"): stmt = ( select(HistoryDatasetCollectionAssociation) .where(HistoryDatasetCollectionAssociation.history_id == self.id) .where(not_(HistoryDatasetCollectionAssociation.deleted)) .where(HistoryDatasetCollectionAssociation.visible) .order_by(HistoryDatasetCollectionAssociation.hid.asc()) .options( joinedload(HistoryDatasetCollectionAssociation.collection), joinedload(HistoryDatasetCollectionAssociation.tags), ) ) self._active_visible_dataset_collections = object_session(self).scalars(stmt).unique().all() return self._active_visible_dataset_collections @property def active_contents(self): """Return all active contents ordered by hid.""" return self.contents_iter(types=["dataset", "dataset_collection"], deleted=False, visible=True)
[docs] def contents_iter(self, **kwds): """ Fetch filtered list of contents of history. """ default_contents_types = [ "dataset", ] types = kwds.get("types", default_contents_types) iters = [] if "dataset" in types: iters.append(self.__dataset_contents_iter(**kwds)) if "dataset_collection" in types: iters.append(self.__collection_contents_iter(**kwds)) return galaxy.util.merge_sorted_iterables(operator.attrgetter("hid"), *iters)
def __dataset_contents_iter(self, **kwds): return self.__filter_contents(HistoryDatasetAssociation, **kwds) def __collection_contents_iter(self, **kwds): return self.__filter_contents(HistoryDatasetCollectionAssociation, **kwds) def __filter_contents(self, content_class, **kwds): session = object_session(self) stmt = select(content_class).where(content_class.history_id == self.id).order_by(content_class.hid.asc()) deleted = galaxy.util.string_as_bool_or_none(kwds.get("deleted", None)) if deleted is not None: stmt = stmt.where(content_class.deleted == deleted) visible = galaxy.util.string_as_bool_or_none(kwds.get("visible", None)) if visible is not None: stmt = stmt.where(content_class.visible == visible) if "object_store_ids" in kwds: if content_class == HistoryDatasetAssociation: stmt = stmt.join(content_class.dataset).where(Dataset.object_store_id.in_(kwds.get("object_store_ids"))) # else ignoring object_store_ids on HDCAs... if "ids" in kwds: assert "object_store_ids" not in kwds ids = kwds["ids"] max_in_filter_length = kwds.get("max_in_filter_length", MAX_IN_FILTER_LENGTH) if len(ids) < max_in_filter_length: stmt = stmt.where(content_class.id.in_(ids)) else: return (content for content in session.scalars(stmt) if content.id in ids) return session.scalars(stmt)
[docs]class UserShareAssociation(RepresentById): user: Mapped[User]
[docs]class HistoryUserShareAssociation(Base, UserShareAssociation): __tablename__ = "history_user_share_association" id: Mapped[int] = mapped_column(primary_key=True) history_id: Mapped[int] = mapped_column(ForeignKey("history.id"), index=True, nullable=True) user_id: Mapped[int] = mapped_column(ForeignKey("galaxy_user.id"), index=True, nullable=True) user: Mapped["User"] = relationship() history: Mapped["History"] = relationship(back_populates="users_shared_with")
[docs]class UserRoleAssociation(Base, RepresentById): __tablename__ = "user_role_association" __table_args__ = (UniqueConstraint("user_id", "role_id"),) id: Mapped[int] = mapped_column(primary_key=True) user_id: Mapped[int] = mapped_column(ForeignKey("galaxy_user.id"), index=True) role_id: Mapped[int] = mapped_column(ForeignKey("role.id"), index=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) user: Mapped["User"] = relationship(back_populates="roles") role: Mapped["Role"] = relationship(back_populates="users") def __init__(self, user, role): add_object_to_object_session(self, user) self.user = user self.role = role
[docs]class GroupRoleAssociation(Base, RepresentById): __tablename__ = "group_role_association" __table_args__ = (UniqueConstraint("group_id", "role_id"),) id: Mapped[int] = mapped_column(primary_key=True) group_id: Mapped[int] = mapped_column(ForeignKey("galaxy_group.id"), index=True) role_id: Mapped[int] = mapped_column(ForeignKey("role.id"), index=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) group: Mapped["Group"] = relationship(back_populates="roles") role: Mapped["Role"] = relationship(back_populates="groups") def __init__(self, group, role): self.group = group ensure_object_added_to_session(self, object_in_session=group) self.role = role
[docs]class Role(Base, Dictifiable, RepresentById): __tablename__ = "role" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) name: Mapped[Optional[str]] = mapped_column(String(255), index=True, unique=True) description: Mapped[Optional[str]] = mapped_column(TEXT) type: Mapped[Optional[str]] = mapped_column(String(40), index=True) deleted: Mapped[Optional[bool]] = mapped_column(index=True, default=False) dataset_actions: Mapped[List["DatasetPermissions"]] = relationship(back_populates="role") groups: Mapped[List["GroupRoleAssociation"]] = relationship(back_populates="role") users: Mapped[List["UserRoleAssociation"]] = relationship(back_populates="role") dict_collection_visible_keys = ["id", "name"] dict_element_visible_keys = ["id", "name", "description", "type"] private_id = None
[docs] class types(str, Enum): PRIVATE = "private" SYSTEM = "system" USER = "user" ADMIN = "admin" SHARING = "sharing"
def __init__(self, name=None, description=None, type=types.SYSTEM, deleted=False): self.name = name self.description = description self.type = type self.deleted = deleted
[docs]class UserQuotaSourceUsage(Base, Dictifiable, RepresentById): __tablename__ = "user_quota_source_usage" __table_args__ = (UniqueConstraint("user_id", "quota_source_label", name="uqsu_unique_label_per_user"),) dict_element_visible_keys = ["disk_usage", "quota_source_label"] id: Mapped[int] = mapped_column(primary_key=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) quota_source_label: Mapped[Optional[str]] = mapped_column(String(32), index=True) # user had an index on disk_usage - does that make any sense? -John disk_usage: Mapped[Decimal] = mapped_column(Numeric(15, 0), default=0) user: Mapped[Optional["User"]] = relationship(back_populates="quota_source_usages")
[docs]class UserQuotaAssociation(Base, Dictifiable, RepresentById): __tablename__ = "user_quota_association" id: Mapped[int] = mapped_column(primary_key=True) user_id: Mapped[int] = mapped_column(ForeignKey("galaxy_user.id"), index=True, nullable=True) quota_id: Mapped[int] = mapped_column(ForeignKey("quota.id"), index=True, nullable=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) user: Mapped["User"] = relationship(back_populates="quotas") quota: Mapped["Quota"] = relationship(back_populates="users") dict_element_visible_keys = ["user"] def __init__(self, user, quota): add_object_to_object_session(self, user) self.user = user self.quota = quota
[docs]class GroupQuotaAssociation(Base, Dictifiable, RepresentById): __tablename__ = "group_quota_association" id: Mapped[int] = mapped_column(primary_key=True) group_id: Mapped[int] = mapped_column(ForeignKey("galaxy_group.id"), index=True, nullable=True) quota_id: Mapped[int] = mapped_column(ForeignKey("quota.id"), index=True, nullable=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) group: Mapped["Group"] = relationship(back_populates="quotas") quota: Mapped["Quota"] = relationship(back_populates="groups") dict_element_visible_keys = ["group"] def __init__(self, group, quota): add_object_to_object_session(self, group) self.group = group self.quota = quota
[docs]class Quota(Base, Dictifiable, RepresentById): __tablename__ = "quota" __table_args__ = (Index("ix_quota_quota_source_label", "quota_source_label"),) id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) name: Mapped[Optional[str]] = mapped_column(String(255), index=True, unique=True) description: Mapped[Optional[str]] = mapped_column(TEXT) bytes: Mapped[Optional[int]] = mapped_column(BigInteger) operation: Mapped[Optional[str]] = mapped_column(String(8)) deleted: Mapped[Optional[bool]] = mapped_column(index=True, default=False) quota_source_label: Mapped[Optional[str]] = mapped_column(String(32), default=None) default: Mapped[List["DefaultQuotaAssociation"]] = relationship( "DefaultQuotaAssociation", back_populates="quota", cascade_backrefs=False ) groups: Mapped[List["GroupQuotaAssociation"]] = relationship(back_populates="quota") users: Mapped[List["UserQuotaAssociation"]] = relationship(back_populates="quota") dict_collection_visible_keys = ["id", "name", "quota_source_label"] dict_element_visible_keys = [ "id", "name", "description", "bytes", "operation", "display_amount", "default", "users", "groups", "quota_source_label", ] valid_operations = ("+", "-", "=") def __init__(self, name=None, description=None, amount=0, operation="=", quota_source_label=None): self.name = name self.description = description if amount is None: self.bytes = -1 else: self.bytes = amount self.operation = operation self.quota_source_label = quota_source_label
[docs] def get_amount(self): if self.bytes == -1: return None return self.bytes
[docs] def set_amount(self, amount): if amount is None: self.bytes = -1 else: self.bytes = amount
amount = property(get_amount, set_amount) @property def display_amount(self): if self.bytes == -1: return "unlimited" else: return galaxy.util.nice_size(self.bytes)
[docs]class DefaultQuotaAssociation(Base, Dictifiable, RepresentById): __tablename__ = "default_quota_association" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) type: Mapped[Optional[str]] = mapped_column(String(32)) quota_id: Mapped[int] = mapped_column(ForeignKey("quota.id"), index=True, nullable=True) quota: Mapped["Quota"] = relationship(back_populates="default") dict_element_visible_keys = ["type"]
[docs] class types(str, Enum): UNREGISTERED = "unregistered" REGISTERED = "registered"
def __init__(self, type, quota): assert type in self.types.__members__.values(), "Invalid type" self.type = type self.quota = quota ensure_object_added_to_session(self, object_in_session=quota)
[docs]class DatasetPermissions(Base, RepresentById): __tablename__ = "dataset_permissions" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) action: Mapped[Optional[str]] = mapped_column(TEXT) dataset_id: Mapped[Optional[int]] = mapped_column(ForeignKey("dataset.id"), index=True) role_id: Mapped[Optional[int]] = mapped_column(ForeignKey("role.id"), index=True) dataset: Mapped[Optional["Dataset"]] = relationship(back_populates="actions") role: Mapped[Optional["Role"]] = relationship(back_populates="dataset_actions") def __init__(self, action, dataset, role=None, role_id=None): self.action = action add_object_to_object_session(self, dataset) self.dataset = dataset if role is not None: self.role = role else: self.role_id = role_id
[docs]class LibraryPermissions(Base, RepresentById): __tablename__ = "library_permissions" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) action: Mapped[Optional[str]] = mapped_column(TEXT) library_id: Mapped[Optional[int]] = mapped_column(ForeignKey("library.id"), index=True) role_id: Mapped[Optional[int]] = mapped_column(ForeignKey("role.id"), index=True) library: Mapped[Optional["Library"]] = relationship(back_populates="actions") role: Mapped[Optional["Role"]] = relationship() def __init__(self, action, library_item, role): self.action = action if isinstance(library_item, Library): self.library = library_item ensure_object_added_to_session(self, object_in_session=library_item) else: raise Exception(f"Invalid Library specified: {library_item.__class__.__name__}") self.role = role
[docs]class LibraryFolderPermissions(Base, RepresentById): __tablename__ = "library_folder_permissions" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) action: Mapped[Optional[str]] = mapped_column(TEXT) library_folder_id: Mapped[Optional[int]] = mapped_column(ForeignKey("library_folder.id"), index=True) role_id: Mapped[Optional[int]] = mapped_column(ForeignKey("role.id"), index=True) folder: Mapped[Optional["LibraryFolder"]] = relationship(back_populates="actions") role: Mapped[Optional["Role"]] = relationship() def __init__(self, action, library_item, role): self.action = action if isinstance(library_item, LibraryFolder): self.folder = library_item ensure_object_added_to_session(self, object_in_session=library_item) else: raise Exception(f"Invalid LibraryFolder specified: {library_item.__class__.__name__}") self.role = role
[docs]class LibraryDatasetPermissions(Base, RepresentById): __tablename__ = "library_dataset_permissions" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) action: Mapped[Optional[str]] = mapped_column(TEXT) library_dataset_id: Mapped[Optional[int]] = mapped_column(ForeignKey("library_dataset.id"), index=True) role_id: Mapped[Optional[int]] = mapped_column(ForeignKey("role.id"), index=True) library_dataset: Mapped[Optional["LibraryDataset"]] = relationship(back_populates="actions") role: Mapped[Optional["Role"]] = relationship() def __init__(self, action, library_item, role): self.action = action if isinstance(library_item, LibraryDataset): self.library_dataset = library_item ensure_object_added_to_session(self, object_in_session=library_item) else: raise Exception(f"Invalid LibraryDataset specified: {library_item.__class__.__name__}") self.role = role
[docs]class LibraryDatasetDatasetAssociationPermissions(Base, RepresentById): __tablename__ = "library_dataset_dataset_association_permissions" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) action: Mapped[Optional[str]] = mapped_column(TEXT) library_dataset_dataset_association_id: Mapped[int] = mapped_column( ForeignKey("library_dataset_dataset_association.id"), index=True, nullable=True ) role_id: Mapped[Optional[int]] = mapped_column(ForeignKey("role.id"), index=True) library_dataset_dataset_association: Mapped["LibraryDatasetDatasetAssociation"] = relationship( back_populates="actions" ) role: Mapped[Optional["Role"]] = relationship() def __init__(self, action, library_item, role): self.action = action if isinstance(library_item, LibraryDatasetDatasetAssociation): add_object_to_object_session(self, library_item) self.library_dataset_dataset_association = library_item else: raise Exception(f"Invalid LibraryDatasetDatasetAssociation specified: {library_item.__class__.__name__}") self.role = role
[docs]class DefaultUserPermissions(Base, RepresentById): __tablename__ = "default_user_permissions" id: Mapped[int] = mapped_column(primary_key=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) action: Mapped[Optional[str]] = mapped_column(TEXT) role_id: Mapped[Optional[int]] = mapped_column(ForeignKey("role.id"), index=True) user: Mapped[Optional["User"]] = relationship(back_populates="default_permissions") role: Mapped[Optional["Role"]] = relationship() def __init__(self, user, action, role): add_object_to_object_session(self, user) self.user = user self.action = action self.role = role
[docs]class DefaultHistoryPermissions(Base, RepresentById): __tablename__ = "default_history_permissions" id: Mapped[int] = mapped_column(primary_key=True) history_id: Mapped[Optional[int]] = mapped_column(ForeignKey("history.id"), index=True) action: Mapped[Optional[str]] = mapped_column(TEXT) role_id: Mapped[Optional[int]] = mapped_column(ForeignKey("role.id"), index=True) history: Mapped[Optional["History"]] = relationship(back_populates="default_permissions") role: Mapped[Optional["Role"]] = relationship() def __init__(self, history, action, role): add_object_to_object_session(self, history) self.history = history self.action = action self.role = role
[docs]class StorableObject:
[docs] def flush(self): if sa_session := object_session(self): with transaction(sa_session): sa_session.commit()
[docs]def setup_global_object_store_for_models(object_store: "BaseObjectStore") -> None: Dataset.object_store = object_store
[docs]class Dataset(Base, StorableObject, Serializable): __tablename__ = "dataset" __table_args__ = (UniqueConstraint("uuid", name="uq_uuid_column"),) id: Mapped[int] = mapped_column(primary_key=True) job_id: Mapped[Optional[int]] = mapped_column(ForeignKey("job.id"), index=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(index=True, default=now, onupdate=now, nullable=True) state: Mapped[Optional[str]] = mapped_column(TrimmedString(64), index=True) deleted: Mapped[Optional[bool]] = mapped_column(index=True, default=False) purged: Mapped[Optional[bool]] = mapped_column(index=True, default=False) purgable: Mapped[Optional[bool]] = mapped_column(default=True) object_store_id: Mapped[Optional[str]] = mapped_column(TrimmedString(255), index=True) external_filename: Mapped[Optional[str]] = mapped_column(TEXT) _extra_files_path: Mapped[Optional[str]] = mapped_column(TEXT) created_from_basename: Mapped[Optional[str]] = mapped_column(TEXT) file_size: Mapped[Optional[Decimal]] = mapped_column(Numeric(15, 0)) total_size: Mapped[Optional[Decimal]] = mapped_column(Numeric(15, 0)) uuid: Mapped[Optional[Union[UUID, str]]] = mapped_column(UUIDType(), unique=True) actions: Mapped[List["DatasetPermissions"]] = relationship(back_populates="dataset") job: Mapped[Optional["Job"]] = relationship(primaryjoin=(lambda: Dataset.job_id == Job.id)) active_history_associations: Mapped[List["HistoryDatasetAssociation"]] = relationship( primaryjoin=( lambda: and_( Dataset.id == HistoryDatasetAssociation.dataset_id, # type: ignore[attr-defined] HistoryDatasetAssociation.deleted == false(), # type: ignore[has-type] HistoryDatasetAssociation.purged == false(), # type: ignore[arg-type] ) ), viewonly=True, ) purged_history_associations: Mapped[List["HistoryDatasetAssociation"]] = relationship( primaryjoin=( lambda: and_( Dataset.id == HistoryDatasetAssociation.dataset_id, # type: ignore[attr-defined] HistoryDatasetAssociation.purged == true(), # type: ignore[arg-type] ) ), viewonly=True, ) active_library_associations: Mapped[List["LibraryDatasetDatasetAssociation"]] = relationship( primaryjoin=( lambda: and_( Dataset.id == LibraryDatasetDatasetAssociation.dataset_id, # type: ignore[attr-defined] LibraryDatasetDatasetAssociation.deleted == false(), # type: ignore[has-type] ) ), viewonly=True, ) hashes: Mapped[List["DatasetHash"]] = relationship(back_populates="dataset", cascade_backrefs=False) sources: Mapped[List["DatasetSource"]] = relationship(back_populates="dataset") history_associations: Mapped[List["HistoryDatasetAssociation"]] = relationship( back_populates="dataset", cascade_backrefs=False ) library_associations: Mapped[List["LibraryDatasetDatasetAssociation"]] = relationship( primaryjoin=(lambda: LibraryDatasetDatasetAssociation.table.c.dataset_id == Dataset.id), back_populates="dataset", cascade_backrefs=False, ) # failed_metadata is only valid as DatasetInstance state currently states = DatasetState non_ready_states = (states.NEW, states.UPLOAD, states.QUEUED, states.RUNNING, states.SETTING_METADATA) ready_states = tuple(set(states.__members__.values()) - set(non_ready_states)) valid_input_states = tuple( set(states.__members__.values()) - {states.ERROR, states.DISCARDED, states.FAILED_METADATA} ) no_data_states = (states.PAUSED, states.DEFERRED, states.DISCARDED, *non_ready_states) terminal_states = ( states.OK, states.EMPTY, states.ERROR, states.DEFERRED, states.DISCARDED, states.FAILED_METADATA, )
[docs] class conversion_messages(str, Enum): PENDING = "pending" NO_DATA = "no data" NO_CHROMOSOME = "no chromosome" NO_CONVERTER = "no converter" NO_TOOL = "no tool" DATA = "data" ERROR = "error" OK = "ok"
permitted_actions = get_permitted_actions(filter="DATASET") file_path = "/tmp/" object_store: ClassVar[Optional["BaseObjectStore"]] = ( None # This get initialized in mapping.py (method init) by app.py ) engine = None def __init__( self, id=None, state=None, external_filename=None, extra_files_path=None, file_size=None, purgable=True, uuid=None, ): self.id = id self.uuid = get_uuid(uuid) self.state = state self.deleted = False self.purged = False self.purgable = purgable self.external_filename = external_filename self.external_extra_files_path = None self._extra_files_path = extra_files_path self.file_size = file_size self.sources = [] self.hashes = [] @property def is_new(self): return self.state == self.states.NEW
[docs] def in_ready_state(self): return self.state in self.ready_states
@property def shareable(self) -> bool: """Return True if placed into an objectstore not labeled as ``private``.""" if self.external_filename: return True else: object_store = self._assert_object_store_set() return not object_store.is_private(self)
[docs] def ensure_shareable(self): if not self.shareable: raise Exception(CANNOT_SHARE_PRIVATE_DATASET_MESSAGE)
[docs] def get_file_name(self, sync_cache: bool = True) -> str: if self.purged: log.warning(f"Attempt to get file name of purged dataset {self.id}") return "" if not self.external_filename: object_store = self._assert_object_store_set() if object_store.exists(self): file_name = object_store.get_filename(self, sync_cache=sync_cache) else: file_name = "" if not file_name and self.state not in (self.states.NEW, self.states.QUEUED): # Queued datasets can be assigned an object store and have a filename, but they aren't guaranteed to. # Anything after queued should have a file name. log.warning(f"Failed to determine file name for dataset {self.id}") return file_name else: filename = self.external_filename # Make filename absolute return os.path.abspath(filename)
@property def quota_source_label(self): return self.quota_source_info.label @property def quota_source_info(self): object_store_id = self.object_store_id quota_source_map = self.object_store.get_quota_source_map() return quota_source_map.get_quota_source_info(object_store_id) @property def device_source_label(self): return self.device_source_info.label @property def device_source_info(self): object_store_id = self.object_store_id device_source_map = self.object_store.get_quota_source_map() return device_source_map.get_device_source_info(object_store_id)
[docs] def set_file_name(self, filename): if not filename: self.external_filename = None else: self.external_filename = filename
def _assert_object_store_set(self) -> "BaseObjectStore": assert self.object_store is not None, "Object Store has not been initialized" return self.object_store
[docs] def get_extra_files_path(self) -> str: # Unlike get_file_name - external_extra_files_path is not backed by an # actual database column so if SA instantiates this object - the # attribute won't exist yet. if not getattr(self, "external_extra_files_path", None): object_store = self._assert_object_store_set() if object_store.exists(self, dir_only=True, extra_dir=self._extra_files_rel_path): return object_store.get_filename(self, dir_only=True, extra_dir=self._extra_files_rel_path) return object_store.construct_path(self, dir_only=True, extra_dir=self._extra_files_rel_path, in_cache=True) else: return os.path.abspath(self.external_extra_files_path)
[docs] def create_extra_files_path(self): if not self.extra_files_path_exists(): self.object_store.create(self, dir_only=True, extra_dir=self._extra_files_rel_path)
[docs] def set_extra_files_path(self, extra_files_path): if not extra_files_path: self.external_extra_files_path = None else: self.external_extra_files_path = extra_files_path
extra_files_path = property(get_extra_files_path, set_extra_files_path)
[docs] def extra_files_path_exists(self): return self.object_store.exists(self, extra_dir=self._extra_files_rel_path, dir_only=True)
@property def store_by(self): store_by = self.object_store.get_store_by(self) return store_by
[docs] def extra_files_path_name_from(self, object_store): if (store_by := self.store_by) is not None: return f"dataset_{getattr(self, store_by)}_files" else: return None
@property def extra_files_path_name(self): return self.extra_files_path_name_from(self.object_store) @property def _extra_files_rel_path(self): return self._extra_files_path or self.extra_files_path_name def _calculate_size(self) -> int: if self.external_filename: try: return os.path.getsize(self.external_filename) except OSError: return 0 assert self.object_store return self.object_store.size(self) @overload def get_size(self, nice_size: Literal[False], calculate_size: bool = True) -> int: ... @overload def get_size(self, nice_size: Literal[True], calculate_size: bool = True) -> str: ...
[docs] def get_size(self, nice_size: bool = False, calculate_size: bool = True) -> Union[int, str]: """Returns the size of the data on disk""" if self.file_size: if nice_size: return galaxy.util.nice_size(self.file_size) else: return cast(int, self.file_size) elif calculate_size: # Hopefully we only reach this branch in sessionless mode if nice_size: return galaxy.util.nice_size(self._calculate_size()) else: return self._calculate_size() else: return cast(int, self.file_size) or 0
[docs] def set_size(self, no_extra_files=False): """Sets the size of the data on disk. If the caller is sure there are no extra files, pass no_extra_files as True to optimize subsequent calls to get_total_size or set_total_size - potentially avoiding both a database flush and check against the file system. """ if not self.file_size: self.file_size = self._calculate_size() if no_extra_files: self.total_size = self.file_size
[docs] def get_total_size(self): if self.total_size is not None: return self.total_size # for backwards compatibility, set if unset self.set_total_size() db_session = object_session(self) with transaction(db_session): db_session.commit() return self.total_size
[docs] def set_total_size(self): if self.file_size is None: self.set_size() self.total_size = self.file_size or 0 if (rel_path := self._extra_files_rel_path) is not None: if self.object_store.exists(self, extra_dir=rel_path, dir_only=True): for root, _, files in os.walk(self.extra_files_path): self.total_size += sum( os.path.getsize(os.path.join(root, file)) for file in files if os.path.exists(os.path.join(root, file)) ) return self.total_size
[docs] def has_data(self): """Detects whether there is any data""" return not self.is_new and self.get_size() > 0
[docs] def mark_deleted(self): self.deleted = True
# FIXME: sqlalchemy will replace this def _delete(self): """Remove the file that corresponds to this data""" self.object_store.delete(self) @property def user_can_purge(self): return ( self.purged is False and not bool(self.library_associations) and len(self.history_associations) == len(self.purged_history_associations) )
[docs] def full_delete(self): """Remove the file and extra files, marks deleted and purged""" # os.unlink( self.file_name ) try: self.object_store.delete(self) except galaxy.exceptions.ObjectNotFound: pass if (rel_path := self._extra_files_rel_path) is not None: if self.object_store.exists(self, extra_dir=rel_path, dir_only=True): self.object_store.delete(self, entire_dir=True, extra_dir=rel_path, dir_only=True) # TODO: purge metadata files self.deleted = True self.purged = True
[docs] def get_access_roles(self, security_agent): roles = [] for dp in self.actions: if dp.action == security_agent.permitted_actions.DATASET_ACCESS.action: roles.append(dp.role) return roles
[docs] def get_manage_permissions_roles(self, security_agent): roles = [] for dp in self.actions: if dp.action == security_agent.permitted_actions.DATASET_MANAGE_PERMISSIONS.action: roles.append(dp.role) return roles
[docs] def has_manage_permissions_roles(self, security_agent): for dp in self.actions: if dp.action == security_agent.permitted_actions.DATASET_MANAGE_PERMISSIONS.action: return True return False
def _serialize(self, id_encoder, serialization_options): # serialize Dataset objects only for jobs that can actually modify these models. assert serialization_options.serialize_dataset_objects def to_int(n) -> Optional[int]: return int(n) if n is not None else None rval = dict_for( self, state=self.state, deleted=self.deleted, purged=self.purged, external_filename=self.external_filename, _extra_files_path=self._extra_files_path, file_size=to_int(self.file_size), object_store_id=self.object_store_id, total_size=to_int(self.total_size), created_from_basename=self.created_from_basename, uuid=str(self.uuid or "") or None, hashes=[h.serialize(id_encoder, serialization_options) for h in self.hashes], sources=[s.serialize(id_encoder, serialization_options) for s in self.sources], ) serialization_options.attach_identifier(id_encoder, self, rval) return rval
[docs]class DatasetSource(Base, Dictifiable, Serializable): __tablename__ = "dataset_source" id: Mapped[int] = mapped_column(primary_key=True) dataset_id: Mapped[Optional[int]] = mapped_column(ForeignKey("dataset.id"), index=True) source_uri: Mapped[Optional[str]] = mapped_column(TEXT) extra_files_path: Mapped[Optional[str]] = mapped_column(TEXT) transform: Mapped[Optional[TRANSFORM_ACTIONS]] = mapped_column(MutableJSONType) dataset: Mapped[Optional["Dataset"]] = relationship(back_populates="sources") hashes: Mapped[List["DatasetSourceHash"]] = relationship(back_populates="source") dict_collection_visible_keys = ["id", "source_uri", "extra_files_path", "transform"] dict_element_visible_keys = [ "id", "source_uri", "extra_files_path", "transform", ] # TODO: implement to_dict and add hashes... def _serialize(self, id_encoder, serialization_options): rval = dict_for( self, source_uri=self.source_uri, extra_files_path=self.extra_files_path, transform=self.transform, hashes=[h.serialize(id_encoder, serialization_options) for h in self.hashes], ) serialization_options.attach_identifier(id_encoder, self, rval) return rval
[docs] def copy(self) -> "DatasetSource": new_source = DatasetSource() new_source.source_uri = self.source_uri new_source.extra_files_path = self.extra_files_path new_source.transform = self.transform new_source.hashes = [h.copy() for h in self.hashes] return new_source
[docs]class HasHashFunctionName: hash_function: Mapped[Optional[str]] @property def hash_func_name(self) -> HashFunctionNameEnum: as_str = self.hash_function assert as_str return HashFunctionNameEnum(self.hash_function)
[docs]class DatasetSourceHash(Base, Serializable, HasHashFunctionName): __tablename__ = "dataset_source_hash" id: Mapped[int] = mapped_column(primary_key=True) dataset_source_id: Mapped[Optional[int]] = mapped_column(ForeignKey("dataset_source.id"), index=True) hash_function: Mapped[Optional[str]] = mapped_column(TEXT) hash_value: Mapped[Optional[str]] = mapped_column(TEXT) source: Mapped[Optional["DatasetSource"]] = relationship(back_populates="hashes") def _serialize(self, id_encoder, serialization_options): rval = dict_for( self, hash_function=self.hash_function, hash_value=self.hash_value, ) serialization_options.attach_identifier(id_encoder, self, rval) return rval
[docs] def copy(self) -> "DatasetSourceHash": new_hash = DatasetSourceHash() new_hash.hash_function = self.hash_function new_hash.hash_value = self.hash_value return new_hash
[docs]class DatasetHash(Base, Dictifiable, Serializable, HasHashFunctionName): __tablename__ = "dataset_hash" id: Mapped[int] = mapped_column(primary_key=True) dataset_id: Mapped[Optional[int]] = mapped_column(ForeignKey("dataset.id"), index=True) hash_function: Mapped[Optional[str]] = mapped_column(TEXT) hash_value: Mapped[Optional[str]] = mapped_column(TEXT) extra_files_path: Mapped[Optional[str]] = mapped_column(TEXT) dataset: Mapped[Optional["Dataset"]] = relationship(back_populates="hashes") dict_collection_visible_keys = ["id", "hash_function", "hash_value", "extra_files_path"] dict_element_visible_keys = ["id", "hash_function", "hash_value", "extra_files_path"] def _serialize(self, id_encoder, serialization_options): rval = dict_for( self, hash_function=self.hash_function, hash_value=self.hash_value, extra_files_path=self.extra_files_path, ) serialization_options.attach_identifier(id_encoder, self, rval) return rval
[docs] def copy(self) -> "DatasetHash": new_hash = DatasetHash() new_hash.hash_function = self.hash_function new_hash.hash_value = self.hash_value new_hash.extra_files_path = self.extra_files_path return new_hash
@property def hash_func_name(self) -> HashFunctionNameEnum: as_str = self.hash_function assert as_str return HashFunctionNameEnum(self.hash_function)
DescribesHash = Union[DatasetSourceHash, DatasetHash]
[docs]def datatype_for_extension(extension, datatypes_registry=None) -> "Data": if extension is not None: extension = extension.lower() if datatypes_registry is None: datatypes_registry = _get_datatypes_registry() if not extension or extension == "auto" or extension == "_sniff_": extension = "data" ret = datatypes_registry.get_datatype_by_extension(extension) if ret is None: log.warning(f"Datatype class not found for extension '{extension}'") return datatypes_registry.get_datatype_by_extension("data") return ret
[docs]class DatasetInstance(RepresentById, UsesCreateAndUpdateTime, _HasTable): """A base class for all 'dataset instances', HDAs, LDDAs, etc""" states = Dataset.states _state: Optional[str] conversion_messages = Dataset.conversion_messages permitted_actions = Dataset.permitted_actions purged: bool creating_job_associations: List[Union[JobToOutputDatasetCollectionAssociation, JobToOutputDatasetAssociation]] copied_from_history_dataset_association: Optional["HistoryDatasetAssociation"] copied_from_library_dataset_dataset_association: Optional["LibraryDatasetDatasetAssociation"] dependent_jobs: List[JobToInputLibraryDatasetAssociation] implicitly_converted_datasets: List["ImplicitlyConvertedDatasetAssociation"] implicitly_converted_parent_datasets: List["ImplicitlyConvertedDatasetAssociation"] validated_states = DatasetValidatedState
[docs] def __init__( self, id=None, hid=None, name=None, info=None, blurb=None, peek=None, tool_version=None, extension=None, dbkey=None, metadata=None, history=None, dataset=None, deleted=False, designation=None, parent_id=None, validated_state=DatasetValidatedState.UNKNOWN, validated_state_message=None, visible=True, create_dataset=False, sa_session=None, extended_metadata=None, flush=True, metadata_deferred=False, creating_job_id=None, copied_from_history_dataset_association=None, copied_from_library_dataset_dataset_association=None, ): self.name = name or "Unnamed dataset" self.id = id self.info = info self.blurb = blurb self.peek = peek self.tool_version = tool_version self.extension = extension self.designation = designation # set private variable to None here, since the attribute may be needed in by MetadataCollection.__init__ self._metadata = None self.metadata = metadata or {} self.metadata_deferred = metadata_deferred self.extended_metadata = extended_metadata if ( dbkey ): # dbkey is stored in metadata, only set if non-zero, or else we could clobber one supplied by input 'metadata' self._metadata["dbkey"] = listify(dbkey) self.deleted = deleted self.visible = visible self.validated_state = validated_state self.validated_state_message = validated_state_message # Relationships if copied_from_history_dataset_association: self.copied_from_history_dataset_association_id = copied_from_history_dataset_association.id if copied_from_library_dataset_dataset_association: self.copied_from_library_dataset_dataset_association_id = copied_from_library_dataset_dataset_association.id if not dataset and create_dataset: # Had to pass the sqlalchemy session in order to create a new dataset dataset = Dataset(state=Dataset.states.NEW) dataset.job_id = creating_job_id if flush: sa_session.add(dataset) with transaction(sa_session): sa_session.commit() elif dataset: add_object_to_object_session(self, dataset) self.dataset = dataset ensure_object_added_to_session(self, object_in_session=dataset) self.parent_id = parent_id
@property def peek(self): return self._peek @peek.setter def peek(self, peek): self._peek = unicodify(peek, strip_null=True) @property def ext(self): return self.extension @property def has_deferred_data(self): return self.dataset.state == Dataset.states.DEFERRED @property def deferred_source_uri(self): if self.has_deferred_data and self.sources: # Assuming the first source is the deferred source return self.sources[0].source_uri return None @property def state(self): # self._state holds state that should only affect this particular dataset association, not the dataset state itself if self._state: return self._state return self.dataset.state @state.setter def state(self, state: Optional[DatasetState]): if state != self.state: if state in (DatasetState.FAILED_METADATA, DatasetState.SETTING_METADATA): self._state = state else: self.set_metadata_success_state() sa_session = object_session(self) if sa_session: sa_session.add(self.dataset) self.dataset.state = state
[docs] def set_metadata_success_state(self): self._state = None
[docs] def get_object_store_id(self): return self.dataset.object_store_id
object_store_id = property(get_object_store_id)
[docs] def get_quota_source_label(self): return self.dataset.quota_source_label
quota_source_label = property(get_quota_source_label)
[docs] def set_skipped(self, object_store_populator: "ObjectStorePopulator") -> None: assert self.dataset object_store_populator.set_object_store_id(self) self.extension = "expression.json" self.state = self.states.OK self.blurb = "skipped" self.visible = False null = json.dumps(None) with open(self.dataset.get_file_name(), "w") as out: out.write(null) self.peek = null self.set_total_size()
[docs] def get_file_name(self, sync_cache: bool = True) -> str: if self.dataset.purged: return "" return self.dataset.get_file_name(sync_cache=sync_cache)
[docs] def set_file_name(self, filename: str): return self.dataset.set_file_name(filename)
@property def extra_files_path(self): return self.dataset.extra_files_path
[docs] def extra_files_path_exists(self): return self.dataset.extra_files_path_exists()
@property def datatype(self) -> "Data": return datatype_for_extension(self.extension)
[docs] def get_metadata(self): # using weakref to store parent (to prevent circ ref), # does a Session.clear() cause parent to be invalidated, while still copying over this non-database attribute? if not hasattr(self, "_metadata_collection") or self._metadata_collection.parent != self: self._metadata_collection = galaxy.model.metadata.MetadataCollection(self) return self._metadata_collection
@property def set_metadata_requires_flush(self): return self.metadata.requires_dataset_id
[docs] def set_metadata(self, bunch): # Needs to accept a MetadataCollection, a bunch, or a dict self._metadata = self.metadata.make_dict_copy(bunch)
metadata = property(get_metadata, set_metadata) @property def has_metadata_files(self): return len(self.metadata_file_types) > 0 @property def metadata_file_types(self): meta_types = [] for meta_type in self.metadata.spec.keys(): if isinstance(self.metadata.spec[meta_type].param, galaxy.model.metadata.FileParameter): meta_types.append(meta_type) return meta_types
[docs] def get_metadata_file_paths_and_extensions(self) -> List[Tuple[str, str]]: metadata = self.metadata metadata_files = [] for metadata_name in self.metadata_file_types: file_ext = metadata.spec[metadata_name].file_ext metadata_file = metadata[metadata_name] if metadata_file: path = metadata_file.get_file_name() metadata_files.append((file_ext, path)) return metadata_files
# This provide backwards compatibility with using the old dbkey # field in the database. That field now maps to "old_dbkey" (see mapping.py).
[docs] def get_dbkey(self): dbkey = self.metadata.dbkey if not isinstance(dbkey, list): dbkey = [dbkey] if dbkey in [[None], []]: return "?" return dbkey[0]
[docs] def set_dbkey(self, value): if "dbkey" in self.datatype.metadata_spec: if not isinstance(value, list): self.metadata.dbkey = [value]
dbkey = property(get_dbkey, set_dbkey)
[docs] def ok_to_edit_metadata(self): """ Prevent modifying metadata when dataset is queued or running as input/output: return `False` if there exists an associated job with a non-terminal state. """ def exists_clause(assoc_model): return ( select(assoc_model.job_id) .join(Job) .where(assoc_model.dataset_id == self.id) .where(Job.state.not_in(Job.terminal_states)) .exists() ) stmt = select( or_( exists_clause(JobToInputDatasetAssociation), exists_clause(JobToOutputDatasetAssociation), ) ) return not object_session(self).scalar(stmt)
[docs] def change_datatype(self, new_ext): self.clear_associated_files() _get_datatypes_registry().change_datatype(self, new_ext)
[docs] def get_size(self, nice_size=False, calculate_size=True): """Returns the size of the data on disk""" if nice_size: return galaxy.util.nice_size(self.dataset.get_size(calculate_size=calculate_size)) return self.dataset.get_size(calculate_size=calculate_size)
[docs] def set_size(self, **kwds): """Sets and gets the size of the data on disk""" return self.dataset.set_size(**kwds)
[docs] def get_total_size(self): return self.dataset.get_total_size()
[docs] def set_total_size(self): return self.dataset.set_total_size()
[docs] def has_data(self): """Detects whether there is any data""" return self.dataset.has_data()
[docs] def get_created_from_basename(self): return self.dataset.created_from_basename
[docs] def set_created_from_basename(self, created_from_basename): if self.dataset.created_from_basename is not None: raise Exception("Underlying dataset already has a created_from_basename set.") self.dataset.created_from_basename = created_from_basename
created_from_basename = property(get_created_from_basename, set_created_from_basename) @property def sources(self): return self.dataset.sources @property def hashes(self): return self.dataset.hashes
[docs] def get_mime(self): """Returns the mime type of the data""" try: return _get_datatypes_registry().get_mimetype_by_extension(self.extension.lower()) except AttributeError: # extension is None return "data"
[docs] def set_peek(self, line_count=None, **kwd): try: # Certain datatype's set_peek methods contain a line_count argument return self.datatype.set_peek(self, line_count=line_count, **kwd) except TypeError: # ... and others don't return self.datatype.set_peek(self, **kwd) except Exception: # Never fail peek setting, but do log exception so datatype logic can be fixed log.exception("Setting peek failed")
[docs] def init_meta(self, copy_from=None): return self.datatype.init_meta(self, copy_from=copy_from)
[docs] def set_meta(self, **kwd): self.clear_associated_files(metadata_safe=True) return self.datatype.set_meta(self, **kwd)
[docs] def missing_meta(self, **kwd): return self.datatype.missing_meta(self, **kwd)
[docs] def as_display_type(self, type, **kwd): return self.datatype.as_display_type(self, type, **kwd)
[docs] def display_peek(self): try: return self.datatype.display_peek(self) except Exception: log.exception("Error occurred while generating dataset peek") return None
[docs] def display_name(self): return self.datatype.display_name(self)
[docs] def display_info(self): return self.datatype.display_info(self)
[docs] def get_converted_files_by_type(self, file_type): for assoc in self.implicitly_converted_datasets: if not assoc.deleted and assoc.type == file_type: item = assoc.dataset or assoc.dataset_ldda if not item.deleted and item.state in Dataset.valid_input_states: return item return None
[docs] def get_converted_dataset_deps(self, trans, target_ext): """ Returns dict of { "dependency" => HDA } """ # List of string of dependencies try: depends_list = trans.app.datatypes_registry.converter_deps[self.extension][target_ext] except KeyError: depends_list = [] return {dep: self.get_converted_dataset(trans, dep) for dep in depends_list}
[docs] def get_converted_dataset(self, trans, target_ext, target_context=None, history=None): """ Return converted dataset(s) if they exist, along with a dict of dependencies. If not converted yet, do so and return None (the first time). If unconvertible, raise exception. """ # See if we can convert the dataset if target_ext not in self.get_converter_types(): raise NoConverterException(f"Conversion from '{self.extension}' to '{target_ext}' not possible") # See if converted dataset already exists, either in metadata in conversions. converted_dataset = self.get_metadata_dataset(target_ext) if converted_dataset: return converted_dataset converted_dataset = self.get_converted_files_by_type(target_ext) if converted_dataset: return converted_dataset deps = {} # List of string of dependencies try: depends_list = trans.app.datatypes_registry.converter_deps[self.extension][target_ext] except KeyError: depends_list = [] # Conversion is possible but hasn't been done yet, run converter. # Check if we have dependencies try: for dependency in depends_list: dep_dataset = self.get_converted_dataset(trans, dependency) if dep_dataset is None: # None means converter is running first time return None elif dep_dataset.state == Job.states.ERROR: raise ConverterDependencyException(f"A dependency ({dependency}) was in an error state.") elif dep_dataset.state != Job.states.OK: # Pending return None deps[dependency] = dep_dataset except NoConverterException: raise NoConverterException(f"A dependency ({dependency}) is missing a converter.") except KeyError: pass # No deps return next( iter( self.datatype.convert_dataset( trans, self, target_ext, return_output=True, visible=False, deps=deps, target_context=target_context, history=history, ).values() ) )
[docs] def attach_implicitly_converted_dataset(self, session, new_dataset, target_ext: str): new_dataset.name = self.name self.copy_attributes(new_dataset) assoc = ImplicitlyConvertedDatasetAssociation( parent=self, file_type=target_ext, dataset=new_dataset, metadata_safe=False ) session.add(new_dataset) session.add(assoc)
[docs] def copy_attributes(self, new_dataset): """ Copies attributes to a new datasets, used for implicit conversions """
[docs] def get_metadata_dataset(self, dataset_ext): """ Returns an HDA that points to a metadata file which contains a converted data with the requested extension. """ for name, value in self.metadata.items(): # HACK: MetadataFile objects do not have a type/ext, so need to use metadata name # to determine type. if dataset_ext == "bai" and name == "bam_index" and isinstance(value, MetadataFile): # HACK: MetadataFile objects cannot be used by tools, so return # a fake HDA that points to metadata file. fake_dataset = Dataset(state=Dataset.states.OK, external_filename=value.get_file_name()) fake_hda = HistoryDatasetAssociation(dataset=fake_dataset) return fake_hda
[docs] def clear_associated_files(self, metadata_safe=False, purge=False): raise Exception("Unimplemented")
[docs] def get_converter_types(self): return self.datatype.get_converter_types(self, _get_datatypes_registry())
[docs] def can_convert_to(self, format): return format in self.get_converter_types()
[docs] def find_conversion_destination( self, accepted_formats: List[str], **kwd ) -> Tuple[bool, Optional[str], Optional["DatasetInstance"]]: """Returns ( target_ext, existing converted dataset )""" return self.datatype.find_conversion_destination( self, accepted_formats, _get_datatypes_registry(), **kwd # type:ignore[arg-type] )
[docs] def add_validation_error(self, validation_error): self.validation_errors.append(validation_error)
[docs] def extend_validation_errors(self, validation_errors): self.validation_errors.extend(validation_errors)
[docs] def mark_deleted(self): self.deleted = True
[docs] def mark_undeleted(self): self.deleted = False
[docs] def mark_unhidden(self): self.visible = True
[docs] def undeletable(self): if self.purged: return False return True
@property def is_ok(self): return self.state == self.states.OK @property def is_pending(self): """ Return true if the dataset is neither ready nor in error """ return self.state in ( self.states.NEW, self.states.UPLOAD, self.states.QUEUED, self.states.RUNNING, self.states.SETTING_METADATA, ) @property def source_library_dataset(self): def get_source(dataset): if isinstance(dataset, LibraryDatasetDatasetAssociation): if dataset.library_dataset: return (dataset, dataset.library_dataset) if dataset.copied_from_library_dataset_dataset_association: source = get_source(dataset.copied_from_library_dataset_dataset_association) if source: return source if dataset.copied_from_history_dataset_association: source = get_source(dataset.copied_from_history_dataset_association) if source: return source return (None, None) return get_source(self) @property def source_dataset_chain(self): def _source_dataset_chain(dataset, lst): try: cp_from_ldda = dataset.copied_from_library_dataset_dataset_association if cp_from_ldda: lst.append((cp_from_ldda, "(Data Library)")) return _source_dataset_chain(cp_from_ldda, lst) except Exception as e: log.warning(e) try: cp_from_hda = dataset.copied_from_history_dataset_association if cp_from_hda: lst.append((cp_from_hda, cp_from_hda.history.name)) return _source_dataset_chain(cp_from_hda, lst) except Exception as e: log.warning(e) return lst return _source_dataset_chain(self, []) @property def creating_job(self) -> Optional[Job]: # TODO this should work with `return self.dataset.job` (revise failing unit tests) creating_job_associations = None if self.creating_job_associations: creating_job_associations = self.creating_job_associations else: inherit_chain = self.source_dataset_chain if inherit_chain: creating_job_associations = inherit_chain[-1][0].creating_job_associations if creating_job_associations: return creating_job_associations[0].job return None
[docs] def get_display_applications(self, trans): return self.datatype.get_display_applications_by_dataset(self, trans)
[docs] def get_datasources(self, trans): """ Returns datasources for dataset; if datasources are not available due to indexing, indexing is started. Return value is a dictionary with entries of type (<datasource_type> : {<datasource_name>, <indexing_message>}). """ data_sources_dict = {} msg = None for source_type, source_list in self.datatype.data_sources.items(): data_source = None if source_type == "data_standalone": # Nothing to do. msg = None data_source = source_list else: # Convert. if isinstance(source_list, str): source_list = [source_list] # Loop through sources until viable one is found. for source in source_list: msg = self.convert_dataset(trans, source) # No message or PENDING means that source is viable. No # message indicates conversion was done and is successful. if not msg or msg == self.conversion_messages.PENDING: data_source = source break # Store msg. data_sources_dict[source_type] = {"name": data_source, "message": msg} return data_sources_dict
[docs] def convert_dataset(self, trans, target_type): """ Converts a dataset to the target_type and returns a message indicating status of the conversion. None is returned to indicate that dataset was converted successfully. """ # Get converted dataset; this will start the conversion if necessary. try: converted_dataset = self.get_converted_dataset(trans, target_type) except NoConverterException: return self.conversion_messages.NO_CONVERTER except ConverterDependencyException as dep_error: return {"kind": self.conversion_messages.ERROR, "message": dep_error.value} # Check dataset state and return any messages. msg = None if converted_dataset and converted_dataset.state == Dataset.states.ERROR: stmt = select(JobToOutputDatasetAssociation.job_id).filter_by(dataset_id=converted_dataset.id).limit(1) job_id = trans.sa_session.scalars(stmt).first() job = trans.sa_session.get(Job, job_id) msg = {"kind": self.conversion_messages.ERROR, "message": job.stderr} elif not converted_dataset or converted_dataset.state != Dataset.states.OK: msg = self.conversion_messages.PENDING return msg
def _serialize(self, id_encoder, serialization_options): metadata = _prepare_metadata_for_serialization(id_encoder, serialization_options, self.metadata) rval = dict_for( self, create_time=self.create_time.__str__(), update_time=self.update_time.__str__(), name=unicodify(self.name), info=unicodify(self.info), blurb=self.blurb, peek=self.peek, extension=self.extension, metadata=metadata, designation=self.designation, deleted=self.deleted, visible=self.visible, dataset_uuid=(lambda uuid: str(uuid) if uuid else None)(self.dataset.uuid), validated_state=self.validated_state, validated_state_message=self.validated_state_message, ) serialization_options.attach_identifier(id_encoder, self, rval) return rval def _handle_serialize_files(self, id_encoder, serialization_options, rval): if serialization_options.serialize_dataset_objects: rval["dataset"] = self.dataset.serialize(id_encoder, serialization_options) else: serialization_options.serialize_files(self, rval) file_metadata = {} dataset = self.dataset hashes = dataset.hashes if hashes: file_metadata["hashes"] = [h.serialize(id_encoder, serialization_options) for h in hashes] if dataset.created_from_basename is not None: file_metadata["created_from_basename"] = dataset.created_from_basename sources = dataset.sources if sources: file_metadata["sources"] = [s.serialize(id_encoder, serialization_options) for s in sources] rval["file_metadata"] = file_metadata
[docs]class HistoryDatasetAssociation(DatasetInstance, HasTags, Dictifiable, UsesAnnotations, HasName, Serializable): """ Resource class that creates a relation between a dataset and a user history. """ history_id: Optional[int] def __init__( self, hid=None, history=None, sa_session=None, **kwd, ): """ Create a a new HDA and associate it with the given history. """ # FIXME: sa_session is must be passed to DataSetInstance if the create_dataset # parameter is True so that the new object can be flushed. Is there a better way? DatasetInstance.__init__(self, sa_session=sa_session, **kwd) self.hid = hid # Relationships self.history = history def __strict_check_before_flush__(self): if self.extension != "len": # TODO: Custom builds (with .len extension) do not get a history or a HID. # These should get some other type of permanent storage, perhaps UserDatasetAssociation ? # Everything else needs to have a hid and a history if not self.history and not getattr(self, "history_id", None): raise Exception(f"HistoryDatasetAssociation {self} without history detected, this is not valid") elif not self.hid: raise Exception(f"HistoryDatasetAssociation {self} without hid, this is not valid") elif self.dataset.file_size is None and self.dataset.state not in self.dataset.no_data_states: raise Exception( f"HistoryDatasetAssociation {self} in state {self.dataset.state} with null file size, this is not valid" ) @property def user(self): if self.history: return self.history.user def __create_version__(self, session): state = inspect(self) changes = {} for attr in state.mapper.columns: # We only create a new version if columns of the HDA table have changed, and ignore relationships. hist = state.get_history(attr.key, True) if not hist.has_changes(): continue # hist.deleted holds old value(s) changes[attr.key] = hist.deleted if changes and self.update_time and self.state == self.states.OK and not self.deleted: # We only record changes to HDAs that exist in the database and have a update_time new_values = {} new_values["name"] = changes.get("name", self.name) new_values["dbkey"] = changes.get("dbkey", self.dbkey) new_values["extension"] = changes.get("extension", self.extension) new_values["extended_metadata_id"] = changes.get("extended_metadata_id", self.extended_metadata_id) for k, v in new_values.items(): if isinstance(v, list): new_values[k] = v[0] new_values["update_time"] = self.update_time new_values["version"] = self.version or 1 new_values["metadata"] = self._metadata past_hda = HistoryDatasetAssociationHistory(history_dataset_association_id=self.id, **new_values) self.version = self.version + 1 if self.version else 1 session.add(past_hda)
[docs] def copy_from(self, other_hda, new_dataset=None, include_tags=True, include_metadata=False): # This deletes the old dataset, so make sure to only call this on new things # in the history (e.g. during job finishing). old_dataset = self.dataset if include_metadata: self._metadata = other_hda._metadata self.metadata_deferred = other_hda.metadata_deferred self.info = other_hda.info self.blurb = other_hda.blurb self.peek = other_hda.peek self.extension = other_hda.extension self.designation = other_hda.designation self.deleted = other_hda.deleted self.visible = other_hda.visible self.validated_state = other_hda.validated_state self.validated_state_message = other_hda.validated_state_message if include_tags and self.history: self.copy_tags_from(self.user, other_hda) self.dataset = new_dataset or other_hda.dataset self.copied_from_history_dataset_association_id = other_hda.id if old_dataset: old_dataset.full_delete()
[docs] def copy(self, parent_id=None, copy_tags=None, flush=True, copy_hid=True, new_name=None): """ Create a copy of this HDA. """ hid = None if copy_hid: hid = self.hid hda = HistoryDatasetAssociation( hid=hid, name=new_name or self.name, info=self.info, blurb=self.blurb, peek=self.peek, tool_version=self.tool_version, extension=self.extension, dbkey=self.dbkey, dataset=self.dataset, visible=self.visible, deleted=self.deleted, parent_id=parent_id, copied_from_history_dataset_association=self, flush=False, ) # update init non-keywords as well hda.purged = self.purged hda.copy_tags_to(copy_tags) object_session(self).add(hda) hda.metadata = self.metadata if flush: session = object_session(self) with transaction(session): session.commit() return hda
[docs] def copy_tags_to(self, copy_tags=None): if copy_tags is not None: if isinstance(copy_tags, dict): copy_tags = copy_tags.values() for tag in copy_tags: copied_tag = tag.copy(cls=HistoryDatasetAssociationTagAssociation) self.tags.append(copied_tag)
[docs] def copy_attributes(self, new_dataset): if new_dataset.hid is None: new_dataset.hid = self.hid
[docs] def to_library_dataset_dataset_association( self, trans, target_folder, replace_dataset=None, parent_id=None, roles=None, ldda_message="", element_identifier=None, ) -> "LibraryDatasetDatasetAssociation": """ Copy this HDA to a library optionally replacing an existing LDDA. """ if not self.dataset.shareable: raise Exception("Attempting to share a non-shareable dataset.") if replace_dataset: # The replace_dataset param ( when not None ) refers to a LibraryDataset that # is being replaced with a new version. library_dataset = replace_dataset else: # If replace_dataset is None, the Library level permissions will be taken from the folder and # applied to the new LibraryDataset, and the current user's DefaultUserPermissions will be applied # to the associated Dataset. library_dataset = LibraryDataset(folder=target_folder, name=self.name, info=self.info) user = trans.user or self.history.user ldda = LibraryDatasetDatasetAssociation( name=element_identifier or self.name, info=self.info, blurb=self.blurb, peek=self.peek, tool_version=self.tool_version, extension=self.extension, dbkey=self.dbkey, dataset=self.dataset, library_dataset=library_dataset, visible=self.visible, deleted=self.deleted, parent_id=parent_id, copied_from_history_dataset_association=self, user=user, ) library_dataset.library_dataset_dataset_association = ldda session = object_session(self) assert session session.add(library_dataset) # If roles were selected on the upload form, restrict access to the Dataset to those roles roles = roles or [] for role in roles: dp = trans.model.DatasetPermissions( trans.app.security_agent.permitted_actions.DATASET_ACCESS.action, ldda.dataset, role ) trans.sa_session.add(dp) # Must set metadata after ldda flushed, as MetadataFiles require ldda.id if self.set_metadata_requires_flush: with transaction(session): session.commit() ldda.metadata = self.metadata # TODO: copy #tags from history if ldda_message: ldda.message = ldda_message if not replace_dataset: target_folder.add_library_dataset(library_dataset, genome_build=ldda.dbkey) session.add(target_folder) session.add(library_dataset) with transaction(session): session.commit() return ldda
[docs] def clear_associated_files(self, metadata_safe=False, purge=False): """ """ # metadata_safe = True means to only clear when assoc.metadata_safe == False for assoc in self.implicitly_converted_datasets: if not assoc.deleted and (not metadata_safe or not assoc.metadata_safe): assoc.clear(purge=purge) for assoc in self.implicitly_converted_parent_datasets: assoc.clear(purge=purge, delete_dataset=False)
[docs] def get_access_roles(self, security_agent): """ Return The access roles associated with this HDA's dataset. """ return self.dataset.get_access_roles(security_agent)
[docs] def purge_usage_from_quota(self, user, quota_source_info): """Remove this HDA's quota_amount from user's quota.""" if user and quota_source_info.use: user.adjust_total_disk_usage(-self.quota_amount(user), quota_source_info.label)
[docs] def quota_amount(self, user): """ Return the disk space used for this HDA relevant to user quotas. If the user has multiple instances of this dataset, it will not affect their disk usage statistic. """ rval = 0 # Anon users are handled just by their single history size. if not user: return rval # Gets an HDA disk usage, if the user does not already # have an association of the same dataset if not self.dataset.library_associations and not self.purged and not self.dataset.purged: for hda in self.dataset.history_associations: if hda.id == self.id: continue if not hda.purged and hda.history and hda.user and hda.user == user: break else: rval += self.get_total_size() return rval
def _serialize(self, id_encoder, serialization_options): rval = super()._serialize(id_encoder, serialization_options) rval["state"] = self.state rval["hid"] = self.hid rval["annotation"] = unicodify(getattr(self, "annotation", "")) rval["tags"] = self.make_tag_string_list() rval["tool_version"] = self.tool_version if self.history: rval["history_encoded_id"] = serialization_options.get_identifier(id_encoder, self.history) # Handle copied_from_history_dataset_association information... copied_from_history_dataset_association_chain = [] src_hda = self while src_hda.copied_from_history_dataset_association: src_hda = src_hda.copied_from_history_dataset_association copied_from_history_dataset_association_chain.append( serialization_options.get_identifier(id_encoder, src_hda) ) rval["copied_from_history_dataset_association_id_chain"] = copied_from_history_dataset_association_chain self._handle_serialize_files(id_encoder, serialization_options, rval) return rval
[docs] def to_dict(self, view="collection", expose_dataset_path=False): """ Return attributes of this HDA that are exposed using the API. """ # Since this class is a proxy to rather complex attributes we want to # display in other objects, we can't use the simpler method used by # other model classes. original_rval = super().to_dict(view=view) hda = self rval = dict( id=hda.id, hda_ldda="hda", uuid=(lambda uuid: str(uuid) if uuid else None)(hda.dataset.uuid), hid=hda.hid, file_ext=hda.ext, peek=unicodify(hda.display_peek()) if hda.peek and hda.peek != "no peek" else None, model_class=self.__class__.__name__, name=hda.name, deleted=hda.deleted, purged=hda.purged, visible=hda.visible, state=hda.state, history_content_type=hda.history_content_type, file_size=int(hda.get_size()), create_time=hda.create_time.isoformat(), update_time=hda.update_time.isoformat(), data_type=f"{hda.datatype.__class__.__module__}.{hda.datatype.__class__.__name__}", genome_build=hda.dbkey, validated_state=hda.validated_state, validated_state_message=hda.validated_state_message, misc_info=hda.info.strip() if isinstance(hda.info, str) else hda.info, misc_blurb=hda.blurb, ) rval.update(original_rval) if hda.copied_from_library_dataset_dataset_association is not None: rval["copied_from_ldda_id"] = hda.copied_from_library_dataset_dataset_association.id if hda.history is not None: rval["history_id"] = hda.history.id if hda.extended_metadata is not None: rval["extended_metadata"] = hda.extended_metadata.data for name in hda.metadata.spec.keys(): val = hda.metadata.get(name) if isinstance(val, MetadataFile): # only when explicitly set: fetching filepaths can be expensive if not expose_dataset_path: continue val = val.get_file_name() # If no value for metadata, look in datatype for metadata. elif not hda.metadata.element_is_set(name) and hasattr(hda.datatype, name): val = getattr(hda.datatype, name) rval[f"metadata_{name}"] = val return rval
[docs] def unpause_dependent_jobs(self, jobs=None): if self.state == self.states.PAUSED: self.state = self.states.NEW self.info = None jobs_to_unpause = jobs or set() for jtida in self.dependent_jobs: if jtida.job not in jobs_to_unpause: jobs_to_unpause.add(jtida.job) for jtoda in jtida.job.output_datasets: jobs_to_unpause.update(jtoda.dataset.unpause_dependent_jobs(jobs=jobs_to_unpause)) return jobs_to_unpause
@property def history_content_type(self): return "dataset" # TODO: down into DatasetInstance content_type = "dataset" @hybrid.hybrid_property def type_id(self): return "-".join((self.content_type, str(self.id))) @type_id.expression # type: ignore[no-redef] def type_id(cls): return (type_coerce(cls.content_type, Unicode) + "-" + type_coerce(cls.id, Unicode)).label("type_id")
[docs]class HistoryDatasetAssociationHistory(Base): __tablename__ = "history_dataset_association_history" id: Mapped[int] = mapped_column(primary_key=True) history_dataset_association_id: Mapped[Optional[int]] = mapped_column( ForeignKey("history_dataset_association.id"), index=True ) update_time: Mapped[Optional[datetime]] = mapped_column(default=now) version: Mapped[Optional[int]] name: Mapped[Optional[str]] = mapped_column(TrimmedString(255)) extension: Mapped[Optional[str]] = mapped_column(TrimmedString(64)) _metadata = Column("metadata", MetadataType) extended_metadata_id: Mapped[Optional[int]] = mapped_column(ForeignKey("extended_metadata.id"), index=True) def __init__( self, history_dataset_association_id, name, dbkey, update_time, version, extension, extended_metadata_id, metadata, ): self.history_dataset_association_id = history_dataset_association_id self.name = name self.dbkey = dbkey self.update_time = update_time self.version = version self.extension = extension self.extended_metadata_id = extended_metadata_id self._metadata = metadata
# hda read access permission given by a user to a specific site (gen. for external display applications)
[docs]class HistoryDatasetAssociationDisplayAtAuthorization(Base, RepresentById): __tablename__ = "history_dataset_association_display_at_authorization" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(index=True, default=now, onupdate=now, nullable=True) history_dataset_association_id: Mapped[Optional[int]] = mapped_column( ForeignKey("history_dataset_association.id"), index=True ) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) site: Mapped[Optional[str]] = mapped_column(TrimmedString(255)) history_dataset_association: Mapped[Optional["HistoryDatasetAssociation"]] = relationship() user: Mapped[Optional["User"]] = relationship() def __init__(self, hda=None, user=None, site=None): self.history_dataset_association = hda self.user = user self.site = site
[docs]class HistoryDatasetAssociationSubset(Base, RepresentById): __tablename__ = "history_dataset_association_subset" id: Mapped[int] = mapped_column(primary_key=True) history_dataset_association_id: Mapped[Optional[int]] = mapped_column( ForeignKey("history_dataset_association.id"), index=True ) history_dataset_association_subset_id: Mapped[Optional[int]] = mapped_column( ForeignKey("history_dataset_association.id"), index=True ) location: Mapped[Optional[str]] = mapped_column(Unicode(255), index=True) hda: Mapped[Optional["HistoryDatasetAssociation"]] = relationship( primaryjoin=( lambda: HistoryDatasetAssociationSubset.history_dataset_association_id == HistoryDatasetAssociation.id ), ) subset: Mapped[Optional["HistoryDatasetAssociation"]] = relationship( primaryjoin=( lambda: HistoryDatasetAssociationSubset.history_dataset_association_subset_id == HistoryDatasetAssociation.id ), ) def __init__(self, hda, subset, location): self.hda = hda self.subset = subset self.location = location
[docs]class Library(Base, Dictifiable, HasName, Serializable): __tablename__ = "library" id: Mapped[int] = mapped_column(primary_key=True) root_folder_id: Mapped[Optional[int]] = mapped_column(ForeignKey("library_folder.id"), index=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) name: Mapped[Optional[str]] = mapped_column(String(255), index=True) deleted: Mapped[Optional[bool]] = mapped_column(index=True, default=False) purged: Mapped[Optional[bool]] = mapped_column(index=True, default=False) description: Mapped[Optional[str]] = mapped_column(TEXT) synopsis: Mapped[Optional[str]] = mapped_column(TEXT) root_folder = relationship("LibraryFolder", back_populates="library_root") actions: Mapped[List["LibraryPermissions"]] = relationship(back_populates="library", cascade_backrefs=False) permitted_actions = get_permitted_actions(filter="LIBRARY") dict_collection_visible_keys = ["id", "name"] dict_element_visible_keys = ["id", "deleted", "name", "description", "synopsis", "root_folder_id", "create_time"] def __init__(self, name=None, description=None, synopsis=None, root_folder=None): self.name = name or "Unnamed library" self.description = description self.synopsis = synopsis self.root_folder = root_folder def _serialize(self, id_encoder, serialization_options): rval = dict_for( self, name=self.name, description=self.description, synopsis=self.synopsis, ) if self.root_folder: rval["root_folder"] = self.root_folder.serialize(id_encoder, serialization_options) serialization_options.attach_identifier(id_encoder, self, rval) return rval
[docs] def to_dict(self, view="collection", value_mapper=None): """ We prepend an F to folders. """ rval = super().to_dict(view=view, value_mapper=value_mapper) return rval
[docs] def get_active_folders(self, folder, folders=None): # TODO: should we make sure the library is not deleted? def sort_by_attr(seq, attr): """ Sort the sequence of objects by object's attribute Arguments: seq - the list or any sequence (including immutable one) of objects to sort. attr - the name of attribute to sort by """ # Use the "Schwartzian transform" # Create the auxiliary list of tuples where every i-th tuple has form # (seq[i].attr, i, seq[i]) and sort it. The second item of tuple is needed not # only to provide stable sorting, but mainly to eliminate comparison of objects # (which can be expensive or prohibited) in case of equal attribute values. intermed = [(getattr(v, attr), i, v) for i, v in enumerate(seq)] intermed.sort() return [_[-1] for _ in intermed] if folders is None: active_folders = [folder] for active_folder in folder.active_folders: active_folders.extend(self.get_active_folders(active_folder, folders)) return sort_by_attr(active_folders, "id")
[docs] def get_access_roles(self, security_agent): roles = [] for lp in self.actions: if lp.action == security_agent.permitted_actions.LIBRARY_ACCESS.action: roles.append(lp.role) return roles
[docs]class LibraryFolder(Base, Dictifiable, HasName, Serializable): __tablename__ = "library_folder" __table_args__ = (Index("ix_library_folder_name", "name", mysql_length=200),) id: Mapped[int] = mapped_column(primary_key=True) parent_id: Mapped[Optional[int]] = mapped_column(ForeignKey("library_folder.id"), index=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) name: Mapped[Optional[str]] = mapped_column(TEXT) description: Mapped[Optional[str]] = mapped_column(TEXT) order_id: Mapped[Optional[int]] # not currently being used, but for possible future use item_count: Mapped[Optional[int]] deleted: Mapped[Optional[bool]] = mapped_column(index=True, default=False) purged: Mapped[Optional[bool]] = mapped_column(index=True, default=False) genome_build: Mapped[Optional[str]] = mapped_column(TrimmedString(40)) folders: Mapped[List["LibraryFolder"]] = relationship( primaryjoin=(lambda: LibraryFolder.id == LibraryFolder.parent_id), order_by=asc(name), back_populates="parent", ) parent: Mapped[Optional["LibraryFolder"]] = relationship(back_populates="folders", remote_side=[id]) active_folders: Mapped[List["LibraryFolder"]] = relationship( primaryjoin=("and_(LibraryFolder.parent_id == LibraryFolder.id, not_(LibraryFolder.deleted))"), order_by=asc(name), # """sqlalchemy.exc.ArgumentError: Error creating eager relationship 'active_folders' # on parent class '<class 'galaxy.model.LibraryFolder'>' to child class '<class 'galaxy.model.LibraryFolder'>': # Cant use eager loading on a self referential relationship.""" # TODO: This is no longer the case. Fix this: https://docs.sqlalchemy.org/en/14/orm/self_referential.html#configuring-self-referential-eager-loading viewonly=True, ) datasets: Mapped[List["LibraryDataset"]] = relationship( primaryjoin=( lambda: LibraryDataset.folder_id == LibraryFolder.id and LibraryDataset.library_dataset_dataset_association_id.isnot(None) ), order_by=(lambda: asc(LibraryDataset._name)), viewonly=True, ) active_datasets: Mapped[List["LibraryDataset"]] = relationship( primaryjoin=( "and_(LibraryDataset.folder_id == LibraryFolder.id, not_(LibraryDataset.deleted), LibraryDataset.library_dataset_dataset_association_id.isnot(None))" ), order_by=(lambda: asc(LibraryDataset._name)), viewonly=True, ) library_root = relationship("Library", back_populates="root_folder") actions: Mapped[List["LibraryFolderPermissions"]] = relationship(back_populates="folder", cascade_backrefs=False) dict_element_visible_keys = [ "id", "parent_id", "name", "description", "item_count", "genome_build", "update_time", "deleted", ] def __init__(self, name=None, description=None, item_count=0, order_id=None, genome_build=None): self.name = name or "Unnamed folder" self.description = description self.item_count = item_count self.order_id = order_id self.genome_build = genome_build
[docs] def add_library_dataset(self, library_dataset, genome_build=None): library_dataset.folder_id = self.id library_dataset.order_id = self.item_count self.item_count += 1 if genome_build not in [None, "?"]: self.genome_build = genome_build
[docs] def add_folder(self, folder): folder.parent_id = self.id folder.order_id = self.item_count self.item_count += 1
@property def activatable_library_datasets(self): # This needs to be a list return [ ld for ld in self.datasets if ld.library_dataset_dataset_association and not ld.library_dataset_dataset_association.dataset.deleted ] def _serialize(self, id_encoder, serialization_options): rval = dict_for( self, id=self.id, # FIXME: serialize only in sessionless export mode name=self.name, description=self.description, genome_build=self.genome_build, item_count=self.item_count, order_id=self.order_id, # update_time=self.update_time, deleted=self.deleted, ) folders = [] for folder in self.folders: folders.append(folder.serialize(id_encoder, serialization_options)) rval["folders"] = folders datasets = [] for dataset in self.datasets: datasets.append(dataset.serialize(id_encoder, serialization_options)) rval["datasets"] = datasets serialization_options.attach_identifier(id_encoder, self, rval) return rval
[docs] def to_dict(self, view="collection", value_mapper=None): rval = super().to_dict(view=view, value_mapper=value_mapper) rval["library_path"] = self.library_path rval["parent_library_id"] = self.parent_library.id return rval
@property def library_path(self): l_path = [] f = self while f.parent: l_path.insert(0, f.name) f = f.parent return l_path @property def parent_library(self): f = self while f.parent: f = f.parent return f.library_root[0]
[docs]class LibraryDataset(Base, Serializable): __tablename__ = "library_dataset" id: Mapped[int] = mapped_column(primary_key=True) # current version of dataset, if null, there is not a current version selected library_dataset_dataset_association_id: Mapped[Optional[int]] = mapped_column( ForeignKey( "library_dataset_dataset_association.id", use_alter=True, name="library_dataset_dataset_association_id_fk" ), index=True, ) folder_id: Mapped[Optional[int]] = mapped_column(ForeignKey("library_folder.id"), index=True) # not currently being used, but for possible future use order_id: Mapped[Optional[int]] create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) # when not None/null this will supercede display in library (but not when imported into user's history?) _name: Mapped[Optional[str]] = mapped_column("name", TrimmedString(255), index=True) # when not None/null this will supercede display in library (but not when imported into user's history?) _info: Mapped[Optional[str]] = mapped_column("info", TrimmedString(255)) deleted: Mapped[Optional[bool]] = mapped_column(index=True, default=False) purged: Mapped[Optional[bool]] = mapped_column(index=True, default=False) folder: Mapped[Optional["LibraryFolder"]] = relationship() library_dataset_dataset_association = relationship( "LibraryDatasetDatasetAssociation", foreign_keys=library_dataset_dataset_association_id, post_update=True ) expired_datasets: Mapped[List["LibraryDatasetDatasetAssociation"]] = relationship( foreign_keys=[id, library_dataset_dataset_association_id], primaryjoin=( "and_(LibraryDataset.id == LibraryDatasetDatasetAssociation.library_dataset_id, \ not_(LibraryDataset.library_dataset_dataset_association_id == LibraryDatasetDatasetAssociation.id))" ), viewonly=True, uselist=True, ) actions: Mapped[List["LibraryDatasetPermissions"]] = relationship( back_populates="library_dataset", cascade_backrefs=False ) # This class acts as a proxy to the currently selected LDDA upload_options = [ ("upload_file", "Upload files"), ("upload_directory", "Upload directory of files"), ("upload_paths", "Upload files from filesystem paths"), ("import_from_history", "Import datasets from your current history"), ]
[docs] def get_info(self): if self.library_dataset_dataset_association: return self.library_dataset_dataset_association.info elif self._info: return self._info else: return "no info"
[docs] def set_info(self, info): self._info = info
info = property(get_info, set_info)
[docs] def get_name(self): if self.library_dataset_dataset_association: return self.library_dataset_dataset_association.name elif self._name: return self._name else: return "Unnamed dataset"
[docs] def set_name(self, name): self._name = name
name = property(get_name, set_name)
[docs] def display_name(self): self.library_dataset_dataset_association.display_name()
def _serialize(self, id_encoder, serialization_options): rval = dict_for( self, name=self.name, info=self.info, order_id=self.order_id, ldda=self.library_dataset_dataset_association.serialize(id_encoder, serialization_options, for_link=True), ) serialization_options.attach_identifier(id_encoder, self, rval) return rval
[docs] def to_dict(self, view="collection"): # Since this class is a proxy to rather complex attributes we want to # display in other objects, we can't use the simpler method used by # other model classes. ldda = self.library_dataset_dataset_association rval = dict( id=self.id, ldda_id=ldda.id, parent_library_id=self.folder.parent_library.id, folder_id=self.folder_id, model_class=self.__class__.__name__, state=ldda.state, name=ldda.name, file_name=ldda.get_file_name(), created_from_basename=ldda.created_from_basename, uploaded_by=ldda.user and ldda.user.email, message=ldda.message, date_uploaded=ldda.create_time.isoformat(), update_time=ldda.update_time.isoformat(), file_size=int(ldda.get_size()), file_ext=ldda.ext, data_type=f"{ldda.datatype.__class__.__module__}.{ldda.datatype.__class__.__name__}", genome_build=ldda.dbkey, misc_info=ldda.info, misc_blurb=ldda.blurb, peek=(lambda ldda: ldda.display_peek() if ldda.peek and ldda.peek != "no peek" else None)(ldda), ) if ldda.dataset.uuid is None: rval["uuid"] = None else: rval["uuid"] = str(ldda.dataset.uuid) for name in ldda.metadata.spec.keys(): val = ldda.metadata.get(name) if isinstance(val, MetadataFile): val = val.get_file_name() elif isinstance(val, list): val = ", ".join(str(v) for v in val) rval[f"metadata_{name}"] = val return rval
[docs]class LibraryDatasetDatasetAssociation(DatasetInstance, HasName, Serializable): message: Mapped[Optional[str]] tags: Mapped[List["LibraryDatasetDatasetAssociationTagAssociation"]] def __init__( self, library_dataset=None, user=None, sa_session=None, **kwd, ): # FIXME: sa_session is must be passed to DataSetInstance if the create_dataset # parameter in kwd is True so that the new object can be flushed. Is there a better way? DatasetInstance.__init__(self, sa_session=sa_session, **kwd) self.library_dataset = library_dataset self.user = user
[docs] def to_history_dataset_association( self, target_history, parent_id=None, add_to_history=False, visible=None, commit=True ): sa_session = object_session(self) hda = HistoryDatasetAssociation( name=self.name, info=self.info, blurb=self.blurb, peek=self.peek, tool_version=self.tool_version, extension=self.extension, dbkey=self.dbkey, dataset=self.dataset, visible=visible if visible is not None else self.visible, deleted=self.deleted, parent_id=parent_id, copied_from_library_dataset_dataset_association=self, history=target_history, ) tag_manager = galaxy.model.tags.GalaxyTagHandler(sa_session) src_ldda_tags = tag_manager.get_tags_str(self.tags) tag_manager.apply_item_tags(user=self.user, item=hda, tags_str=src_ldda_tags, flush=False) sa_session.add(hda) hda.metadata = self.metadata if add_to_history and target_history: target_history.stage_addition(hda) if commit: target_history.add_pending_items() if commit: with transaction(sa_session): sa_session.commit() return hda
[docs] def copy(self, parent_id=None, target_folder=None, flush=True): sa_session = object_session(self) ldda = LibraryDatasetDatasetAssociation( name=self.name, info=self.info, blurb=self.blurb, peek=self.peek, tool_version=self.tool_version, extension=self.extension, dbkey=self.dbkey, dataset=self.dataset, visible=self.visible, deleted=self.deleted, parent_id=parent_id, copied_from_library_dataset_dataset_association=self, folder=target_folder, ) tag_manager = galaxy.model.tags.GalaxyTagHandler(sa_session) src_ldda_tags = tag_manager.get_tags_str(self.tags) tag_manager.apply_item_tags(user=self.user, item=ldda, tags_str=src_ldda_tags) sa_session.add(ldda) with transaction(sa_session): sa_session.commit() # Need to set after flushed, as MetadataFiles require dataset.id ldda.metadata = self.metadata with transaction(sa_session): sa_session.commit() return ldda
[docs] def clear_associated_files(self, metadata_safe=False, purge=False): return
[docs] def get_access_roles(self, security_agent): return self.dataset.get_access_roles(security_agent)
[docs] def get_manage_permissions_roles(self, security_agent): return self.dataset.get_manage_permissions_roles(security_agent)
[docs] def has_manage_permissions_roles(self, security_agent): return self.dataset.has_manage_permissions_roles(security_agent)
def _serialize(self, id_encoder, serialization_options): rval = super()._serialize(id_encoder, serialization_options) self._handle_serialize_files(id_encoder, serialization_options, rval) return rval
[docs] def to_dict(self, view="collection"): # Since this class is a proxy to rather complex attributes we want to # display in other objects, we can't use the simpler method used by # other model classes. ldda = self try: file_size = int(ldda.get_size()) except OSError: file_size = 0 # TODO: render tags here rval = dict( id=ldda.id, hda_ldda="ldda", model_class=self.__class__.__name__, name=ldda.name, deleted=ldda.deleted, visible=ldda.visible, state=ldda.state, library_dataset_id=ldda.library_dataset_id, file_size=file_size, file_name=ldda.get_file_name(), update_time=ldda.update_time.isoformat(), file_ext=ldda.ext, data_type=f"{ldda.datatype.__class__.__module__}.{ldda.datatype.__class__.__name__}", genome_build=ldda.dbkey, misc_info=ldda.info, misc_blurb=ldda.blurb, created_from_basename=ldda.created_from_basename, ) if ldda.dataset.uuid is None: rval["uuid"] = None else: rval["uuid"] = str(ldda.dataset.uuid) rval["parent_library_id"] = ldda.library_dataset.folder.parent_library.id if ldda.extended_metadata is not None: rval["extended_metadata"] = ldda.extended_metadata.data for name in ldda.metadata.spec.keys(): val = ldda.metadata.get(name) if isinstance(val, MetadataFile): val = val.get_file_name() # If no value for metadata, look in datatype for metadata. elif val is None and hasattr(ldda.datatype, name): val = getattr(ldda.datatype, name) rval[f"metadata_{name}"] = val return rval
[docs] def update_parent_folder_update_times(self): # sets the update_time for all continaing folders up the tree ldda = self sql = text( """ WITH RECURSIVE parent_folders_of(folder_id) AS (SELECT folder_id FROM library_dataset WHERE id = :library_dataset_id UNION ALL SELECT library_folder.parent_id FROM library_folder, parent_folders_of WHERE library_folder.id = parent_folders_of.folder_id ) UPDATE library_folder SET update_time = (SELECT update_time FROM library_dataset_dataset_association WHERE id = :ldda_id) WHERE exists (SELECT 1 FROM parent_folders_of WHERE library_folder.id = parent_folders_of.folder_id) """ ) with object_session(self).bind.connect() as conn, conn.begin(): ret = conn.execute(sql, {"library_dataset_id": ldda.library_dataset_id, "ldda_id": ldda.id}) if ret.rowcount < 1: log.warning(f"Attempt to updated parent folder times failed: {ret.rowcount} records updated.")
[docs]class ExtendedMetadata(Base, RepresentById): __tablename__ = "extended_metadata" id: Mapped[int] = mapped_column(primary_key=True) data: Mapped[Optional[bytes]] = mapped_column(MutableJSONType) children: Mapped[List["ExtendedMetadataIndex"]] = relationship(back_populates="extended_metadata") def __init__(self, data): self.data = data
[docs]class ExtendedMetadataIndex(Base, RepresentById): __tablename__ = "extended_metadata_index" id: Mapped[int] = mapped_column(primary_key=True) extended_metadata_id: Mapped[Optional[int]] = mapped_column( ForeignKey("extended_metadata.id", onupdate="CASCADE", ondelete="CASCADE"), index=True ) path: Mapped[Optional[str]] = mapped_column(String(255)) value: Mapped[Optional[str]] = mapped_column(TEXT) extended_metadata: Mapped[Optional["ExtendedMetadata"]] = relationship(back_populates="children") def __init__(self, extended_metadata, path, value): self.extended_metadata = extended_metadata self.path = path self.value = value
[docs]class LibraryInfoAssociation(Base, RepresentById): __tablename__ = "library_info_association" id: Mapped[int] = mapped_column(primary_key=True) library_id: Mapped[Optional[int]] = mapped_column(ForeignKey("library.id"), index=True) form_definition_id: Mapped[Optional[int]] = mapped_column(ForeignKey("form_definition.id"), index=True) form_values_id: Mapped[Optional[int]] = mapped_column(ForeignKey("form_values.id"), index=True) inheritable: Mapped[Optional[bool]] = mapped_column(index=True, default=False) deleted: Mapped[Optional[bool]] = mapped_column(index=True, default=False) library: Mapped[Optional["Library"]] = relationship( primaryjoin=( lambda: and_( LibraryInfoAssociation.library_id == Library.id, not_(LibraryInfoAssociation.deleted), # type:ignore[arg-type] ) ), ) template: Mapped[Optional["FormDefinition"]] = relationship( primaryjoin=lambda: LibraryInfoAssociation.form_definition_id == FormDefinition.id ) info: Mapped[Optional["FormValues"]] = relationship( primaryjoin=lambda: LibraryInfoAssociation.form_values_id == FormValues.id ) def __init__(self, library, form_definition, info, inheritable=False): self.library = library self.template = form_definition self.info = info self.inheritable = inheritable
[docs]class LibraryFolderInfoAssociation(Base, RepresentById): __tablename__ = "library_folder_info_association" id: Mapped[int] = mapped_column(primary_key=True) library_folder_id: Mapped[Optional[int]] = mapped_column(ForeignKey("library_folder.id"), index=True) form_definition_id: Mapped[Optional[int]] = mapped_column(ForeignKey("form_definition.id"), index=True) form_values_id: Mapped[Optional[int]] = mapped_column(ForeignKey("form_values.id"), index=True) inheritable: Mapped[Optional[bool]] = mapped_column(index=True, default=False) deleted: Mapped[Optional[bool]] = mapped_column(index=True, default=False) folder: Mapped[Optional["LibraryFolder"]] = relationship( primaryjoin=( lambda: (LibraryFolderInfoAssociation.library_folder_id == LibraryFolder.id) & (not_(LibraryFolderInfoAssociation.deleted)) ), ) template: Mapped[Optional["FormDefinition"]] = relationship( primaryjoin=(lambda: LibraryFolderInfoAssociation.form_definition_id == FormDefinition.id) ) info: Mapped[Optional["FormValues"]] = relationship( primaryjoin=(lambda: LibraryFolderInfoAssociation.form_values_id == FormValues.id) ) def __init__(self, folder, form_definition, info, inheritable=False): self.folder = folder self.template = form_definition self.info = info self.inheritable = inheritable
[docs]class LibraryDatasetDatasetInfoAssociation(Base, RepresentById): __tablename__ = "library_dataset_dataset_info_association" id: Mapped[int] = mapped_column(primary_key=True) library_dataset_dataset_association_id: Mapped[Optional[int]] = mapped_column( ForeignKey("library_dataset_dataset_association.id"), index=True ) form_definition_id: Mapped[Optional[int]] = mapped_column(ForeignKey("form_definition.id"), index=True) form_values_id: Mapped[Optional[int]] = mapped_column(ForeignKey("form_values.id"), index=True) deleted: Mapped[Optional[bool]] = mapped_column(index=True, default=False) library_dataset_dataset_association: Mapped[Optional["LibraryDatasetDatasetAssociation"]] = relationship( primaryjoin=( lambda: ( LibraryDatasetDatasetInfoAssociation.library_dataset_dataset_association_id == LibraryDatasetDatasetAssociation.id ) & (not_(LibraryDatasetDatasetInfoAssociation.deleted)) ), ) template: Mapped[Optional["FormDefinition"]] = relationship( primaryjoin=(lambda: LibraryDatasetDatasetInfoAssociation.form_definition_id == FormDefinition.id), ) info: Mapped[Optional["FormValues"]] = relationship( primaryjoin=(lambda: LibraryDatasetDatasetInfoAssociation.form_values_id == FormValues.id) ) def __init__(self, library_dataset_dataset_association, form_definition, info): # TODO: need to figure out if this should be inheritable to the associated LibraryDataset self.library_dataset_dataset_association = library_dataset_dataset_association self.template = form_definition self.info = info @property def inheritable(self): return True # always allow inheriting, used for replacement
[docs]class ImplicitlyConvertedDatasetAssociation(Base, Serializable): __tablename__ = "implicitly_converted_dataset_association" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) hda_id: Mapped[Optional[int]] = mapped_column(ForeignKey("history_dataset_association.id"), index=True) ldda_id: Mapped[Optional[int]] = mapped_column(ForeignKey("library_dataset_dataset_association.id"), index=True) hda_parent_id: Mapped[Optional[int]] = mapped_column(ForeignKey("history_dataset_association.id"), index=True) ldda_parent_id: Mapped[Optional[int]] = mapped_column( ForeignKey("library_dataset_dataset_association.id"), index=True ) deleted: Mapped[Optional[bool]] = mapped_column(index=True, default=False) metadata_safe: Mapped[Optional[bool]] = mapped_column(index=True, default=True) type: Mapped[Optional[str]] = mapped_column(TrimmedString(255)) parent_hda: Mapped[Optional["HistoryDatasetAssociation"]] = relationship( primaryjoin=(lambda: ImplicitlyConvertedDatasetAssociation.hda_parent_id == HistoryDatasetAssociation.id), back_populates="implicitly_converted_datasets", ) dataset_ldda: Mapped[Optional["LibraryDatasetDatasetAssociation"]] = relationship( primaryjoin=(lambda: ImplicitlyConvertedDatasetAssociation.ldda_id == LibraryDatasetDatasetAssociation.id), back_populates="implicitly_converted_parent_datasets", ) dataset: Mapped[Optional["HistoryDatasetAssociation"]] = relationship( primaryjoin=(lambda: ImplicitlyConvertedDatasetAssociation.hda_id == HistoryDatasetAssociation.id), back_populates="implicitly_converted_parent_datasets", ) parent_ldda: Mapped[Optional["LibraryDatasetDatasetAssociation"]] = relationship( primaryjoin=( lambda: ImplicitlyConvertedDatasetAssociation.ldda_parent_id == LibraryDatasetDatasetAssociation.table.c.id ), back_populates="implicitly_converted_datasets", ) def __init__( self, id=None, parent=None, dataset=None, file_type=None, deleted=False, purged=False, metadata_safe=True, for_import=False, ): self.id = id add_object_to_object_session(self, dataset) if isinstance(dataset, HistoryDatasetAssociation): self.dataset = dataset elif isinstance(dataset, LibraryDatasetDatasetAssociation): self.dataset_ldda = dataset elif not for_import: raise AttributeError(f"Unknown dataset type provided for dataset: {type(dataset)}") # else if for import - these connections might not have been included in the store, # recover the data we can? if isinstance(parent, HistoryDatasetAssociation): self.parent_hda = parent elif isinstance(parent, LibraryDatasetDatasetAssociation): self.parent_ldda = parent elif not for_import: raise AttributeError(f"Unknown dataset type provided for parent: {type(parent)}") # else if for import - these connections might not have been included in the store, # recover the data we can? self.type = file_type self.deleted = deleted self.purged = purged self.metadata_safe = metadata_safe
[docs] def clear(self, purge=False, delete_dataset=True): self.deleted = True if self.dataset: if delete_dataset: self.dataset.deleted = True if purge: self.dataset.purged = True if purge and self.dataset.deleted: # do something with purging self.purged = True try: os.unlink(self.get_file_name()) except Exception as e: log.error(f"Failed to purge associated file ({self.get_file_name()}) from disk: {unicodify(e)}")
def _serialize(self, id_encoder, serialization_options): rval = dict_for( self, file_type=self.type, ) if self.parent_hda: rval["parent_hda"] = serialization_options.get_identifier(id_encoder, self.parent_hda) if self.dataset: rval["hda"] = serialization_options.get_identifier(id_encoder, self.dataset) serialization_options.attach_identifier(id_encoder, self, rval) return rval
DEFAULT_COLLECTION_NAME = "Unnamed Collection"
[docs]class InnerCollectionFilter(NamedTuple): column: str operator_function: Callable expected_value: Union[str, int, float, bool]
[docs] def produce_filter(self, table): return self.operator_function(getattr(table, self.column), self.expected_value)
[docs]class DatasetCollection(Base, Dictifiable, UsesAnnotations, Serializable): __tablename__ = "dataset_collection" id: Mapped[int] = mapped_column(primary_key=True) collection_type: Mapped[str] = mapped_column(Unicode(255)) populated_state: Mapped[str] = mapped_column(TrimmedString(64), default="ok") populated_state_message: Mapped[Optional[str]] = mapped_column(TEXT) element_count: Mapped[Optional[int]] create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) elements: Mapped[List["DatasetCollectionElement"]] = relationship( primaryjoin=(lambda: DatasetCollection.id == DatasetCollectionElement.dataset_collection_id), back_populates="collection", order_by=lambda: DatasetCollectionElement.element_index, ) dict_collection_visible_keys = ["id", "collection_type"] dict_element_visible_keys = ["id", "collection_type"] populated_states = DatasetCollectionPopulatedState def __init__(self, id=None, collection_type=None, populated=True, element_count=None): self.id = id self.collection_type = collection_type if not populated: self.populated_state = DatasetCollection.populated_states.NEW self.element_count = element_count def _build_nested_collection_attributes_stmt( self, collection_attributes: Optional[Iterable[str]] = None, element_attributes: Optional[Iterable[str]] = None, hda_attributes: Optional[Iterable[str]] = None, dataset_attributes: Optional[Iterable[str]] = None, dataset_permission_attributes: Optional[Iterable[str]] = None, return_entities: Optional[ Iterable[ Union[ Type[HistoryDatasetAssociation], Type[Dataset], Type[DatasetPermissions], Type["DatasetCollection"], Type["DatasetCollectionElement"], ] ] ] = None, inner_filter: Optional[InnerCollectionFilter] = None, ): collection_attributes = collection_attributes or () element_attributes = element_attributes or () hda_attributes = hda_attributes or () dataset_attributes = dataset_attributes or () dataset_permission_attributes = dataset_permission_attributes or () return_entities = return_entities or () dataset_collection = self dc = alias(DatasetCollection.__table__) dce = alias(DatasetCollectionElement.__table__) depth_collection_type = dataset_collection.collection_type order_by_columns = [dce.c.element_index] nesting_level = 0 def attribute_columns(column_collection, attributes, nesting_level=None): label_fragment = f"_{nesting_level}" if nesting_level is not None else "" return [getattr(column_collection, a).label(f"{a}{label_fragment}") for a in attributes] q = ( select( *attribute_columns(dce.c, element_attributes, nesting_level), *attribute_columns(dc.c, collection_attributes, nesting_level), ) .select_from(dce, dc) .join(dce, dce.c.dataset_collection_id == dc.c.id) .filter(dc.c.id == dataset_collection.id) ) while ":" in depth_collection_type: nesting_level += 1 inner_dce = alias(DatasetCollectionElement.__table__) inner_dc = alias(DatasetCollection.__table__) order_by_columns.append(inner_dce.c.element_index) q = q.outerjoin(inner_dce, inner_dce.c.dataset_collection_id == dce.c.child_collection_id) if collection_attributes: q = q.join(inner_dc, inner_dc.c.id == dce.c.child_collection_id) q = q.add_columns( *attribute_columns(inner_dc.c, collection_attributes, nesting_level), ) q = q.add_columns(*attribute_columns(inner_dce.c, element_attributes, nesting_level)) dce = inner_dce dc = inner_dc depth_collection_type = depth_collection_type.split(":", 1)[1] if inner_filter: q = q.filter(inner_filter.produce_filter(dc.c)) if ( hda_attributes or dataset_attributes or dataset_permission_attributes or return_entities and not return_entities == (DatasetCollectionElement,) ): q = q.join(HistoryDatasetAssociation).join(Dataset) if dataset_permission_attributes: q = q.join(DatasetPermissions) q = ( q.add_columns(*attribute_columns(HistoryDatasetAssociation, hda_attributes)) .add_columns(*attribute_columns(Dataset, dataset_attributes)) .add_columns(*attribute_columns(DatasetPermissions, dataset_permission_attributes)) ) for entity in return_entities: q = q.add_columns(entity) if entity == DatasetCollectionElement: q = q.filter(entity.id == dce.c.id) q = q.order_by(*order_by_columns) return q @property def elements_deleted(self): if not hasattr(self, "_elements_deleted"): if session := object_session(self): stmt = self._build_nested_collection_attributes_stmt( hda_attributes=("deleted",), dataset_attributes=("deleted",) ) stmt = stmt.exists().where(or_(HistoryDatasetAssociation.deleted == true(), Dataset.deleted == true())) self._elements_deleted = session.execute(select(stmt)).scalar() else: self._elements_deleted = False for dataset_instance in self.dataset_instances: if dataset_instance.deleted or dataset_instance.dataset.deleted: self._elements_deleted = True break return self._elements_deleted @property def dataset_states_and_extensions_summary(self): if not hasattr(self, "_dataset_states_and_extensions_summary"): stmt = self._build_nested_collection_attributes_stmt( hda_attributes=("extension",), dataset_attributes=("state",) ) # With DISTINCT, all columns that appear in the ORDER BY clause must appear in the SELECT clause. stmt = stmt.add_columns(*stmt._order_by_clauses) stmt = stmt.distinct() tuples = object_session(self).execute(stmt) extensions = set() states = set() for extension, state, *_ in tuples: # we discard the added columns from the order-by clause states.add(state) extensions.add(extension) self._dataset_states_and_extensions_summary = (states, extensions) return self._dataset_states_and_extensions_summary @property def has_deferred_data(self): if not hasattr(self, "_has_deferred_data"): has_deferred_data = False if object_session(self): # TODO: Optimize by just querying without returning the states... stmt = self._build_nested_collection_attributes_stmt(dataset_attributes=("state",)) tuples = object_session(self).execute(stmt) for (state,) in tuples: if state == Dataset.states.DEFERRED: has_deferred_data = True break else: # This will be in a remote tool evaluation context, so can't query database for dataset_element in self.dataset_elements_and_identifiers(): if dataset_element.hda.state == Dataset.states.DEFERRED: has_deferred_data = True break self._has_deferred_data = has_deferred_data return self._has_deferred_data @property def populated_optimized(self): if not hasattr(self, "_populated_optimized"): if not self.id: return self.populated _populated_optimized = True if ":" not in self.collection_type: _populated_optimized = self.populated_state == DatasetCollection.populated_states.OK else: stmt = self._build_nested_collection_attributes_stmt( collection_attributes=("populated_state",), inner_filter=InnerCollectionFilter( "populated_state", operator.__ne__, DatasetCollection.populated_states.OK ), ) stmt = stmt.subquery() stmt = select(~exists(stmt)) session = object_session(self) _populated_optimized = session.scalar(stmt) self._populated_optimized = _populated_optimized return self._populated_optimized @property def populated(self): top_level_populated = self.populated_state == DatasetCollection.populated_states.OK if top_level_populated and self.has_subcollections: return all(e.child_collection and e.child_collection.populated for e in self.elements) return top_level_populated @property def dataset_action_tuples(self): if not hasattr(self, "_dataset_action_tuples"): stmt = self._build_nested_collection_attributes_stmt(dataset_permission_attributes=("action", "role_id")) tuples = object_session(self).execute(stmt) self._dataset_action_tuples = [(action, role_id) for action, role_id in tuples if action is not None] return self._dataset_action_tuples @property def element_identifiers_extensions_paths_and_metadata_files( self, ) -> List[List[Any]]: results = [] if object_session(self): stmt = self._build_nested_collection_attributes_stmt( element_attributes=("element_identifier",), hda_attributes=("extension",), return_entities=(HistoryDatasetAssociation, Dataset), ) tuples = object_session(self).execute(stmt) # type:ignore[union-attr] # element_identifiers, extension, path for row in tuples: result = [row[:-3], row.extension, row.Dataset.get_file_name()] hda = row.HistoryDatasetAssociation result.append(hda.get_metadata_file_paths_and_extensions()) results.append(result) else: # This will be in a remote tool evaluation context, so can't query database for dataset_element in self.dataset_elements_and_identifiers(): # Let's pretend name is element identifier results.append( [ dataset_element._identifiers, dataset_element.hda.extension, dataset_element.hda.get_file_name(), dataset_element.hda.get_metadata_file_paths_and_extensions(), ] ) return results @property def waiting_for_elements(self): top_level_waiting = self.populated_state == DatasetCollection.populated_states.NEW if not top_level_waiting and self.has_subcollections: return any(e.child_collection.waiting_for_elements for e in self.elements) return top_level_waiting
[docs] def mark_as_populated(self): self.populated_state = DatasetCollection.populated_states.OK
[docs] def handle_population_failed(self, message): self.populated_state = DatasetCollection.populated_states.FAILED self.populated_state_message = message
[docs] def finalize(self, collection_type_description): # All jobs have written out their elements - everything should be populated # but might not be - check that second case! (TODO) self.mark_as_populated() if self.has_subcollections and collection_type_description.has_subcollections(): for element in self.elements: element.child_collection.finalize(collection_type_description.child_collection_type_description())
@property def dataset_instances(self): db_session = object_session(self) if db_session and self.id: stmt = self._build_nested_collection_attributes_stmt(return_entities=(HistoryDatasetAssociation,)) tuples = db_session.execute(stmt).all() return [tuple[0] for tuple in tuples] else: # Sessionless context instances = [] for element in self.elements: if element.is_collection: instances.extend(element.child_collection.dataset_instances) else: instance = element.dataset_instance instances.append(instance) return instances @property def dataset_elements(self): db_session = object_session(self) if db_session and self.id: stmt = self._build_nested_collection_attributes_stmt(return_entities=(DatasetCollectionElement,)) tuples = db_session.execute(stmt).all() return [tuple[0] for tuple in tuples] elements = [] for element in self.elements: if element.is_collection: elements.extend(element.child_collection.dataset_elements) else: elements.append(element) return elements
[docs] def dataset_elements_and_identifiers(self, identifiers=None): # Used only in remote tool evaluation context elements = [] if identifiers is None: identifiers = [] for element in self.elements: _identifiers = identifiers[:] _identifiers.append(element.element_identifier) if element.is_collection: elements.extend(element.child_collection.dataset_elements_and_identifiers(_identifiers)) else: element._identifiers = _identifiers elements.append(element) return elements
@property def first_dataset_element(self) -> Optional["DatasetCollectionElement"]: for element in self.elements: if element.is_collection: first_element = element.child_collection.first_dataset_element if first_element: return first_element else: return element return None @property def state(self): # TODO: DatasetCollection state handling... return "ok"
[docs] def validate(self): if self.collection_type is None: raise Exception("Each dataset collection must define a collection type.")
def __getitem__(self, key): if isinstance(key, int): try: return self.elements[key] except IndexError: pass else: # This might be a peformance issue for large collection, but we don't use this a lot for element in self.elements: if element.element_identifier == key: return element get_by_attribute = "element_index" if isinstance(key, int) else "element_identifier" error_message = f"Dataset collection has no {get_by_attribute} with key {key}." raise KeyError(error_message)
[docs] def copy( self, destination=None, element_destination=None, dataset_instance_attributes=None, flush=True, minimize_copies=False, ): new_collection = DatasetCollection(collection_type=self.collection_type, element_count=self.element_count) for element in self.elements: element.copy_to_collection( new_collection, destination=destination, element_destination=element_destination, dataset_instance_attributes=dataset_instance_attributes, flush=flush, minimize_copies=minimize_copies, ) object_session(self).add(new_collection) if flush: session = object_session(self) with transaction(session): session.commit() return new_collection
[docs] def replace_failed_elements(self, replacements): stmt = self._build_nested_collection_attributes_stmt( return_entities=[DatasetCollectionElement], hda_attributes=["id"] ) tuples = object_session(self).execute(stmt).all() hda_id_to_element = dict(tuples) for failed, replacement in replacements.items(): element = hda_id_to_element.get(failed.id) if element: element.hda = replacement
[docs] def set_from_dict(self, new_data): # Nothing currently editable in this class. return {}
@property def has_subcollections(self): return ":" in self.collection_type def _serialize(self, id_encoder, serialization_options): rval = dict_for( self, type=self.collection_type, populated_state=self.populated_state, populated_state_message=self.populated_state_message, elements=[e.serialize(id_encoder, serialization_options) for e in self.elements], ) serialization_options.attach_identifier(id_encoder, self, rval) return rval
[docs]class DatasetCollectionInstance(HasName, UsesCreateAndUpdateTime): @property def state(self): return self.collection.state @property def populated(self): return self.collection.populated @property def dataset_instances(self): return self.collection.dataset_instances
[docs] def display_name(self): return self.get_display_name()
def _base_to_dict(self, view): return dict( id=self.id, name=self.name, collection_id=self.collection_id, collection_type=self.collection.collection_type, populated=self.populated, populated_state=self.collection.populated_state, populated_state_message=self.collection.populated_state_message, element_count=self.collection.element_count, elements_datatypes=list(self.dataset_dbkeys_and_extensions_summary[1]), type="collection", # contents type (distinguished from file or folder (in case of library)) )
[docs] def set_from_dict(self, new_data): """ Set object attributes to the values in dictionary new_data limiting to only those keys in dict_element_visible_keys. Returns a dictionary of the keys, values that have been changed. """ # precondition: keys are proper, values are parsed and validated changed = self.collection.set_from_dict(new_data) # unknown keys are ignored here for key in (k for k in new_data.keys() if k in self.editable_keys): new_val = new_data[key] old_val = self.__getattribute__(key) if new_val == old_val: continue self.__setattr__(key, new_val) changed[key] = new_val return changed
@property def has_deferred_data(self): return self.collection.has_deferred_data
[docs]class HistoryDatasetCollectionAssociation( Base, DatasetCollectionInstance, HasTags, Dictifiable, UsesAnnotations, Serializable, ): """Associates a DatasetCollection with a History.""" __tablename__ = "history_dataset_collection_association" id: Mapped[int] = mapped_column(primary_key=True) collection_id: Mapped[Optional[int]] = mapped_column(ForeignKey("dataset_collection.id"), index=True) history_id: Mapped[Optional[int]] = mapped_column(ForeignKey("history.id"), index=True) name: Mapped[Optional[str]] = mapped_column(TrimmedString(255)) hid: Mapped[Optional[int]] visible: Mapped[Optional[bool]] deleted: Mapped[Optional[bool]] = mapped_column(default=False) copied_from_history_dataset_collection_association_id: Mapped[Optional[int]] = mapped_column( ForeignKey("history_dataset_collection_association.id") ) implicit_output_name: Mapped[Optional[str]] = mapped_column(Unicode(255)) job_id: Mapped[Optional[int]] = mapped_column(ForeignKey("job.id"), index=True) implicit_collection_jobs_id: Mapped[Optional[int]] = mapped_column( ForeignKey("implicit_collection_jobs.id"), index=True ) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, index=True, nullable=True) collection = relationship("DatasetCollection") history: Mapped[Optional["History"]] = relationship(back_populates="dataset_collections") copied_from_history_dataset_collection_association = relationship( "HistoryDatasetCollectionAssociation", primaryjoin=copied_from_history_dataset_collection_association_id == id, remote_side=[id], uselist=False, back_populates="copied_to_history_dataset_collection_association", ) copied_to_history_dataset_collection_association = relationship( "HistoryDatasetCollectionAssociation", back_populates="copied_from_history_dataset_collection_association", ) implicit_input_collections: Mapped[List["ImplicitlyCreatedDatasetCollectionInput"]] = relationship( primaryjoin=( lambda: HistoryDatasetCollectionAssociation.id == ImplicitlyCreatedDatasetCollectionInput.dataset_collection_id ), ) implicit_collection_jobs = relationship("ImplicitCollectionJobs", uselist=False) job: Mapped[Optional["Job"]] = relationship( back_populates="history_dataset_collection_associations", uselist=False, ) tags: Mapped[List["HistoryDatasetCollectionTagAssociation"]] = relationship( order_by=lambda: HistoryDatasetCollectionTagAssociation.id, back_populates="dataset_collection", ) annotations: Mapped[List["HistoryDatasetCollectionAssociationAnnotationAssociation"]] = relationship( order_by=lambda: HistoryDatasetCollectionAssociationAnnotationAssociation.id, back_populates="history_dataset_collection", ) ratings: Mapped[List["HistoryDatasetCollectionRatingAssociation"]] = relationship( order_by=lambda: HistoryDatasetCollectionRatingAssociation.id, back_populates="dataset_collection", ) creating_job_associations: Mapped[List["JobToOutputDatasetCollectionAssociation"]] = relationship(viewonly=True) dict_dbkeysandextensions_visible_keys = ["dbkeys", "extensions"] editable_keys = ("name", "deleted", "visible") _job_state_summary = None def __init__(self, deleted=False, visible=True, **kwd): super().__init__(**kwd) # Since deleted property is shared between history and dataset collections, # it could be on either table - some places in the code however it is convient # it is on instance instead of collection. self.deleted = deleted self.visible = visible self.implicit_input_collections = self.implicit_input_collections or [] @property def history_content_type(self): return "dataset_collection" # TODO: down into DatasetCollectionInstance content_type = "dataset_collection" @hybrid.hybrid_property def type_id(self): return "-".join((self.content_type, str(self.id))) @type_id.expression # type: ignore[no-redef] def type_id(cls): return (type_coerce(cls.content_type, Unicode) + "-" + type_coerce(cls.id, Unicode)).label("type_id") @property def job_source_type(self): if self.implicit_collection_jobs_id: return "ImplicitCollectionJobs" elif self.job_id: return "Job" else: return None @property def job_state_summary(self): """ Aggregate counts of jobs by state, stored in a JobStateSummary object. """ if not self._job_state_summary: self._job_state_summary = self._get_job_state_summary() # if summary exists, but there are no jobs, load zeroes for all other counts (otherwise they will be None) if self._job_state_summary and self._job_state_summary.all_jobs == 0: zeroes = [0] * (len(Job.states) + 1) self._job_state_summary = JobStateSummary._make(zeroes) return self._job_state_summary def _get_job_state_summary(self): def build_statement(): state_label = "state" # used to generate `SELECT job.state AS state`, and then refer to it in aggregates. # Select job states joining on icjja > icj > hdca # (We are selecting Job.id in addition to Job.state because otherwise the UNION operation # will get rid of duplicates, making aggregates meaningless.) subq1 = ( select(Job.id, Job.state.label(state_label)) .join(ImplicitCollectionJobsJobAssociation, ImplicitCollectionJobsJobAssociation.job_id == Job.id) .join( ImplicitCollectionJobs, ImplicitCollectionJobs.id == ImplicitCollectionJobsJobAssociation.implicit_collection_jobs_id, ) .join( HistoryDatasetCollectionAssociation, HistoryDatasetCollectionAssociation.implicit_collection_jobs_id == ImplicitCollectionJobs.id, ) .where(HistoryDatasetCollectionAssociation.id == self.id) ) # Select job states joining on hdca subq2 = ( select(Job.id, Job.state.label(state_label)) .join(HistoryDatasetCollectionAssociation, HistoryDatasetCollectionAssociation.job_id == Job.id) .where(HistoryDatasetCollectionAssociation.id == self.id) ) # Combine subqueries subq = subq1.union(subq2) # Build and return final query stm = select().select_from(subq) # Add aggregate columns for each job state for state in enum_values(Job.states): col = func.sum(case((column(state_label) == state, 1), else_=0)).label(state) stm = stm.add_columns(col) # Add aggregate column for all jobs col = func.count().label("all_jobs") stm = stm.add_columns(col) return stm if not object_session(self): return None # no session means object is not persistant; therefore, it has no associated jobs. engine = object_session(self).bind with engine.connect() as conn: counts = conn.execute(build_statement()).one() assert len(counts) == len(Job.states) + 1 # Verify all job states + all jobs are counted return JobStateSummary._make(counts) @property def job_state_summary_dict(self): if self.job_state_summary: return self.job_state_summary._asdict() @property def dataset_dbkeys_and_extensions_summary(self): if not hasattr(self, "_dataset_dbkeys_and_extensions_summary"): stmt = self.collection._build_nested_collection_attributes_stmt(hda_attributes=("_metadata", "extension")) tuples = object_session(self).execute(stmt) extensions = set() dbkeys = set() for row in tuples: if row is not None: dbkey_field = row._metadata.get("dbkey") if isinstance(dbkey_field, list): for dbkey in dbkey_field: dbkeys.add(dbkey) else: dbkeys.add(dbkey_field) if row.extension: extensions.add(row.extension) self._dataset_dbkeys_and_extensions_summary = (dbkeys, extensions) return self._dataset_dbkeys_and_extensions_summary @property def job_source_id(self): return self.implicit_collection_jobs_id or self.job_id
[docs] def touch(self): # cause an update to be emitted, so that e.g. update_time is incremented and triggers are notified if getattr(self, "name", None): # attribute to flag doesn't really matter as long as it's not null (and primary key also doesn't work) flag_modified(self, "name") if self.collection: flag_modified(self.collection, "collection_type")
@overload def to_hda_representative(self, multiple: Literal[False] = False) -> Optional[HistoryDatasetAssociation]: ... @overload def to_hda_representative(self, multiple: Literal[True]) -> List[HistoryDatasetAssociation]: ...
[docs] def to_hda_representative( self, multiple: bool = False ) -> Union[List[HistoryDatasetAssociation], Optional[HistoryDatasetAssociation]]: rval = [] for dataset in self.collection.dataset_elements: rval.append(dataset.dataset_instance) if multiple is False: break if multiple: return rval return rval[0] if rval else None
def _serialize(self, id_encoder, serialization_options): rval = dict_for( self, display_name=self.display_name(), state=self.state, hid=self.hid, collection=self.collection.serialize(id_encoder, serialization_options), implicit_output_name=self.implicit_output_name, ) if self.history: rval["history_encoded_id"] = serialization_options.get_identifier(id_encoder, self.history) implicit_input_collections = [] for implicit_input_collection in self.implicit_input_collections: input_hdca = implicit_input_collection.input_dataset_collection implicit_input_collections.append( { "name": implicit_input_collection.name, "input_dataset_collection": serialization_options.get_identifier(id_encoder, input_hdca), } ) if implicit_input_collections: rval["implicit_input_collections"] = implicit_input_collections # Handle copied_from_history_dataset_association information... copied_from_history_dataset_collection_association_chain = [] src_hdca = self while src_hdca.copied_from_history_dataset_collection_association: src_hdca = src_hdca.copied_from_history_dataset_collection_association copied_from_history_dataset_collection_association_chain.append( serialization_options.get_identifier(id_encoder, src_hdca) ) rval["copied_from_history_dataset_collection_association_id_chain"] = ( copied_from_history_dataset_collection_association_chain ) serialization_options.attach_identifier(id_encoder, self, rval) return rval
[docs] def to_dict(self, view="collection"): original_dict_value = super().to_dict(view=view) if view == "dbkeysandextensions": (dbkeys, extensions) = self.dataset_dbkeys_and_extensions_summary dict_value = dict( dbkey=dbkeys.pop() if len(dbkeys) == 1 else "?", extension=extensions.pop() if len(extensions) == 1 else "auto", ) else: dict_value = dict( hid=self.hid, history_id=self.history.id, history_content_type=self.history_content_type, visible=self.visible, deleted=self.deleted, job_source_id=self.job_source_id, job_source_type=self.job_source_type, job_state_summary=self.job_state_summary_dict, create_time=self.create_time.isoformat(), update_time=self.update_time.isoformat(), **self._base_to_dict(view=view), ) dict_value.update(original_dict_value) return dict_value
[docs] def add_implicit_input_collection(self, name, history_dataset_collection): self.implicit_input_collections.append( ImplicitlyCreatedDatasetCollectionInput(name, history_dataset_collection) )
[docs] def find_implicit_input_collection(self, name): matching_collection = None for implicit_input_collection in self.implicit_input_collections: if implicit_input_collection.name == name: matching_collection = implicit_input_collection.input_dataset_collection break return matching_collection
[docs] def copy( self, element_destination=None, dataset_instance_attributes=None, flush=True, set_hid=True, minimize_copies=False, ): """ Create a copy of this history dataset collection association. Copy underlying collection. """ hdca = HistoryDatasetCollectionAssociation( hid=self.hid, collection=None, visible=self.visible, deleted=self.deleted, name=self.name, copied_from_history_dataset_collection_association=self, ) if self.implicit_collection_jobs_id: hdca.implicit_collection_jobs_id = self.implicit_collection_jobs_id elif self.job_id: hdca.job_id = self.job_id collection_copy = self.collection.copy( destination=hdca, element_destination=element_destination, dataset_instance_attributes=dataset_instance_attributes, flush=False, minimize_copies=minimize_copies, ) hdca.collection = collection_copy object_session(self).add(hdca) hdca.copy_tags_from(self.history.user, self) if element_destination and set_hid: element_destination.stage_addition(hdca) element_destination.add_pending_items() if flush: session = object_session(self) with transaction(session): session.commit() return hdca
@property def waiting_for_elements(self): summary = self.job_state_summary if summary.all_jobs > 0 and summary.deleted + summary.error + summary.failed + summary.ok == summary.all_jobs: return False else: return self.collection.waiting_for_elements
[docs] def contains_collection(self, collection_id): """Checks to see that the indicated collection is a member of the hdca by using a recursive CTE sql query to find the collection's parents and checking to see if any of the parents are associated with this hdca""" if collection_id == self.collection_id: # collection_id is root collection return True sa_session = object_session(self) DCE = DatasetCollectionElement HDCA = HistoryDatasetCollectionAssociation # non-recursive part of the cte (starting point) parents_cte = ( Query(DCE.dataset_collection_id) .filter(or_(DCE.child_collection_id == collection_id, DCE.dataset_collection_id == collection_id)) .cte(name="element_parents", recursive="True") ) ep = aliased(parents_cte, name="ep") # add the recursive part of the cte expression dce = aliased(DCE, name="dce") rec = Query(dce.dataset_collection_id.label("dataset_collection_id")).filter( dce.child_collection_id == ep.c.dataset_collection_id ) parents_cte = parents_cte.union(rec) # join parents to hdca, look for matching hdca_id hdca = aliased(HDCA, name="hdca") jointohdca = parents_cte.join(hdca, hdca.collection_id == parents_cte.c.dataset_collection_id) qry = Query(hdca.id).select_from(jointohdca).filter(hdca.id == self.id) results = qry.with_session(sa_session).all() return len(results) > 0
HistoryItem: TypeAlias = Union[HistoryDatasetAssociation, HistoryDatasetCollectionAssociation]
[docs]class LibraryDatasetCollectionAssociation(Base, DatasetCollectionInstance, RepresentById): """Associates a DatasetCollection with a library folder.""" __tablename__ = "library_dataset_collection_association" id: Mapped[int] = mapped_column(primary_key=True) collection_id: Mapped[Optional[int]] = mapped_column(ForeignKey("dataset_collection.id"), index=True) folder_id: Mapped[Optional[int]] = mapped_column(ForeignKey("library_folder.id"), index=True) name: Mapped[Optional[str]] = mapped_column(TrimmedString(255)) deleted: Mapped[Optional[bool]] = mapped_column(default=False) collection = relationship("DatasetCollection") folder = relationship("LibraryFolder") tags: Mapped[List["LibraryDatasetCollectionTagAssociation"]] = relationship( order_by=lambda: LibraryDatasetCollectionTagAssociation.id, back_populates="dataset_collection", ) annotations: Mapped[List["LibraryDatasetCollectionAnnotationAssociation"]] = relationship( order_by=lambda: LibraryDatasetCollectionAnnotationAssociation.id, back_populates="dataset_collection", ) ratings: Mapped[List["LibraryDatasetCollectionRatingAssociation"]] = relationship( order_by=lambda: LibraryDatasetCollectionRatingAssociation.id, back_populates="dataset_collection", ) editable_keys = ("name", "deleted") def __init__(self, deleted=False, **kwd): super().__init__(**kwd) # Since deleted property is shared between history and dataset collections, # it could be on either table - some places in the code however it is convient # it is on instance instead of collection. self.deleted = deleted
[docs] def to_dict(self, view="collection"): dict_value = dict(folder_id=self.folder.id, **self._base_to_dict(view=view)) return dict_value
[docs]class DatasetCollectionElement(Base, Dictifiable, Serializable): """Associates a DatasetInstance (hda or ldda) with a DatasetCollection.""" __tablename__ = "dataset_collection_element" id: Mapped[int] = mapped_column(primary_key=True) # Parent collection id describing what collection this element belongs to. dataset_collection_id: Mapped[int] = mapped_column(ForeignKey("dataset_collection.id"), index=True) # Child defined by this association - HDA, LDDA, or another dataset association... hda_id: Mapped[Optional[int]] = mapped_column(ForeignKey("history_dataset_association.id"), index=True) ldda_id: Mapped[Optional[int]] = mapped_column(ForeignKey("library_dataset_dataset_association.id"), index=True) child_collection_id: Mapped[Optional[int]] = mapped_column(ForeignKey("dataset_collection.id"), index=True) # Element index and identifier to define this parent-child relationship. element_index: Mapped[Optional[int]] element_identifier: Mapped[Optional[str]] = mapped_column(Unicode(255)) hda = relationship( "HistoryDatasetAssociation", primaryjoin=(lambda: DatasetCollectionElement.hda_id == HistoryDatasetAssociation.id), ) ldda = relationship( "LibraryDatasetDatasetAssociation", primaryjoin=(lambda: DatasetCollectionElement.ldda_id == LibraryDatasetDatasetAssociation.id), ) child_collection = relationship( "DatasetCollection", primaryjoin=(lambda: DatasetCollectionElement.child_collection_id == DatasetCollection.id) ) collection = relationship( "DatasetCollection", primaryjoin=(lambda: DatasetCollection.id == DatasetCollectionElement.dataset_collection_id), back_populates="elements", ) # actionable dataset id needs to be available via API... dict_collection_visible_keys = ["id", "element_type", "element_index", "element_identifier"] dict_element_visible_keys = ["id", "element_type", "element_index", "element_identifier"] UNINITIALIZED_ELEMENT = object() def __init__( self, id=None, collection=None, element=None, element_index=None, element_identifier=None, ): if isinstance(element, HistoryDatasetAssociation): self.hda = element elif isinstance(element, LibraryDatasetDatasetAssociation): self.ldda = element elif isinstance(element, DatasetCollection): self.child_collection = element elif element != self.UNINITIALIZED_ELEMENT: raise AttributeError(f"Unknown element type provided: {type(element)}") self.id = id add_object_to_object_session(self, collection) self.collection = collection self.element_index = element_index self.element_identifier = element_identifier or str(element_index) def __strict_check_before_flush__(self): if self.collection.populated_optimized: assert self.element_object, "Dataset Collection Element without child entity detected, this is not valid" @property def element_type(self): if self.hda: return "hda" elif self.ldda: return "ldda" elif self.child_collection: # TOOD: Rename element_type to element_type. return "dataset_collection" else: return None @property def is_collection(self): return self.element_type == "dataset_collection" @property def element_object( self, ) -> Optional[Union[HistoryDatasetAssociation, LibraryDatasetDatasetAssociation, DatasetCollection]]: if self.hda: return self.hda elif self.ldda: return self.ldda elif self.child_collection: return self.child_collection else: return None @property def dataset_instance(self): element_object = self.element_object if isinstance(element_object, DatasetCollection): raise AttributeError("Nested collection has no associated dataset_instance.") return element_object @property def dataset(self): return self.dataset_instance.dataset
[docs] def first_dataset_instance(self): element_object = self.element_object if isinstance(element_object, DatasetCollection): return element_object.dataset_instances[0] else: return element_object
@property def dataset_instances(self): element_object = self.element_object if isinstance(element_object, DatasetCollection): return element_object.dataset_instances else: return [element_object] @property def has_deferred_data(self): return self.element_object.has_deferred_data
[docs] def copy_to_collection( self, collection, destination=None, element_destination=None, dataset_instance_attributes=None, flush=True, minimize_copies=False, ): dataset_instance_attributes = dataset_instance_attributes or {} element_object = self.element_object if element_destination: if self.is_collection: element_object = element_object.copy( destination=destination, element_destination=element_destination, dataset_instance_attributes=dataset_instance_attributes, flush=flush, minimize_copies=minimize_copies, ) else: new_element_object = None if minimize_copies: new_element_object = element_destination.get_dataset_by_hid(element_object.hid) if ( new_element_object and new_element_object.dataset and new_element_object.dataset.id == element_object.dataset_id ): element_object = new_element_object else: new_element_object = element_object.copy(flush=flush, copy_tags=element_object.tags) for attribute, value in dataset_instance_attributes.items(): setattr(new_element_object, attribute, value) new_element_object.visible = False if destination is not None and element_object.hidden_beneath_collection_instance: new_element_object.hidden_beneath_collection_instance = destination # Ideally we would not need to give the following # element an HID and it would exist in the history only # as an element of the containing collection. element_destination.stage_addition(new_element_object) element_object = new_element_object new_element = DatasetCollectionElement( element=element_object, collection=collection, element_index=self.element_index, element_identifier=self.element_identifier, ) return new_element
def _serialize(self, id_encoder, serialization_options): rval = dict_for( self, element_type=self.element_type, element_index=self.element_index, element_identifier=self.element_identifier, ) serialization_options.attach_identifier(id_encoder, self, rval) element_obj = self.element_object if isinstance(element_obj, HistoryDatasetAssociation): rval["hda"] = element_obj.serialize(id_encoder, serialization_options, for_link=True) else: rval["child_collection"] = element_obj.serialize(id_encoder, serialization_options) return rval
[docs]class Event(Base, RepresentById): __tablename__ = "event" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) history_id: Mapped[Optional[int]] = mapped_column(ForeignKey("history.id"), index=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) message: Mapped[Optional[str]] = mapped_column(TrimmedString(1024)) session_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_session.id"), index=True) tool_id: Mapped[Optional[str]] = mapped_column(String(255)) history: Mapped[Optional["History"]] = relationship() user: Mapped[Optional["User"]] = relationship() galaxy_session: Mapped[Optional["GalaxySession"]] = relationship()
[docs]class GalaxySession(Base, RepresentById): __tablename__ = "galaxy_session" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) remote_host: Mapped[Optional[str]] = mapped_column(String(255)) remote_addr: Mapped[Optional[str]] = mapped_column(String(255)) referer: Mapped[Optional[str]] = mapped_column(TEXT) current_history_id: Mapped[Optional[int]] = mapped_column(ForeignKey("history.id")) # unique 128 bit random number coerced to a string session_key: Mapped[Optional[str]] = mapped_column(TrimmedString(255), index=True, unique=True) is_valid: Mapped[Optional[bool]] = mapped_column(default=False) # saves a reference to the previous session so we have a way to chain them together prev_session_id: Mapped[Optional[int]] disk_usage: Mapped[Optional[Decimal]] = mapped_column(Numeric(15, 0), index=True) last_action: Mapped[Optional[datetime]] current_history: Mapped[Optional["History"]] = relationship() histories: Mapped[List["GalaxySessionToHistoryAssociation"]] = relationship( back_populates="galaxy_session", cascade_backrefs=False ) user: Mapped[Optional["User"]] = relationship(back_populates="galaxy_sessions") def __init__(self, is_valid=False, **kwd): super().__init__(**kwd) self.is_valid = is_valid self.last_action = self.last_action or now()
[docs] def add_history(self, history, association=None): if association is None: self.histories.append(GalaxySessionToHistoryAssociation(self, history)) else: self.histories.append(association)
[docs] def get_disk_usage(self): if self.disk_usage is None: return 0 return self.disk_usage
[docs] def set_disk_usage(self, bytes): self.disk_usage = bytes
total_disk_usage = property(get_disk_usage, set_disk_usage)
[docs]class GalaxySessionToHistoryAssociation(Base, RepresentById): __tablename__ = "galaxy_session_to_history" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) session_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_session.id"), index=True) history_id: Mapped[Optional[int]] = mapped_column(ForeignKey("history.id"), index=True) galaxy_session: Mapped[Optional["GalaxySession"]] = relationship(back_populates="histories") history: Mapped[Optional["History"]] = relationship(back_populates="galaxy_sessions") def __init__(self, galaxy_session, history): self.galaxy_session = galaxy_session ensure_object_added_to_session(self, object_in_session=galaxy_session) add_object_to_object_session(self, history) self.history = history
[docs]class StoredWorkflow(Base, HasTags, Dictifiable, RepresentById, UsesCreateAndUpdateTime): """ StoredWorkflow represents the root node of a tree of objects that compose a workflow, including workflow revisions, steps, and subworkflows. It is responsible for the metadata associated with a workflow including owner, name, published, and create/update time. Each time a workflow is modified a revision is created, represented by a new :class:`galaxy.model.Workflow` instance. See :class:`galaxy.model.Workflow` for more information """ __tablename__ = "stored_workflow" __table_args__ = (Index("ix_stored_workflow_slug", "slug", mysql_length=200),) id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, index=True, nullable=True) user_id: Mapped[int] = mapped_column(ForeignKey("galaxy_user.id"), index=True) latest_workflow_id: Mapped[Optional[int]] = mapped_column( ForeignKey("workflow.id", use_alter=True, name="stored_workflow_latest_workflow_id_fk"), index=True ) name: Mapped[Optional[str]] = mapped_column(TEXT) deleted: Mapped[Optional[bool]] = mapped_column(default=False) hidden: Mapped[Optional[bool]] = mapped_column(default=False) importable: Mapped[Optional[bool]] = mapped_column(default=False) slug: Mapped[Optional[str]] = mapped_column(TEXT) from_path: Mapped[Optional[str]] = mapped_column(TEXT) published: Mapped[Optional[bool]] = mapped_column(index=True, default=False) user: Mapped["User"] = relationship( primaryjoin=(lambda: User.id == StoredWorkflow.user_id), back_populates="stored_workflows" ) workflows: Mapped[List["Workflow"]] = relationship( back_populates="stored_workflow", cascade="all, delete-orphan", primaryjoin=(lambda: StoredWorkflow.id == Workflow.stored_workflow_id), order_by=lambda: -Workflow.id, cascade_backrefs=False, ) latest_workflow = relationship( "Workflow", post_update=True, primaryjoin=(lambda: StoredWorkflow.latest_workflow_id == Workflow.id), lazy=False, ) tags: Mapped[List["StoredWorkflowTagAssociation"]] = relationship( order_by=lambda: StoredWorkflowTagAssociation.id, back_populates="stored_workflow", ) owner_tags: Mapped[List["StoredWorkflowTagAssociation"]] = relationship( primaryjoin=( lambda: and_( StoredWorkflow.id == StoredWorkflowTagAssociation.stored_workflow_id, StoredWorkflow.user_id == StoredWorkflowTagAssociation.user_id, ) ), viewonly=True, order_by=lambda: StoredWorkflowTagAssociation.id, ) annotations: Mapped[List["StoredWorkflowAnnotationAssociation"]] = relationship( order_by=lambda: StoredWorkflowAnnotationAssociation.id, back_populates="stored_workflow", ) ratings: Mapped[List["StoredWorkflowRatingAssociation"]] = relationship( order_by=lambda: StoredWorkflowRatingAssociation.id, back_populates="stored_workflow", ) users_shared_with: Mapped[List["StoredWorkflowUserShareAssociation"]] = relationship( back_populates="stored_workflow" ) average_rating = None # Set up proxy so that # StoredWorkflow.users_shared_with # returns a list of users that workflow is shared with. users_shared_with_dot_users = association_proxy("users_shared_with", "user") dict_collection_visible_keys = [ "id", "name", "create_time", "update_time", "published", "importable", "deleted", "hidden", ] dict_element_visible_keys = [ "id", "name", "create_time", "update_time", "published", "importable", "deleted", "hidden", ] def __init__( self, user=None, name=None, slug=None, create_time=None, update_time=None, published=False, latest_workflow_id=None, workflow=None, hidden=False, ): add_object_to_object_session(self, user) self.user = user self.name = name self.slug = slug self.create_time = create_time self.update_time = update_time self.published = published self.latest_workflow = workflow self.workflows = listify(workflow) self.hidden = hidden
[docs] def get_internal_version(self, version): if version is None: return self.latest_workflow if len(self.workflows) <= version: raise galaxy.exceptions.RequestParameterInvalidException("Version does not exist") return list(reversed(self.workflows))[version]
[docs] def get_internal_version_by_id(self, workflow_instance_id: int): sa_session = object_session(self) assert sa_session workflow = sa_session.get(Workflow, workflow_instance_id) if not workflow: raise galaxy.exceptions.ObjectNotFound() elif workflow.stored_workflow != self: raise galaxy.exceptions.RequestParameterInvalidException() return workflow
[docs] def version_of(self, workflow): for version, workflow_instance in enumerate(reversed(self.workflows)): if workflow_instance.id == workflow.id: return version raise KeyError("Failed to find a version of target workflow instance in stored workflow.")
[docs] def show_in_tool_panel(self, user_id): sa_session = object_session(self) stmt = ( select(func.count()) .select_from(StoredWorkflowMenuEntry) .where(StoredWorkflowMenuEntry.stored_workflow_id == self.id) .where(StoredWorkflowMenuEntry.user_id == user_id) ) return bool(sa_session.scalar(stmt))
[docs] def copy_tags_from(self, target_user, source_workflow): # Override to only copy owner tags. for src_swta in source_workflow.owner_tags: new_swta = src_swta.copy() new_swta.user = target_user self.tags.append(new_swta)
[docs] def invocation_counts(self) -> InvocationsStateCounts: sa_session = object_session(self) assert sa_session stmt = ( select(WorkflowInvocation.state, func.count(WorkflowInvocation.state)) .select_from(StoredWorkflow) .join(Workflow, Workflow.stored_workflow_id == StoredWorkflow.id) .join(WorkflowInvocation, WorkflowInvocation.workflow_id == Workflow.id) .group_by(WorkflowInvocation.state) .where(StoredWorkflow.id == self.id) ) rows = sa_session.execute(stmt).all() rows_as_dict = dict(r for r in rows if r[0] is not None) # type:ignore[arg-type, var-annotated] return InvocationsStateCounts(rows_as_dict)
[docs] def to_dict(self, view="collection", value_mapper=None): rval = super().to_dict(view=view, value_mapper=value_mapper) rval["latest_workflow_uuid"] = (lambda uuid: str(uuid) if self.latest_workflow.uuid else None)( self.latest_workflow.uuid ) return rval
[docs]class Workflow(Base, Dictifiable, RepresentById): """ Workflow represents a revision of a :class:`galaxy.model.StoredWorkflow`. A new instance is created for each workflow revision and provides a common parent for the workflow steps. See :class:`galaxy.model.WorkflowStep` for more information """ __tablename__ = "workflow" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) # workflows will belong to either a stored workflow or a parent/nesting workflow. stored_workflow_id: Mapped[Optional[int]] = mapped_column(ForeignKey("stored_workflow.id"), index=True) parent_workflow_id: Mapped[Optional[int]] = mapped_column(ForeignKey("workflow.id"), index=True) name: Mapped[Optional[str]] = mapped_column(TEXT) has_cycles: Mapped[Optional[bool]] has_errors: Mapped[Optional[bool]] reports_config: Mapped[Optional[bytes]] = mapped_column(JSONType) creator_metadata: Mapped[Optional[bytes]] = mapped_column(JSONType) license: Mapped[Optional[str]] = mapped_column(TEXT) source_metadata: Mapped[Optional[Dict[str, str]]] = mapped_column(JSONType) uuid: Mapped[Optional[Union[UUID, str]]] = mapped_column(UUIDType) steps = relationship( "WorkflowStep", back_populates="workflow", primaryjoin=(lambda: Workflow.id == WorkflowStep.workflow_id), order_by=lambda: asc(WorkflowStep.order_index), cascade="all, delete-orphan", lazy=False, ) comments: Mapped[List["WorkflowComment"]] = relationship( back_populates="workflow", primaryjoin=(lambda: Workflow.id == WorkflowComment.workflow_id), cascade="all, delete-orphan", lazy=False, ) parent_workflow_steps = relationship( "WorkflowStep", primaryjoin=(lambda: Workflow.id == WorkflowStep.subworkflow_id), back_populates="subworkflow", cascade_backrefs=False, ) stored_workflow = relationship( "StoredWorkflow", primaryjoin=(lambda: StoredWorkflow.id == Workflow.stored_workflow_id), back_populates="workflows", ) step_count = None dict_collection_visible_keys = ["name", "has_cycles", "has_errors"] dict_element_visible_keys = ["name", "has_cycles", "has_errors"] input_step_types = ["data_input", "data_collection_input", "parameter_input"] def __init__(self, uuid=None): self.user = None self.uuid = get_uuid(uuid)
[docs] def has_outputs_defined(self): """ Returns true or false indicating whether or not a workflow has outputs defined. """ for step in self.steps: if step.workflow_outputs: return True return False
[docs] def to_dict(self, view="collection", value_mapper=None): rval = super().to_dict(view=view, value_mapper=value_mapper) rval["uuid"] = (lambda uuid: str(uuid) if uuid else None)(self.uuid) return rval
@property def steps_by_id(self): steps = {} for step in self.steps: step_id = step.id steps[step_id] = step return steps
[docs] def step_by_index(self, order_index: int): for step in self.steps: if order_index == step.order_index: return step raise KeyError(f"Workflow has no step with order_index '{order_index}'")
[docs] def step_by_label(self, label): for step in self.steps: if label == step.label: return step raise KeyError(f"Workflow has no step with label '{label}'")
@property def input_steps(self): for step in self.steps: if step.type in Workflow.input_step_types: yield step @property def workflow_outputs(self): for step in self.steps: yield from step.workflow_outputs
[docs] def workflow_output_for(self, output_label): target_output = None for workflow_output in self.workflow_outputs: if workflow_output.label == output_label: target_output = workflow_output break return target_output
@property def workflow_output_labels(self): names = [] for workflow_output in self.workflow_outputs: names.append(workflow_output.label) return names @property def top_level_workflow(self): """If this workflow is not attached to stored workflow directly, recursively grab its parents until it is the top level workflow which must have a stored workflow associated with it. """ top_level_workflow = self if self.stored_workflow is None: # TODO: enforce this at creation... assert len({w.uuid for w in self.parent_workflow_steps}) == 1 return self.parent_workflow_steps[0].workflow.top_level_workflow return top_level_workflow @property def top_level_stored_workflow(self): """If this workflow is not attached to stored workflow directly, recursively grab its parents until it is the top level workflow which must have a stored workflow associated with it and then grab that stored workflow. """ return self.top_level_workflow.stored_workflow
[docs] def copy(self, user=None): """Copy a workflow for a new StoredWorkflow object. Pass user if user-specific information needed. """ copied_workflow = Workflow() copied_workflow.name = self.name copied_workflow.has_cycles = self.has_cycles copied_workflow.has_errors = self.has_errors copied_workflow.reports_config = self.reports_config copied_workflow.license = self.license copied_workflow.creator_metadata = self.creator_metadata # Map old step ids to new steps step_mapping = {} copied_steps = [] for step in self.steps: copied_step = WorkflowStep() copied_steps.append(copied_step) step_mapping[step.id] = copied_step for old_step, new_step in zip(self.steps, copied_steps): old_step.copy_to(new_step, step_mapping, user=user) copied_workflow.steps = copied_steps copied_comments = [comment.copy() for comment in self.comments] steps_by_id = {s.order_index: s for s in copied_workflow.steps} comments_by_id = {c.order_index: c for c in copied_comments} # copy comment relationships for old_comment, new_comment in zip(self.comments, copied_comments): for step_id in [step.order_index for step in old_comment.child_steps]: child_step = steps_by_id.get(step_id) if child_step: child_step.parent_comment = new_comment for comment_id in [comment.order_index for comment in old_comment.child_comments]: child_comment = comments_by_id.get(comment_id) if child_comment: child_comment.parent_comment = new_comment copied_workflow.comments = copied_comments return copied_workflow
@property def version(self): return self.stored_workflow.version_of(self)
[docs] def log_str(self): extra = "" if self.stored_workflow: extra = f",name={self.stored_workflow.name}" return "Workflow[id=%d%s]" % (self.id, extra)
InputConnDictType = Dict[str, Union[Dict[str, Any], List[Dict[str, Any]]]]
[docs]class WorkflowStep(Base, RepresentById, UsesCreateAndUpdateTime): """ WorkflowStep represents a tool or subworkflow, its inputs, annotations, and any outputs that are flagged as workflow outputs. See :class:`galaxy.model.WorkflowStepInput` and :class:`galaxy.model.WorkflowStepConnection` for more information. """ __tablename__ = "workflow_step" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) workflow_id: Mapped[int] = mapped_column(ForeignKey("workflow.id"), index=True) subworkflow_id: Mapped[Optional[int]] = mapped_column(ForeignKey("workflow.id"), index=True) dynamic_tool_id: Mapped[Optional[int]] = mapped_column(ForeignKey("dynamic_tool.id"), index=True) type: Mapped[Optional[str]] = mapped_column(String(64)) tool_id: Mapped[Optional[str]] = mapped_column(TEXT) tool_version: Mapped[Optional[str]] = mapped_column(TEXT) tool_inputs: Mapped[Optional[Dict[str, Any]]] = mapped_column(JSONType) tool_errors: Mapped[Optional[bytes]] = mapped_column(JSONType) position: Mapped[Optional[bytes]] = mapped_column(MutableJSONType) config: Mapped[Optional[bytes]] = mapped_column(JSONType) order_index: Mapped[Optional[int]] when_expression: Mapped[Optional[bytes]] = mapped_column(JSONType) uuid: Mapped[Optional[Union[UUID, str]]] = mapped_column(UUIDType) label: Mapped[Optional[str]] = mapped_column(Unicode(255)) temp_input_connections = None parent_comment_id: Mapped[Optional[int]] = mapped_column(ForeignKey("workflow_comment.id"), index=True) parent_comment: Mapped[Optional["WorkflowComment"]] = relationship( primaryjoin=(lambda: WorkflowComment.id == WorkflowStep.parent_comment_id), back_populates="child_steps", ) subworkflow: Mapped[Optional["Workflow"]] = relationship( primaryjoin=(lambda: Workflow.id == WorkflowStep.subworkflow_id), back_populates="parent_workflow_steps", ) dynamic_tool: Mapped[Optional["DynamicTool"]] = relationship( primaryjoin=(lambda: DynamicTool.id == WorkflowStep.dynamic_tool_id) ) tags: Mapped[List["WorkflowStepTagAssociation"]] = relationship( order_by=lambda: WorkflowStepTagAssociation.id, back_populates="workflow_step" ) annotations: Mapped[List["WorkflowStepAnnotationAssociation"]] = relationship( order_by=lambda: WorkflowStepAnnotationAssociation.id, back_populates="workflow_step", ) post_job_actions = relationship("PostJobAction", back_populates="workflow_step", cascade_backrefs=False) inputs = relationship("WorkflowStepInput", back_populates="workflow_step") workflow_outputs: Mapped[List["WorkflowOutput"]] = relationship( back_populates="workflow_step", cascade_backrefs=False ) output_connections: Mapped[List["WorkflowStepConnection"]] = relationship( primaryjoin=(lambda: WorkflowStepConnection.output_step_id == WorkflowStep.id) ) workflow: Mapped["Workflow"] = relationship( primaryjoin=(lambda: Workflow.id == WorkflowStep.workflow_id), back_populates="steps", cascade_backrefs=False, ) STEP_TYPE_TO_INPUT_TYPE = { "data_input": "dataset", "data_collection_input": "dataset_collection", "parameter_input": "parameter", } DEFAULT_POSITION = {"left": 0, "top": 0} def __init__(self): self.uuid = uuid4() self._input_connections_by_name = None self._inputs_by_name = None # Injected attributes # TODO: code using these should be refactored to not depend on these non-persistent fields self.module: Optional[WorkflowModule] self.state: Optional[DefaultToolState] self.upgrade_messages: Optional[Dict]
[docs] @reconstructor def init_on_load(self): self._input_connections_by_name = None self._inputs_by_name = None
@property def tool_uuid(self): return self.dynamic_tool and self.dynamic_tool.uuid @property def is_input_type(self) -> bool: return bool(self.type and self.type in self.STEP_TYPE_TO_INPUT_TYPE) @property def input_type(self): assert self.is_input_type, "step.input_type can only be called on input step types" return self.STEP_TYPE_TO_INPUT_TYPE[self.type] @property def input_default_value(self): self.get_input_default_value(None)
[docs] def get_input_default_value(self, default_default): # parameter_input and the data parameters handle this slightly differently # unfortunately. if self.type == "parameter_input": tool_state = self.tool_inputs default_value = tool_state.get("default", default_default) else: default_value = default_default for step_input in self.inputs: if step_input.name == "input" and step_input.default_value_set: default_value = step_input.default_value break return default_value
@property def input_optional(self): tool_state = self.tool_inputs return tool_state.get("optional") or False
[docs] def setup_inputs_by_name(self): # Ensure input_connections has already been set. # Make connection information available on each step by input name. inputs_by_name = {} for step_input in self.inputs: input_name = step_input.name assert input_name not in inputs_by_name inputs_by_name[input_name] = step_input self._inputs_by_name = inputs_by_name
@property def inputs_by_name(self): if self._inputs_by_name is None: self.setup_inputs_by_name() return self._inputs_by_name
[docs] def get_input(self, input_name): for step_input in self.inputs: if step_input.name == input_name: return step_input return None
[docs] def get_or_add_input(self, input_name): step_input = self.get_input(input_name) if step_input is None: step_input = WorkflowStepInput(self) step_input.name = input_name return step_input
[docs] def add_connection(self, input_name, output_name, output_step, input_subworkflow_step_index=None): step_input = self.get_or_add_input(input_name) conn = WorkflowStepConnection() conn.input_step_input = step_input ensure_object_added_to_session(conn, object_in_session=step_input) conn.output_name = output_name add_object_to_object_session(conn, output_step) conn.output_step = output_step if self.subworkflow: if input_subworkflow_step_index is not None: input_subworkflow_step = self.subworkflow.step_by_index(input_subworkflow_step_index) else: input_subworkflow_steps = [step for step in self.subworkflow.input_steps if step.label == input_name] if not input_subworkflow_steps: inferred_order_index = input_name.split(":", 1)[0] if inferred_order_index.isdigit(): input_subworkflow_steps = [self.subworkflow.step_by_index(int(inferred_order_index))] if len(input_subworkflow_steps) != 1: # `when` expression inputs don't need to be passed into subworkflow # In the absence of formal extra step inputs this seems like the best we can do. # A better way to do these validations is to validate that all required subworkflow inputs # are connected. if input_name not in (self.when_expression or ""): raise galaxy.exceptions.MessageException( f"Invalid subworkflow connection at step index {self.order_index + 1}" ) else: input_subworkflow_steps = [None] input_subworkflow_step = input_subworkflow_steps[0] conn.input_subworkflow_step = input_subworkflow_step return conn
@property def input_connections(self): connections = [_ for step_input in self.inputs for _ in step_input.connections] return connections @property def unique_workflow_outputs(self): # Older Galaxy workflows may have multiple WorkflowOutputs # per "output_name", when serving these back to the editor # feed only a "best" output per "output_name."" outputs = {} for workflow_output in self.workflow_outputs: output_name = workflow_output.output_name if output_name in outputs: found_output = outputs[output_name] if found_output.label is None and workflow_output.label is not None: outputs[output_name] = workflow_output else: outputs[output_name] = workflow_output return list(outputs.values()) @property def content_id(self): content_id = None if self.type == "tool": content_id = self.tool_id elif self.type == "subworkflow": content_id = self.subworkflow.id else: content_id = None return content_id @property def input_connections_by_name(self): if self._input_connections_by_name is None: self.setup_input_connections_by_name() return self._input_connections_by_name
[docs] def setup_input_connections_by_name(self): # Ensure input_connections has already been set. # Make connection information available on each step by input name. input_connections_by_name = {} for conn in self.input_connections: input_name = conn.input_name if input_name not in input_connections_by_name: input_connections_by_name[input_name] = [] input_connections_by_name[input_name].append(conn) self._input_connections_by_name = input_connections_by_name
[docs] def create_or_update_workflow_output(self, output_name, label, uuid): output = self.workflow_output_for(output_name) if output is None: output = WorkflowOutput(workflow_step=self, output_name=output_name) if uuid is not None: output.uuid = uuid if label is not None: output.label = label return output
[docs] def workflow_output_for(self, output_name): target_output = None for workflow_output in self.workflow_outputs: if workflow_output.output_name == output_name: target_output = workflow_output break return target_output
[docs] def copy_to(self, copied_step, step_mapping, user=None): copied_step.order_index = self.order_index copied_step.type = self.type copied_step.tool_id = self.tool_id copied_step.tool_version = self.tool_version copied_step.tool_inputs = self.tool_inputs copied_step.tool_errors = self.tool_errors copied_step.position = self.position copied_step.config = self.config copied_step.label = self.label copied_step.when_expression = self.when_expression copied_step.inputs = copy_list(self.inputs, copied_step) subworkflow_step_mapping = {} if user is not None and self.annotations: annotations = [] for annotation in self.annotations: association = WorkflowStepAnnotationAssociation() association.user = user association.workflow_step = copied_step association.annotation = annotation.annotation annotations.append(association) copied_step.annotations = annotations if subworkflow := self.subworkflow: stored_subworkflow = subworkflow.stored_workflow if stored_subworkflow and stored_subworkflow.user == user: # This should be fine and reduces the number of stored subworkflows copied_step.subworkflow = subworkflow copied_subworkflow = subworkflow else: # Can this even happen, building a workflow with a subworkflow you don't own ? copied_subworkflow = subworkflow.copy() stored_workflow = StoredWorkflow( user, name=copied_subworkflow.name, workflow=copied_subworkflow, hidden=True ) copied_subworkflow.stored_workflow = stored_workflow copied_step.subworkflow = copied_subworkflow for subworkflow_step, copied_subworkflow_step in zip(subworkflow.steps, copied_subworkflow.steps): subworkflow_step_mapping[subworkflow_step.id] = copied_subworkflow_step for old_conn, new_conn in zip(self.input_connections, copied_step.input_connections): new_conn.input_step_input = copied_step.get_or_add_input(old_conn.input_name) new_conn.output_step = step_mapping[old_conn.output_step_id] if old_conn.input_subworkflow_step_id: new_conn.input_subworkflow_step = subworkflow_step_mapping[old_conn.input_subworkflow_step_id] for orig_pja in self.post_job_actions: PostJobAction( orig_pja.action_type, copied_step, output_name=orig_pja.output_name, action_arguments=orig_pja.action_arguments, ) copied_step.workflow_outputs = copy_list(self.workflow_outputs, copied_step)
[docs] def log_str(self): return ( f"WorkflowStep[index={self.order_index},type={self.type},label={self.label},uuid={self.uuid},id={self.id}]" )
@property def effective_label(self) -> Optional[str]: label = self.label if label is not None: return label elif self.is_input_type: tool_inputs = self.tool_inputs if tool_inputs is not None: return cast(Optional[str], tool_inputs.get("name")) return None
[docs] def clear_module_extras(self): # the module code adds random dynamic state to the step, this # attempts to clear that. for module_attribute in ["module"]: try: delattr(self, module_attribute) except AttributeError: pass
[docs]class WorkflowStepInput(Base, RepresentById): __tablename__ = "workflow_step_input" __table_args__ = ( Index( "ix_workflow_step_input_workflow_step_id_name_unique", "workflow_step_id", "name", unique=True, mysql_length={"name": 200}, ), ) id: Mapped[int] = mapped_column(primary_key=True) workflow_step_id: Mapped[Optional[int]] = mapped_column(ForeignKey("workflow_step.id"), index=True) name: Mapped[Optional[str]] = mapped_column(TEXT) merge_type: Mapped[Optional[str]] = mapped_column(TEXT) scatter_type: Mapped[Optional[str]] = mapped_column(TEXT) value_from: Mapped[Optional[bytes]] = mapped_column(MutableJSONType) value_from_type: Mapped[Optional[str]] = mapped_column(TEXT) default_value: Mapped[Optional[bytes]] = mapped_column(MutableJSONType) default_value_set: Mapped[Optional[bool]] = mapped_column(default=False) runtime_value: Mapped[Optional[bool]] = mapped_column(default=False) workflow_step: Mapped[Optional["WorkflowStep"]] = relationship( back_populates="inputs", cascade="all", primaryjoin=(lambda: WorkflowStepInput.workflow_step_id == WorkflowStep.id), ) connections: Mapped[List["WorkflowStepConnection"]] = relationship( back_populates="input_step_input", primaryjoin=(lambda: WorkflowStepConnection.input_step_input_id == WorkflowStepInput.id), cascade_backrefs=False, ) def __init__(self, workflow_step): add_object_to_object_session(self, workflow_step) self.workflow_step = workflow_step self.default_value_set = False
[docs] def copy(self, copied_step): copied_step_input = WorkflowStepInput(copied_step) copied_step_input.name = self.name copied_step_input.default_value = self.default_value copied_step_input.default_value_set = self.default_value_set copied_step_input.merge_type = self.merge_type copied_step_input.scatter_type = self.scatter_type copied_step_input.connections = copy_list(self.connections) return copied_step_input
[docs]class WorkflowStepConnection(Base, RepresentById): __tablename__ = "workflow_step_connection" id: Mapped[int] = mapped_column(primary_key=True) output_step_id: Mapped[Optional[int]] = mapped_column(ForeignKey("workflow_step.id"), index=True) input_step_input_id: Mapped[Optional[int]] = mapped_column(ForeignKey("workflow_step_input.id"), index=True) output_name: Mapped[Optional[str]] = mapped_column(TEXT) input_subworkflow_step_id: Mapped[Optional[int]] = mapped_column(ForeignKey("workflow_step.id"), index=True) input_step_input = relationship( "WorkflowStepInput", back_populates="connections", cascade="all", primaryjoin=(lambda: WorkflowStepConnection.input_step_input_id == WorkflowStepInput.id), ) input_subworkflow_step = relationship( "WorkflowStep", primaryjoin=(lambda: WorkflowStepConnection.input_subworkflow_step_id == WorkflowStep.id) ) output_step = relationship( "WorkflowStep", back_populates="output_connections", cascade="all", primaryjoin=(lambda: WorkflowStepConnection.output_step_id == WorkflowStep.id), ) # Constant used in lieu of output_name and input_name to indicate an # implicit connection between two steps that is not dependent on a dataset # or a dataset collection. Allowing for instance data manager steps to setup # index data before a normal tool runs or for workflows that manage data # outside of Galaxy. NON_DATA_CONNECTION = "__NO_INPUT_OUTPUT_NAME__" @property def non_data_connection(self): return self.output_name == self.input_name == WorkflowStepConnection.NON_DATA_CONNECTION @property def input_name(self): return self.input_step_input.name @property def input_step(self) -> Optional[WorkflowStep]: return self.input_step_input and self.input_step_input.workflow_step @property def input_step_id(self): input_step = self.input_step return input_step and input_step.id
[docs] def copy(self): # TODO: handle subworkflow ids... copied_connection = WorkflowStepConnection() copied_connection.output_name = self.output_name return copied_connection
[docs]class WorkflowOutput(Base, Serializable): __tablename__ = "workflow_output" id: Mapped[int] = mapped_column(primary_key=True) workflow_step_id: Mapped[int] = mapped_column(ForeignKey("workflow_step.id"), index=True) output_name: Mapped[Optional[str]] = mapped_column(String(255)) label: Mapped[Optional[str]] = mapped_column(Unicode(255)) uuid: Mapped[Optional[Union[UUID, str]]] = mapped_column(UUIDType) workflow_step: Mapped["WorkflowStep"] = relationship( back_populates="workflow_outputs", primaryjoin=(lambda: WorkflowStep.id == WorkflowOutput.workflow_step_id), ) def __init__(self, workflow_step, output_name=None, label=None, uuid=None): self.workflow_step = workflow_step ensure_object_added_to_session(self, object_in_session=workflow_step) self.output_name = output_name self.label = label self.uuid = get_uuid(uuid)
[docs] def copy(self, copied_step): copied_output = WorkflowOutput(copied_step) copied_output.output_name = self.output_name copied_output.label = self.label return copied_output
def _serialize(self, id_encoder, serialization_options): return dict_for( self, output_name=self.output_name, label=self.label, uuid=str(self.uuid), )
[docs]class WorkflowComment(Base, RepresentById): """ WorkflowComment represents an in-editor comment which is no associated to any WorkflowStep. It is purely decorative, and should not influence how a workflow is ran. """ __tablename__ = "workflow_comment" id: Mapped[int] = mapped_column(primary_key=True) order_index: Mapped[Optional[int]] workflow_id: Mapped[int] = mapped_column(ForeignKey("workflow.id"), index=True) position: Mapped[Optional[bytes]] = mapped_column(MutableJSONType) size: Mapped[Optional[bytes]] = mapped_column(JSONType) type: Mapped[Optional[str]] = mapped_column(String(16)) color: Mapped[Optional[str]] = mapped_column(String(16)) data: Mapped[Optional[bytes]] = mapped_column(JSONType) parent_comment_id: Mapped[Optional[int]] = mapped_column(ForeignKey("workflow_comment.id"), index=True) workflow: Mapped["Workflow"] = relationship( primaryjoin=(lambda: Workflow.id == WorkflowComment.workflow_id), back_populates="comments", ) child_steps = relationship( "WorkflowStep", primaryjoin=(lambda: WorkflowStep.parent_comment_id == WorkflowComment.id), back_populates="parent_comment", ) parent_comment: Mapped[Optional["WorkflowComment"]] = relationship( primaryjoin=(lambda: WorkflowComment.id == WorkflowComment.parent_comment_id), back_populates="child_comments", remote_side=[id], ) child_comments: Mapped[List["WorkflowComment"]] = relationship( primaryjoin=(lambda: WorkflowComment.parent_comment_id == WorkflowComment.id), back_populates="parent_comment", )
[docs] def to_dict(self): comment_dict = { "id": self.order_index, "position": self.position, "size": self.size, "type": self.type, "color": self.color, "data": self.data, } if self.child_steps: comment_dict["child_steps"] = [step.order_index for step in self.child_steps] if self.child_comments: comment_dict["child_comments"] = [comment.order_index for comment in self.child_comments] WorkflowCommentModel(root=comment_dict) return comment_dict
[docs] def from_dict(dict): WorkflowCommentModel(root=dict) comment = WorkflowComment() comment.order_index = dict.get("id", 0) comment.type = dict.get("type", "text") comment.position = dict.get("position", None) comment.size = dict.get("size", None) comment.color = dict.get("color", "none") comment.data = dict.get("data", None) return comment
[docs] def copy(self): comment = WorkflowComment() comment.order_index = self.order_index comment.type = self.type comment.position = self.position comment.size = self.size comment.color = self.color comment.data = self.data return comment
[docs]class StoredWorkflowUserShareAssociation(Base, UserShareAssociation): __tablename__ = "stored_workflow_user_share_connection" id: Mapped[int] = mapped_column(primary_key=True) stored_workflow_id: Mapped[int] = mapped_column(ForeignKey("stored_workflow.id"), index=True, nullable=True) user_id: Mapped[int] = mapped_column(ForeignKey("galaxy_user.id"), index=True, nullable=True) user: Mapped[User] = relationship() stored_workflow: Mapped["StoredWorkflow"] = relationship(back_populates="users_shared_with")
[docs]class StoredWorkflowMenuEntry(Base, RepresentById): __tablename__ = "stored_workflow_menu_entry" id: Mapped[int] = mapped_column(primary_key=True) stored_workflow_id: Mapped[Optional[int]] = mapped_column(ForeignKey("stored_workflow.id"), index=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) order_index: Mapped[Optional[int]] stored_workflow: Mapped[Optional["StoredWorkflow"]] = relationship() user: Mapped[Optional["User"]] = relationship( back_populates="stored_workflow_menu_entries", primaryjoin=( lambda: (StoredWorkflowMenuEntry.user_id == User.id) & (StoredWorkflowMenuEntry.stored_workflow_id == StoredWorkflow.id) & not_(StoredWorkflow.deleted) ), )
[docs]@dataclass class InputWithRequest: input: Any request: Dict[str, Any]
[docs]@dataclass class InputToMaterialize: hda: "HistoryDatasetAssociation" input_dataset: "WorkflowRequestToInputDatasetAssociation"
[docs]class WorkflowInvocation(Base, UsesCreateAndUpdateTime, Dictifiable, Serializable): __tablename__ = "workflow_invocation" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, index=True, nullable=True) workflow_id: Mapped[int] = mapped_column(ForeignKey("workflow.id"), index=True) state: Mapped[Optional[str]] = mapped_column(TrimmedString(64), index=True) scheduler: Mapped[Optional[str]] = mapped_column(TrimmedString(255), index=True) handler: Mapped[Optional[str]] = mapped_column(TrimmedString(255), index=True) uuid: Mapped[Optional[Union[UUID, str]]] = mapped_column(UUIDType()) history_id: Mapped[Optional[int]] = mapped_column(ForeignKey("history.id"), index=True) history = relationship("History", back_populates="workflow_invocations") input_parameters = relationship( "WorkflowRequestInputParameter", back_populates="workflow_invocation", cascade_backrefs=False ) step_states = relationship("WorkflowRequestStepState", back_populates="workflow_invocation", cascade_backrefs=False) input_step_parameters = relationship( "WorkflowRequestInputStepParameter", back_populates="workflow_invocation", cascade_backrefs=False ) input_datasets = relationship( "WorkflowRequestToInputDatasetAssociation", back_populates="workflow_invocation", cascade_backrefs=False ) input_dataset_collections = relationship( "WorkflowRequestToInputDatasetCollectionAssociation", back_populates="workflow_invocation", cascade_backrefs=False, ) subworkflow_invocations = relationship( "WorkflowInvocationToSubworkflowInvocationAssociation", primaryjoin=( lambda: WorkflowInvocationToSubworkflowInvocationAssociation.workflow_invocation_id == WorkflowInvocation.id ), back_populates="parent_workflow_invocation", uselist=True, ) steps = relationship( "WorkflowInvocationStep", back_populates="workflow_invocation", order_by=lambda: WorkflowInvocationStep.order_index, cascade_backrefs=False, ) workflow = relationship("Workflow") output_dataset_collections = relationship( "WorkflowInvocationOutputDatasetCollectionAssociation", back_populates="workflow_invocation", cascade_backrefs=False, ) output_datasets = relationship( "WorkflowInvocationOutputDatasetAssociation", back_populates="workflow_invocation", cascade_backrefs=False ) output_values = relationship( "WorkflowInvocationOutputValue", back_populates="workflow_invocation", cascade_backrefs=False ) messages = relationship("WorkflowInvocationMessage", back_populates="workflow_invocation") dict_collection_visible_keys = [ "id", "update_time", "create_time", "workflow_id", "history_id", "uuid", "state", ] dict_element_visible_keys = [ "id", "update_time", "create_time", "workflow_id", "history_id", "uuid", "state", ] states = InvocationState non_terminal_states = [states.NEW, states.READY]
[docs] def create_subworkflow_invocation_for_step(self, step): assert step.type == "subworkflow" subworkflow_invocation = WorkflowInvocation() self.attach_subworkflow_invocation_for_step(step, subworkflow_invocation) return subworkflow_invocation
[docs] def attach_subworkflow_invocation_for_step(self, step, subworkflow_invocation): assert step.type == "subworkflow" assoc = WorkflowInvocationToSubworkflowInvocationAssociation() assoc.workflow_invocation = self assoc.workflow_step = step add_object_to_object_session(subworkflow_invocation, self.history) subworkflow_invocation.history = self.history subworkflow_invocation.workflow = step.subworkflow assoc.subworkflow_invocation = subworkflow_invocation self.subworkflow_invocations.append(assoc) return assoc
[docs] def get_subworkflow_invocation_for_step(self, step): assoc = self.get_subworkflow_invocation_association_for_step(step) return assoc.subworkflow_invocation
[docs] def get_subworkflow_invocation_association_for_step(self, step): assert step.type == "subworkflow" assoc = None for subworkflow_invocation in self.subworkflow_invocations: if subworkflow_invocation.workflow_step == step: assoc = subworkflow_invocation break return assoc
@property def active(self): """Indicates the workflow invocation is somehow active - and in particular valid actions may be performed on its WorkflowInvocationSteps. """ states = WorkflowInvocation.states return self.state in [states.NEW, states.READY]
[docs] def set_state(self, state: InvocationState): session = object_session(self) priority_states = (WorkflowInvocation.states.CANCELLING, WorkflowInvocation.states.CANCELLED) if session and self.id and state not in priority_states: # generate statement that will not revert CANCELLING or CANCELLED back to anything non-terminal session.execute( update(WorkflowInvocation) .where( WorkflowInvocation.id == self.id, or_(~WorkflowInvocation.state.in_(priority_states), WorkflowInvocation.state.is_(None)), ) .values(state=state) ) else: # Not bound to a session, or setting cancelling/cancelled self.state = state
[docs] def cancel(self): if self.state not in [WorkflowInvocation.states.CANCELLING, WorkflowInvocation.states.CANCELLED]: # No use cancelling workflow again, for all others we may still want to be able to cancel # remaining tool and workflow steps self.state = WorkflowInvocation.states.CANCELLING return True return False
[docs] def cancel_invocation_steps(self): sa_session = object_session(self) assert sa_session job_subq = ( select(Job.id) .join(WorkflowInvocationStep) .filter(WorkflowInvocationStep.workflow_invocation_id == self.id) .filter(~Job.state.in_(Job.finished_states)) .with_for_update() ) sa_session.execute(update(Job).where(Job.id.in_(job_subq)).values({"state": Job.states.DELETING})) job_collection_subq = ( select(Job.id) .join(ImplicitCollectionJobsJobAssociation) .join(ImplicitCollectionJobs) .join( WorkflowInvocationStep, WorkflowInvocationStep.implicit_collection_jobs_id == ImplicitCollectionJobs.id ) .filter(WorkflowInvocationStep.workflow_invocation_id == self.id) .filter(~Job.state.in_(Job.finished_states)) .with_for_update() .subquery() ) sa_session.execute( update(Job).where(Job.id.in_(job_collection_subq.element)).values({"state": Job.states.DELETING}) ) for invocation in self.subworkflow_invocations: subworkflow_invocation = invocation.subworkflow_invocation cancelled = subworkflow_invocation.cancel() if cancelled: subworkflow_invocation.add_message(InvocationCancellationUserRequest(reason="user_request")) sa_session.add(subworkflow_invocation) sa_session.commit()
[docs] def mark_cancelled(self): self.state = WorkflowInvocation.states.CANCELLED
[docs] def fail(self): self.state = WorkflowInvocation.states.FAILED
[docs] def step_states_by_step_id(self): step_states = {} for step_state in self.step_states: step_id = step_state.workflow_step_id step_states[step_id] = step_state return step_states
[docs] def step_invocations_by_step_id(self): step_invocations = {} for invocation_step in self.steps: step_id = invocation_step.workflow_step_id assert step_id not in step_invocations step_invocations[step_id] = invocation_step return step_invocations
[docs] def step_invocation_for_step_id(self, step_id: int) -> Optional["WorkflowInvocationStep"]: target_invocation_step = None for invocation_step in self.steps: if step_id == invocation_step.workflow_step_id: target_invocation_step = invocation_step return target_invocation_step
[docs] def step_invocation_for_label(self, label): target_invocation_step = None for invocation_step in self.steps: if label == invocation_step.workflow_step.label: target_invocation_step = invocation_step return target_invocation_step
[docs] @staticmethod def poll_unhandled_workflow_ids(sa_session): stmt = ( select(WorkflowInvocation.id) .where(WorkflowInvocation.state == WorkflowInvocation.states.NEW) .where(WorkflowInvocation.handler.is_(None)) .order_by(WorkflowInvocation.id.asc()) ) return list(sa_session.scalars(stmt))
[docs] @staticmethod def poll_active_workflow_ids(engine, scheduler=None, handler=None): and_conditions = [ or_( WorkflowInvocation.state == WorkflowInvocation.states.NEW, WorkflowInvocation.state == WorkflowInvocation.states.REQUIRES_MATERIALIZATION, WorkflowInvocation.state == WorkflowInvocation.states.READY, WorkflowInvocation.state == WorkflowInvocation.states.CANCELLING, ), ] if scheduler is not None: and_conditions.append(WorkflowInvocation.scheduler == scheduler) if handler is not None: and_conditions.append(WorkflowInvocation.handler == handler) stmt = select(WorkflowInvocation.id).filter(and_(*and_conditions)).order_by(WorkflowInvocation.id.asc()) # Immediately just load all ids into memory so time slicing logic # is relatively intutitive. with engine.connect() as conn: return conn.scalars(stmt).all()
[docs] def add_output(self, workflow_output, step, output_object): if not hasattr(output_object, "history_content_type"): # assuming this is a simple type, just JSON-ify it and stick in the database. In the future # I'd like parameter_inputs to have datasets and collections as valid parameter types so # dispatch on actual object and not step type. output_assoc = WorkflowInvocationOutputValue() output_assoc.workflow_invocation = self ensure_object_added_to_session(output_assoc, object_in_session=self) output_assoc.workflow_output = workflow_output output_assoc.workflow_step = step output_assoc.value = output_object self.output_values.append(output_assoc) elif output_object.history_content_type == "dataset": output_assoc = WorkflowInvocationOutputDatasetAssociation() output_assoc.workflow_invocation = self ensure_object_added_to_session(output_assoc, object_in_session=self) output_assoc.workflow_output = workflow_output output_assoc.workflow_step = step output_assoc.dataset = output_object self.output_datasets.append(output_assoc) elif output_object.history_content_type == "dataset_collection": output_assoc = WorkflowInvocationOutputDatasetCollectionAssociation() output_assoc.workflow_invocation = self ensure_object_added_to_session(output_assoc, object_in_session=self) output_assoc.workflow_output = workflow_output output_assoc.workflow_step = step output_assoc.dataset_collection = output_object self.output_dataset_collections.append(output_assoc) else: raise Exception("Unknown output type encountered")
[docs] def get_output_object(self, label): for output_dataset_assoc in self.output_datasets: if output_dataset_assoc.workflow_output.label == label: return output_dataset_assoc.dataset for output_dataset_collection_assoc in self.output_dataset_collections: if output_dataset_collection_assoc.workflow_output.label == label: return output_dataset_collection_assoc.dataset_collection # That probably isn't good. workflow_output = self.workflow.workflow_output_for(label) if workflow_output: raise galaxy.exceptions.MessageException( f"Failed to find workflow output named [{label}], one was defined but none registered during execution." ) else: raise galaxy.exceptions.MessageException( f"Failed to find workflow output named [{label}], workflow doesn't define output by that name - valid names are {self.workflow.workflow_output_labels}." )
[docs] def get_input_object(self, label): for input_dataset_assoc in self.input_datasets: if input_dataset_assoc.workflow_step.label == label: return input_dataset_assoc.dataset for input_dataset_collection_assoc in self.input_dataset_collections: if input_dataset_collection_assoc.workflow_step.label == label: return input_dataset_collection_assoc.dataset_collection raise Exception(f"Failed to find input with label {label}")
@property def output_associations(self): outputs = [] for output_dataset_assoc in self.output_datasets: outputs.append(output_dataset_assoc) for output_dataset_collection_assoc in self.output_dataset_collections: outputs.append(output_dataset_collection_assoc) return outputs @property def input_associations(self): inputs = [] for input_dataset_assoc in self.input_datasets: inputs.append(input_dataset_assoc) for input_dataset_collection_assoc in self.input_dataset_collections: inputs.append(input_dataset_collection_assoc) return inputs
[docs] def inputs_requiring_materialization(self) -> List[InputToMaterialize]: hdas_to_materialize: List[InputToMaterialize] = [] for input_dataset_assoc in self.input_datasets: request = input_dataset_assoc.request if request: deferred = request.get("deferred", False) if not deferred: hdas_to_materialize.append( InputToMaterialize( input_dataset_assoc.dataset, input_dataset_assoc, ) ) return hdas_to_materialize
def _serialize(self, id_encoder, serialization_options): invocation_attrs = dict_for(self) invocation_attrs["state"] = self.state invocation_attrs["create_time"] = self.create_time.__str__() invocation_attrs["update_time"] = self.update_time.__str__() steps = [] for step in self.steps: steps.append(step.serialize(id_encoder, serialization_options)) invocation_attrs["steps"] = steps input_parameters = [] for input_parameter in self.input_parameters: input_parameters.append(input_parameter.serialize(id_encoder, serialization_options)) invocation_attrs["input_parameters"] = input_parameters step_states = [] for step_state in self.step_states: step_states.append(step_state.serialize(id_encoder, serialization_options)) invocation_attrs["step_states"] = step_states input_step_parameters = [] for input_step_parameter in self.input_step_parameters: input_step_parameters.append(input_step_parameter.serialize(id_encoder, serialization_options)) invocation_attrs["input_step_parameters"] = input_step_parameters input_datasets = [] for input_dataset in self.input_datasets: input_datasets.append(input_dataset.serialize(id_encoder, serialization_options)) invocation_attrs["input_datasets"] = input_datasets input_dataset_collections = [] for input_dataset_collection in self.input_dataset_collections: input_dataset_collections.append(input_dataset_collection.serialize(id_encoder, serialization_options)) invocation_attrs["input_dataset_collections"] = input_dataset_collections output_dataset_collections = [] for output_dataset_collection in self.output_dataset_collections: output_dataset_collections.append(output_dataset_collection.serialize(id_encoder, serialization_options)) invocation_attrs["output_dataset_collections"] = output_dataset_collections output_datasets = [] for output_dataset in self.output_datasets: output_datasets.append(output_dataset.serialize(id_encoder, serialization_options)) invocation_attrs["output_datasets"] = output_datasets output_values = [] for output_value in self.output_values: output_values.append(output_value.serialize(id_encoder, serialization_options)) invocation_attrs["output_values"] = output_values serialization_options.attach_identifier(id_encoder, self, invocation_attrs) return invocation_attrs
[docs] def to_dict(self, view="collection", value_mapper=None, step_details=False, legacy_job_state=False): rval = super().to_dict(view=view, value_mapper=value_mapper) if rval["state"] is None: # bugs could result in no state being set rval["state"] = self.states.FAILED if view == "element": steps = [] for step in self.steps: if step_details: v = step.to_dict(view="element") else: v = step.to_dict(view="collection") if legacy_job_state: step_jobs = step.jobs if step_jobs: for step_job in step_jobs: v_clone = v.copy() v_clone["state"] = step_job.state v_clone["job_id"] = step_job.id steps.append(v_clone) else: v["state"] = None steps.append(v) else: v["implicit_collection_jobs_id"] = step.implicit_collection_jobs_id steps.append(v) rval["steps"] = steps inputs = {} for input_item_association in self.input_datasets + self.input_dataset_collections: if input_item_association.history_content_type == "dataset": src = "hda" item = input_item_association.dataset elif input_item_association.history_content_type == "dataset_collection": src = "hdca" item = input_item_association.dataset_collection else: # TODO: LDDAs are not implemented in workflow_request_to_input_dataset table raise Exception(f"Unknown history content type '{input_item_association.history_content_type}'") # Should this maybe also be by label ? Would break backwards compatibility though inputs[str(input_item_association.workflow_step.order_index)] = { "id": item.id, "src": src, "label": input_item_association.workflow_step.label, "workflow_step_id": input_item_association.workflow_step_id, } rval["inputs"] = inputs input_parameters = {} for input_step_parameter in self.input_step_parameters: label = input_step_parameter.workflow_step.label if not label: label = f"{input_step_parameter.workflow_step.order_index + 1}: Unnamed parameter" input_parameters[label] = { "parameter_value": input_step_parameter.parameter_value, "label": label, "workflow_step_id": input_step_parameter.workflow_step_id, } rval["input_step_parameters"] = input_parameters outputs = {} for output_assoc in self.output_datasets: # TODO: does this work correctly if outputs are mapped over? label = output_assoc.workflow_output.label if not label: label = f"{output_assoc.workflow_output.output_name} (Step {output_assoc.workflow_output.workflow_step.order_index + 1})" outputs[label] = { "src": "hda", "id": output_assoc.dataset_id, "workflow_step_id": output_assoc.workflow_step_id, } output_collections = {} for output_assoc in self.output_dataset_collections: label = output_assoc.workflow_output.label if not label: label = f"{output_assoc.workflow_output.output_name} (Step {output_assoc.workflow_output.workflow_step.order_index + 1})" output_collections[label] = { "src": "hdca", "id": output_assoc.dataset_collection_id, "workflow_step_id": output_assoc.workflow_step_id, } rval["outputs"] = outputs rval["output_collections"] = output_collections output_values = {} for output_param in self.output_values: label = output_param.workflow_output.label if not label: continue output_values[label] = output_param.value rval["output_values"] = output_values return rval
[docs] def add_input(self, content, step_id=None, step=None): assert step_id is not None or step is not None def attach_step(request_to_content): if step_id is not None: request_to_content.workflow_step_id = step_id else: request_to_content.workflow_step = step request: Optional[Dict[str, Any]] = None if isinstance(content, InputWithRequest): request = content.request content = content.input history_content_type = getattr(content, "history_content_type", None) if history_content_type == "dataset": request_to_content = WorkflowRequestToInputDatasetAssociation() request_to_content.dataset = content request_to_content.request = request attach_step(request_to_content) self.input_datasets.append(request_to_content) elif history_content_type == "dataset_collection": request_to_content = WorkflowRequestToInputDatasetCollectionAssociation() request_to_content.dataset_collection = content request_to_content.request = request attach_step(request_to_content) self.input_dataset_collections.append(request_to_content) else: request_to_content = WorkflowRequestInputStepParameter() request_to_content.parameter_value = content request_to_content.request = request attach_step(request_to_content) self.input_step_parameters.append(request_to_content)
[docs] def recover_inputs(self) -> Tuple[Dict[str, Any], str]: inputs = {} inputs_by = "name" have_referenced_steps_by_order_index = False def best_step_reference(workflow_step: "WorkflowStep") -> str: label = workflow_step.effective_label if label is not None: return label nonlocal have_referenced_steps_by_order_index have_referenced_steps_by_order_index = True return str(workflow_step.order_index) def ensure_step(step: Optional["WorkflowStep"]) -> "WorkflowStep": if step is None: raise galaxy.exceptions.InconsistentDatabase( "workflow input found without step definition, this should not happen" ) assert step return step for input_dataset_assoc in self.input_datasets: workflow_step = ensure_step(input_dataset_assoc.workflow_step) input_dataset = input_dataset_assoc.dataset input_index = best_step_reference(workflow_step) inputs[input_index] = input_dataset for input_dataset_collection_assoc in self.input_dataset_collections: workflow_step = ensure_step(input_dataset_collection_assoc.workflow_step) input_dataset_collection = input_dataset_collection_assoc.dataset_collection input_index = best_step_reference(workflow_step) inputs[input_index] = input_dataset_collection for input_step_parameter_assoc in self.input_step_parameters: workflow_step = ensure_step(input_step_parameter_assoc.workflow_step) value = input_step_parameter_assoc.parameter_value input_index = best_step_reference(workflow_step) inputs[input_index] = value if have_referenced_steps_by_order_index: inputs_by = "name|step_index" return inputs, inputs_by
[docs] def add_message(self, message: "InvocationMessageUnion"): self.messages.append( WorkflowInvocationMessage( # type:ignore[abstract] workflow_invocation_id=self.id, **message.dict( exclude_unset=True, exclude={ "history_id" }, # history_id comes in through workflow_invocation and isn't persisted in database ), ) )
@property def resource_parameters(self): resource_type = WorkflowRequestInputParameter.types.RESOURCE_PARAMETERS _resource_parameters = {} for input_parameter in self.input_parameters: if input_parameter.type == resource_type: _resource_parameters[input_parameter.name] = input_parameter.value return _resource_parameters
[docs] def has_input_for_step(self, step_id): for content in self.input_datasets: if content.workflow_step_id == step_id: return True for content in self.input_dataset_collections: if content.workflow_step_id == step_id: return True return False
[docs] def set_handler(self, handler): self.handler = handler
[docs] def log_str(self): extra = "" if (safe_id := getattr(self, "id", None)) is not None: extra += f"id={safe_id}" else: extra += "unflushed" return f"{self.__class__.__name__}[{extra}]"
[docs]class WorkflowInvocationToSubworkflowInvocationAssociation(Base, Dictifiable, RepresentById): __tablename__ = "workflow_invocation_to_subworkflow_invocation_association" id: Mapped[int] = mapped_column(primary_key=True) workflow_invocation_id: Mapped[Optional[int]] = mapped_column( ForeignKey("workflow_invocation.id", name="fk_wfi_swi_wfi"), index=True ) subworkflow_invocation_id: Mapped[Optional[int]] = mapped_column( ForeignKey("workflow_invocation.id", name="fk_wfi_swi_swi"), index=True ) workflow_step_id: Mapped[Optional[int]] = mapped_column(ForeignKey("workflow_step.id", name="fk_wfi_swi_ws")) subworkflow_invocation = relationship( "WorkflowInvocation", primaryjoin=( lambda: WorkflowInvocationToSubworkflowInvocationAssociation.subworkflow_invocation_id == WorkflowInvocation.id ), uselist=False, ) workflow_step = relationship("WorkflowStep") parent_workflow_invocation = relationship( "WorkflowInvocation", primaryjoin=( lambda: WorkflowInvocationToSubworkflowInvocationAssociation.workflow_invocation_id == WorkflowInvocation.id ), back_populates="subworkflow_invocations", uselist=False, ) dict_collection_visible_keys = ["id", "workflow_step_id", "workflow_invocation_id", "subworkflow_invocation_id"] dict_element_visible_keys = ["id", "workflow_step_id", "workflow_invocation_id", "subworkflow_invocation_id"]
[docs]class WorkflowInvocationMessage(Base, Dictifiable, Serializable): __tablename__ = "workflow_invocation_message" id: Mapped[int] = mapped_column(primary_key=True) workflow_invocation_id: Mapped[int] = mapped_column(ForeignKey("workflow_invocation.id"), index=True) reason: Mapped[Optional[str]] = mapped_column(String(32)) details: Mapped[Optional[str]] = mapped_column(TrimmedString(255)) output_name: Mapped[Optional[str]] = mapped_column(String(255)) workflow_step_id: Mapped[Optional[int]] = mapped_column(ForeignKey("workflow_step.id")) dependent_workflow_step_id: Mapped[Optional[int]] = mapped_column(ForeignKey("workflow_step.id")) job_id: Mapped[Optional[int]] = mapped_column(ForeignKey("job.id")) hda_id: Mapped[Optional[int]] = mapped_column(ForeignKey("history_dataset_association.id")) hdca_id: Mapped[Optional[int]] = mapped_column(ForeignKey("history_dataset_collection_association.id")) workflow_invocation: Mapped["WorkflowInvocation"] = relationship(back_populates="messages", lazy=True) workflow_step: Mapped[Optional["WorkflowStep"]] = relationship(foreign_keys=workflow_step_id, lazy=True) dependent_workflow_step: Mapped[Optional["WorkflowStep"]] = relationship( foreign_keys=dependent_workflow_step_id, lazy=True ) @property def workflow_step_index(self): return self.workflow_step and self.workflow_step.order_index @property def dependent_workflow_step_index(self): return self.dependent_workflow_step and self.dependent_workflow_step.order_index @property def history_id(self): return self.workflow_invocation.history_id
[docs]class EffectiveOutput(TypedDict): """An output for the sake or determining full workflow outputs. A workflow output might not be an effective output if it is an output on a subworkflow or a parent workflow that doesn't declare it an output. This is currently only used for determining object store selections. We don't want to capture subworkflow outputs that the user would like to ignore and discard as effective workflow outputs. """ output_name: str step_id: int
[docs]class WorkflowInvocationStepObjectStores(NamedTuple): preferred_object_store_id: Optional[str] preferred_outputs_object_store_id: Optional[str] preferred_intermediate_object_store_id: Optional[str] step_effective_outputs: Optional[List["EffectiveOutput"]]
[docs] def is_output_name_an_effective_output(self, output_name: str) -> bool: if self.step_effective_outputs is None: return True else: for effective_output in self.step_effective_outputs: if effective_output["output_name"] == output_name: return True return False
@property def is_split_configuration(self): preferred_outputs_object_store_id = self.preferred_outputs_object_store_id preferred_intermediate_object_store_id = self.preferred_intermediate_object_store_id has_typed_preferences = ( preferred_outputs_object_store_id is not None or preferred_intermediate_object_store_id is not None ) return has_typed_preferences and preferred_outputs_object_store_id != preferred_intermediate_object_store_id
[docs]class WorkflowInvocationStep(Base, Dictifiable, Serializable): __tablename__ = "workflow_invocation_step" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) workflow_invocation_id: Mapped[int] = mapped_column(ForeignKey("workflow_invocation.id"), index=True) workflow_step_id: Mapped[int] = mapped_column(ForeignKey("workflow_step.id"), index=True) state: Mapped[Optional[str]] = mapped_column(TrimmedString(64), index=True) job_id: Mapped[Optional[int]] = mapped_column(ForeignKey("job.id"), index=True) implicit_collection_jobs_id: Mapped[Optional[int]] = mapped_column( ForeignKey("implicit_collection_jobs.id"), index=True ) action: Mapped[Optional[bytes]] = mapped_column(MutableJSONType) workflow_step = relationship("WorkflowStep") job: Mapped[Optional["Job"]] = relationship(back_populates="workflow_invocation_step", uselist=False) implicit_collection_jobs = relationship("ImplicitCollectionJobs", uselist=False) output_dataset_collections = relationship( "WorkflowInvocationStepOutputDatasetCollectionAssociation", back_populates="workflow_invocation_step", cascade_backrefs=False, ) output_datasets = relationship( "WorkflowInvocationStepOutputDatasetAssociation", back_populates="workflow_invocation_step", cascade_backrefs=False, ) workflow_invocation: Mapped["WorkflowInvocation"] = relationship(back_populates="steps") output_value = relationship( "WorkflowInvocationOutputValue", foreign_keys="[WorkflowInvocationStep.workflow_invocation_id, WorkflowInvocationStep.workflow_step_id]", primaryjoin=( lambda: and_( WorkflowInvocationStep.workflow_invocation_id == WorkflowInvocationOutputValue.workflow_invocation_id, WorkflowInvocationStep.workflow_step_id == WorkflowInvocationOutputValue.workflow_step_id, ) ), back_populates="workflow_invocation_step", viewonly=True, ) order_index: Mapped[int] = column_property( select(WorkflowStep.order_index).where(WorkflowStep.id == workflow_step_id).scalar_subquery() ) dict_collection_visible_keys = [ "id", "update_time", "job_id", "workflow_step_id", "subworkflow_invocation_id", "state", "action", ] dict_element_visible_keys = [ "id", "update_time", "job_id", "workflow_step_id", "subworkflow_invocation_id", "state", "action", ] states = InvocationStepState @property def is_new(self): return self.state == self.states.NEW
[docs] def add_output(self, output_name, output_object): if output_object.history_content_type == "dataset": output_assoc = WorkflowInvocationStepOutputDatasetAssociation() output_assoc.workflow_invocation_step = self ensure_object_added_to_session(output_assoc, object_in_session=self) output_assoc.dataset = output_object output_assoc.output_name = output_name self.output_datasets.append(output_assoc) elif output_object.history_content_type == "dataset_collection": output_assoc = WorkflowInvocationStepOutputDatasetCollectionAssociation() output_assoc.workflow_invocation_step = self ensure_object_added_to_session(output_assoc, object_in_session=self) output_assoc.dataset_collection = output_object output_assoc.output_name = output_name self.output_dataset_collections.append(output_assoc) else: raise Exception("Unknown output type encountered")
@property def jobs(self): if self.job: return [self.job] elif self.implicit_collection_jobs: return self.implicit_collection_jobs.job_list else: return [] @property def preferred_object_stores(self) -> WorkflowInvocationStepObjectStores: meta_type = WorkflowRequestInputParameter.types.META_PARAMETERS preferred_object_store_id = None preferred_outputs_object_store_id = None preferred_intermediate_object_store_id = None step_effective_outputs: Optional[List[EffectiveOutput]] = None workflow_invocation = self.workflow_invocation for input_parameter in workflow_invocation.input_parameters: if input_parameter.type != meta_type: continue if input_parameter.name == "preferred_object_store_id": preferred_object_store_id = input_parameter.value elif input_parameter.name == "preferred_outputs_object_store_id": preferred_outputs_object_store_id = input_parameter.value elif input_parameter.name == "preferred_intermediate_object_store_id": preferred_intermediate_object_store_id = input_parameter.value elif input_parameter.name == "effective_outputs": all_effective_outputs = json.loads(input_parameter.value) step_id = self.workflow_step_id step_effective_outputs = [e for e in all_effective_outputs if e["step_id"] == step_id] return WorkflowInvocationStepObjectStores( preferred_object_store_id, preferred_outputs_object_store_id, preferred_intermediate_object_store_id, step_effective_outputs, ) def _serialize(self, id_encoder, serialization_options): step_attrs = dict_for(self) step_attrs["state"] = self.state step_attrs["create_time"] = self.create_time.__str__() step_attrs["update_time"] = self.update_time.__str__() step_attrs["order_index"] = self.workflow_step.order_index step_attrs["action"] = self.action if self.job: step_attrs["job"] = self.job.serialize(id_encoder, serialization_options, for_link=True) elif self.implicit_collection_jobs: step_attrs["implicit_collection_jobs"] = self.implicit_collection_jobs.serialize( id_encoder, serialization_options, for_link=True ) outputs = [] for output_dataset_assoc in self.output_datasets: output = dict( output_name=output_dataset_assoc.output_name, ) dataset = output_dataset_assoc.dataset if dataset: output["dataset"] = dataset.serialize(id_encoder, serialization_options, for_link=True) outputs.append(output) step_attrs["outputs"] = outputs output_collections = [] for output_dataset_collection_assoc in self.output_dataset_collections: output_collection = dict( output_name=output_dataset_collection_assoc.output_name, ) dataset_collection = output_dataset_collection_assoc.dataset_collection if dataset_collection: output_collection["dataset_collection"] = dataset_collection.serialize( id_encoder, serialization_options, for_link=True ) output_collections.append(output_collection) step_attrs["output_collections"] = output_collections return step_attrs
[docs] def to_dict(self, view="collection", value_mapper=None): rval = super().to_dict(view=view, value_mapper=value_mapper) rval["order_index"] = self.workflow_step.order_index rval["workflow_step_label"] = self.workflow_step.label rval["workflow_step_uuid"] = str(self.workflow_step.uuid) # Following no longer makes sense... # rval['state'] = self.job.state if self.job is not None else None if view == "element": jobs = [] for job in self.jobs: jobs.append(job.to_dict()) outputs = {} for output_assoc in self.output_datasets: name = output_assoc.output_name outputs[name] = { "src": "hda", "id": output_assoc.dataset.id, "uuid": ( str(output_assoc.dataset.dataset.uuid) if output_assoc.dataset.dataset.uuid is not None else None ), } output_collections = {} for output_assoc in self.output_dataset_collections: name = output_assoc.output_name output_collections[name] = { "src": "hdca", "id": output_assoc.dataset_collection.id, } rval["outputs"] = outputs rval["output_collections"] = output_collections rval["jobs"] = jobs return rval
[docs]class WorkflowRequestInputParameter(Base, Dictifiable, Serializable): """Workflow-related parameters not tied to steps or inputs.""" __tablename__ = "workflow_request_input_parameters" id: Mapped[int] = mapped_column(primary_key=True) workflow_invocation_id: Mapped[Optional[int]] = mapped_column( ForeignKey("workflow_invocation.id", onupdate="CASCADE", ondelete="CASCADE"), index=True ) name: Mapped[Optional[str]] = mapped_column(Unicode(255)) value: Mapped[Optional[str]] = mapped_column(TEXT) type: Mapped[Optional[str]] = mapped_column(Unicode(255)) workflow_invocation: Mapped[Optional["WorkflowInvocation"]] = relationship(back_populates="input_parameters") dict_collection_visible_keys = ["id", "name", "value", "type"]
[docs] class types(str, Enum): REPLACEMENT_PARAMETERS = "replacements" STEP_PARAMETERS = "step" META_PARAMETERS = "meta" RESOURCE_PARAMETERS = "resource"
def _serialize(self, id_encoder, serialization_options): request_input_parameter_attrs = dict_for(self) request_input_parameter_attrs["name"] = self.name request_input_parameter_attrs["value"] = self.value request_input_parameter_attrs["type"] = self.type return request_input_parameter_attrs
[docs]class WorkflowRequestStepState(Base, Dictifiable, Serializable): """Workflow step value parameters.""" __tablename__ = "workflow_request_step_states" id: Mapped[int] = mapped_column(primary_key=True) workflow_invocation_id: Mapped[Optional[int]] = mapped_column( ForeignKey("workflow_invocation.id", onupdate="CASCADE", ondelete="CASCADE"), index=True ) workflow_step_id: Mapped[Optional[int]] = mapped_column(ForeignKey("workflow_step.id")) value: Mapped[Optional[bytes]] = mapped_column(MutableJSONType) workflow_step: Mapped[Optional["WorkflowStep"]] = relationship() workflow_invocation: Mapped[Optional["WorkflowInvocation"]] = relationship(back_populates="step_states") dict_collection_visible_keys = ["id", "name", "value", "workflow_step_id"] def _serialize(self, id_encoder, serialization_options): request_step_state = dict_for(self) request_step_state["value"] = self.value request_step_state["order_index"] = self.workflow_step.order_index return request_step_state
[docs]class WorkflowRequestToInputDatasetAssociation(Base, Dictifiable, Serializable): """Workflow step input dataset parameters.""" __tablename__ = "workflow_request_to_input_dataset" id: Mapped[int] = mapped_column(primary_key=True) name: Mapped[Optional[str]] = mapped_column(String(255)) workflow_invocation_id: Mapped[Optional[int]] = mapped_column(ForeignKey("workflow_invocation.id"), index=True) workflow_step_id: Mapped[Optional[int]] = mapped_column(ForeignKey("workflow_step.id")) dataset_id: Mapped[Optional[int]] = mapped_column(ForeignKey("history_dataset_association.id"), index=True) request: Mapped[Optional[Dict]] = mapped_column(JSONType) workflow_step: Mapped[Optional["WorkflowStep"]] = relationship() dataset: Mapped[Optional["HistoryDatasetAssociation"]] = relationship() workflow_invocation: Mapped[Optional["WorkflowInvocation"]] = relationship(back_populates="input_datasets") history_content_type = "dataset" dict_collection_visible_keys = ["id", "workflow_invocation_id", "workflow_step_id", "dataset_id", "name"] def _serialize(self, id_encoder, serialization_options): request_input_dataset_attrs = dict_for(self) request_input_dataset_attrs["name"] = self.name request_input_dataset_attrs["dataset"] = self.dataset.serialize( id_encoder, serialization_options, for_link=True ) request_input_dataset_attrs["order_index"] = self.workflow_step.order_index return request_input_dataset_attrs
[docs]class WorkflowRequestToInputDatasetCollectionAssociation(Base, Dictifiable, Serializable): """Workflow step input dataset collection parameters.""" __tablename__ = "workflow_request_to_input_collection_dataset" id: Mapped[int] = mapped_column(primary_key=True) name: Mapped[Optional[str]] = mapped_column(String(255)) workflow_invocation_id: Mapped[Optional[int]] = mapped_column(ForeignKey("workflow_invocation.id"), index=True) workflow_step_id: Mapped[Optional[int]] = mapped_column(ForeignKey("workflow_step.id")) dataset_collection_id: Mapped[Optional[int]] = mapped_column( ForeignKey("history_dataset_collection_association.id"), index=True ) workflow_step: Mapped[Optional["WorkflowStep"]] = relationship() dataset_collection: Mapped[Optional["HistoryDatasetCollectionAssociation"]] = relationship() workflow_invocation: Mapped[Optional["WorkflowInvocation"]] = relationship( back_populates="input_dataset_collections" ) request: Mapped[Optional[Dict]] = mapped_column(JSONType) history_content_type = "dataset_collection" dict_collection_visible_keys = ["id", "workflow_invocation_id", "workflow_step_id", "dataset_collection_id", "name"] def _serialize(self, id_encoder, serialization_options): request_input_collection_attrs = dict_for(self) request_input_collection_attrs["name"] = self.name request_input_collection_attrs["dataset_collection"] = self.dataset_collection.serialize( id_encoder, serialization_options, for_link=True ) request_input_collection_attrs["order_index"] = self.workflow_step.order_index return request_input_collection_attrs
[docs]class WorkflowRequestInputStepParameter(Base, Dictifiable, Serializable): """Workflow step parameter inputs.""" __tablename__ = "workflow_request_input_step_parameter" id: Mapped[int] = mapped_column(primary_key=True) workflow_invocation_id: Mapped[Optional[int]] = mapped_column(ForeignKey("workflow_invocation.id"), index=True) workflow_step_id: Mapped[Optional[int]] = mapped_column(ForeignKey("workflow_step.id")) parameter_value: Mapped[Optional[bytes]] = mapped_column(MutableJSONType) request: Mapped[Optional[Dict]] = mapped_column(JSONType) workflow_step: Mapped[Optional["WorkflowStep"]] = relationship() workflow_invocation: Mapped[Optional["WorkflowInvocation"]] = relationship(back_populates="input_step_parameters") dict_collection_visible_keys = ["id", "workflow_invocation_id", "workflow_step_id", "parameter_value"] def _serialize(self, id_encoder, serialization_options): request_input_step_parameter_attrs = dict_for(self) request_input_step_parameter_attrs["parameter_value"] = self.parameter_value request_input_step_parameter_attrs["order_index"] = self.workflow_step.order_index return request_input_step_parameter_attrs
[docs]class WorkflowInvocationOutputDatasetAssociation(Base, Dictifiable, Serializable): """Represents links to output datasets for the workflow.""" __tablename__ = "workflow_invocation_output_dataset_association" id: Mapped[int] = mapped_column(primary_key=True) workflow_invocation_id: Mapped[Optional[int]] = mapped_column(ForeignKey("workflow_invocation.id"), index=True) workflow_step_id: Mapped[Optional[int]] = mapped_column(ForeignKey("workflow_step.id"), index=True) dataset_id: Mapped[Optional[int]] = mapped_column(ForeignKey("history_dataset_association.id"), index=True) workflow_output_id: Mapped[Optional[int]] = mapped_column(ForeignKey("workflow_output.id"), index=True) workflow_invocation: Mapped[Optional["WorkflowInvocation"]] = relationship(back_populates="output_datasets") workflow_step: Mapped[Optional["WorkflowStep"]] = relationship() dataset: Mapped[Optional["HistoryDatasetAssociation"]] = relationship() workflow_output: Mapped[Optional["WorkflowOutput"]] = relationship() history_content_type = "dataset" dict_collection_visible_keys = ["id", "workflow_invocation_id", "workflow_step_id", "dataset_id", "name"] def _serialize(self, id_encoder, serialization_options): output_dataset_attrs = dict_for(self) output_dataset_attrs["dataset"] = self.dataset.serialize(id_encoder, serialization_options, for_link=True) output_dataset_attrs["order_index"] = self.workflow_step.order_index output_dataset_attrs["workflow_output"] = self.workflow_output.serialize(id_encoder, serialization_options) return output_dataset_attrs
[docs]class WorkflowInvocationOutputDatasetCollectionAssociation(Base, Dictifiable, Serializable): """Represents links to output dataset collections for the workflow.""" __tablename__ = "workflow_invocation_output_dataset_collection_association" id: Mapped[int] = mapped_column(primary_key=True) workflow_invocation_id: Mapped[Optional[int]] = mapped_column( ForeignKey("workflow_invocation.id", name="fk_wiodca_wii"), index=True ) workflow_step_id: Mapped[Optional[int]] = mapped_column( ForeignKey("workflow_step.id", name="fk_wiodca_wsi"), index=True ) dataset_collection_id: Mapped[Optional[int]] = mapped_column( ForeignKey("history_dataset_collection_association.id", name="fk_wiodca_dci"), index=True ) workflow_output_id: Mapped[Optional[int]] = mapped_column( ForeignKey("workflow_output.id", name="fk_wiodca_woi"), index=True ) workflow_invocation: Mapped[Optional["WorkflowInvocation"]] = relationship( back_populates="output_dataset_collections" ) workflow_step: Mapped[Optional["WorkflowStep"]] = relationship() dataset_collection: Mapped[Optional["HistoryDatasetCollectionAssociation"]] = relationship() workflow_output: Mapped[Optional["WorkflowOutput"]] = relationship() history_content_type = "dataset_collection" dict_collection_visible_keys = ["id", "workflow_invocation_id", "workflow_step_id", "dataset_collection_id", "name"] def _serialize(self, id_encoder, serialization_options): output_collection_attrs = dict_for(self) output_collection_attrs["dataset_collection"] = self.dataset_collection.serialize( id_encoder, serialization_options, for_link=True ) output_collection_attrs["order_index"] = self.workflow_step.order_index output_collection_attrs["workflow_output"] = self.workflow_output.serialize(id_encoder, serialization_options) return output_collection_attrs
[docs]class WorkflowInvocationOutputValue(Base, Dictifiable, Serializable): """Represents a link to a specified or computed workflow parameter.""" __tablename__ = "workflow_invocation_output_value" id: Mapped[int] = mapped_column(primary_key=True) workflow_invocation_id: Mapped[Optional[int]] = mapped_column(ForeignKey("workflow_invocation.id"), index=True) workflow_step_id: Mapped[Optional[int]] = mapped_column(ForeignKey("workflow_step.id")) workflow_output_id: Mapped[Optional[int]] = mapped_column(ForeignKey("workflow_output.id"), index=True) value: Mapped[Optional[bytes]] = mapped_column(MutableJSONType) workflow_invocation: Mapped[Optional["WorkflowInvocation"]] = relationship(back_populates="output_values") workflow_invocation_step: Mapped[Optional["WorkflowInvocationStep"]] = relationship( foreign_keys="[WorkflowInvocationStep.workflow_invocation_id, WorkflowInvocationStep.workflow_step_id]", primaryjoin=( lambda: and_( WorkflowInvocationStep.workflow_invocation_id == WorkflowInvocationOutputValue.workflow_invocation_id, WorkflowInvocationStep.workflow_step_id == WorkflowInvocationOutputValue.workflow_step_id, ) ), back_populates="output_value", viewonly=True, ) workflow_step: Mapped[Optional["WorkflowStep"]] = relationship() workflow_output: Mapped[Optional["WorkflowOutput"]] = relationship() dict_collection_visible_keys = ["id", "workflow_invocation_id", "workflow_step_id", "value"] def _serialize(self, id_encoder, serialization_options): output_value_attrs = dict_for(self) output_value_attrs["value"] = self.value output_value_attrs["order_index"] = self.workflow_step.order_index output_value_attrs["workflow_output"] = self.workflow_output.serialize(id_encoder, serialization_options) return output_value_attrs
[docs]class WorkflowInvocationStepOutputDatasetAssociation(Base, Dictifiable, RepresentById): """Represents links to output datasets for the workflow.""" __tablename__ = "workflow_invocation_step_output_dataset_association" id: Mapped[int] = mapped_column(primary_key=True) workflow_invocation_step_id: Mapped[Optional[int]] = mapped_column( ForeignKey("workflow_invocation_step.id"), index=True ) dataset_id: Mapped[Optional[int]] = mapped_column(ForeignKey("history_dataset_association.id"), index=True) output_name: Mapped[Optional[str]] = mapped_column(String(255)) workflow_invocation_step: Mapped[Optional["WorkflowInvocationStep"]] = relationship( back_populates="output_datasets" ) dataset: Mapped[Optional["HistoryDatasetAssociation"]] = relationship() dict_collection_visible_keys = ["id", "workflow_invocation_step_id", "dataset_id", "output_name"]
[docs]class WorkflowInvocationStepOutputDatasetCollectionAssociation(Base, Dictifiable, RepresentById): """Represents links to output dataset collections for the workflow.""" __tablename__ = "workflow_invocation_step_output_dataset_collection_association" id: Mapped[int] = mapped_column(primary_key=True) workflow_invocation_step_id: Mapped[Optional[int]] = mapped_column( ForeignKey("workflow_invocation_step.id", name="fk_wisodca_wisi"), index=True ) workflow_step_id: Mapped[Optional[int]] = mapped_column( ForeignKey("workflow_step.id", name="fk_wisodca_wsi"), index=True ) dataset_collection_id: Mapped[Optional[int]] = mapped_column( ForeignKey("history_dataset_collection_association.id", name="fk_wisodca_dci"), index=True ) output_name: Mapped[Optional[str]] = mapped_column(String(255)) workflow_invocation_step: Mapped[Optional["WorkflowInvocationStep"]] = relationship( back_populates="output_dataset_collections" ) dataset_collection: Mapped[Optional["HistoryDatasetCollectionAssociation"]] = relationship() dict_collection_visible_keys = ["id", "workflow_invocation_step_id", "dataset_collection_id", "output_name"]
[docs]class MetadataFile(Base, StorableObject, Serializable): __tablename__ = "metadata_file" id: Mapped[int] = mapped_column(primary_key=True) name: Mapped[Optional[str]] = mapped_column(TEXT) hda_id: Mapped[Optional[int]] = mapped_column(ForeignKey("history_dataset_association.id"), index=True) lda_id: Mapped[Optional[int]] = mapped_column(ForeignKey("library_dataset_dataset_association.id"), index=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(index=True, default=now, onupdate=now, nullable=True) object_store_id: Mapped[Optional[str]] = mapped_column(TrimmedString(255), index=True) uuid: Mapped[Optional[Union[UUID, str]]] = mapped_column(UUIDType(), index=True) deleted: Mapped[Optional[bool]] = mapped_column(index=True, default=False) purged: Mapped[Optional[bool]] = mapped_column(index=True, default=False) history_dataset: Mapped[Optional["HistoryDatasetAssociation"]] = relationship() library_dataset: Mapped[Optional["LibraryDatasetDatasetAssociation"]] = relationship() def __init__(self, dataset=None, name=None, uuid=None): self.uuid = get_uuid(uuid) if isinstance(dataset, HistoryDatasetAssociation): self.history_dataset = dataset elif isinstance(dataset, LibraryDatasetDatasetAssociation): self.library_dataset = dataset self.name = name @property def dataset(self) -> Optional["DatasetInstance"]: da = self.history_dataset or self.library_dataset return da and da.dataset
[docs] def update_from_file(self, file_name): if not self.dataset: raise Exception("Attempted to write MetadataFile, but no DatasetAssociation set") if not self.dataset.purged: self.dataset.object_store.update_from_file( self, file_name=file_name, extra_dir="_metadata_files", extra_dir_at_root=True, alt_name=os.path.basename(self.get_file_name()), )
[docs] def get_file_name(self, sync_cache: bool = True) -> str: # Ensure the directory structure and the metadata file object exist try: da = self.history_dataset or self.library_dataset assert da is not None if self.object_store_id is None: self.object_store_id = da.dataset.object_store_id object_store = da.dataset.object_store store_by = object_store.get_store_by(da.dataset) if store_by == "id" and self.id is None: self.flush() # type:ignore[unreachable] identifier = getattr(self, store_by) alt_name = f"metadata_{identifier}.dat" if not object_store.exists(self, extra_dir="_metadata_files", extra_dir_at_root=True, alt_name=alt_name): object_store.create(self, extra_dir="_metadata_files", extra_dir_at_root=True, alt_name=alt_name) path = object_store.get_filename( self, extra_dir="_metadata_files", extra_dir_at_root=True, alt_name=alt_name, sync_cache=sync_cache ) return path except (AssertionError, AttributeError): assert ( self.id is not None ), "ID must be set before MetadataFile used without an HDA/LDDA (commit the object)" # In case we're not working with the history_dataset path = os.path.join(Dataset.file_path, "_metadata_files", *directory_hash_id(self.id)) # Create directory if it does not exist try: os.makedirs(path) except OSError as e: # File Exists is okay, otherwise reraise if e.errno != errno.EEXIST: raise # Return filename inside hashed directory return os.path.abspath(os.path.join(path, "metadata_%d.dat" % self.id))
def _serialize(self, id_encoder, serialization_options): as_dict = dict_for(self) serialization_options.attach_identifier(id_encoder, self, as_dict) as_dict["uuid"] = str(self.uuid or "") or None return as_dict
[docs]class FormDefinition(Base, Dictifiable, RepresentById): __tablename__ = "form_definition" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[Optional[datetime]] = mapped_column(default=now) update_time: Mapped[Optional[datetime]] = mapped_column(default=now, onupdate=now) name: Mapped[str] = mapped_column(TrimmedString(255)) desc: Mapped[Optional[str]] = mapped_column(TEXT) form_definition_current_id: Mapped[int] = mapped_column( ForeignKey("form_definition_current.id", use_alter=True), index=True ) fields: Mapped[Optional[bytes]] = mapped_column(MutableJSONType) type: Mapped[Optional[str]] = mapped_column(TrimmedString(255), index=True) layout: Mapped[Optional[bytes]] = mapped_column(MutableJSONType) form_definition_current: Mapped["FormDefinitionCurrent"] = relationship( back_populates="forms", primaryjoin=(lambda: FormDefinitionCurrent.id == FormDefinition.form_definition_current_id), ) # The following form_builder classes are supported by the FormDefinition class. supported_field_types = [ AddressField, CheckboxField, PasswordField, SelectField, TextArea, TextField, WorkflowField, WorkflowMappingField, HistoryField, ]
[docs] class types(str, Enum): USER_INFO = "User Information"
dict_collection_visible_keys = ["id", "name"] dict_element_visible_keys = ["id", "name", "desc", "form_definition_current_id", "fields", "layout"]
[docs] def populate(self, user=None, values=None, security=None): values = values or {} form_def = {"id": security.encode_id(self.id) if security else self.id, "name": self.name, "inputs": []} for field in self.fields: FieldClass = ( { "AddressField": AddressField, "CheckboxField": CheckboxField, "HistoryField": HistoryField, "PasswordField": PasswordField, "SelectField": SelectField, "TextArea": TextArea, "TextField": TextField, "WorkflowField": WorkflowField, } ).get(field["type"], TextField) form_def["inputs"].append( FieldClass( user=user, value=values.get(field["name"], field.get("default")), security=security, **field ).to_dict() ) return form_def
[docs] def grid_fields(self, grid_index): # Returns a dictionary whose keys are integers corresponding to field positions # on the grid and whose values are the field. gridfields = {} for i, f in enumerate(self.fields): if str(f["layout"]) == str(grid_index): gridfields[i] = f return gridfields
[docs]class FormDefinitionCurrent(Base, RepresentById): __tablename__ = "form_definition_current" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) latest_form_id: Mapped[Optional[int]] = mapped_column(ForeignKey("form_definition.id"), index=True) deleted: Mapped[Optional[bool]] = mapped_column(index=True, default=False) forms: Mapped[List["FormDefinition"]] = relationship( back_populates="form_definition_current", cascade="all, delete-orphan", primaryjoin=(lambda: FormDefinitionCurrent.id == FormDefinition.form_definition_current_id), ) latest_form: Mapped[Optional["FormDefinition"]] = relationship( post_update=True, primaryjoin=(lambda: FormDefinitionCurrent.latest_form_id == FormDefinition.id), ) def __init__(self, form_definition=None): self.latest_form = form_definition
[docs]class FormValues(Base, RepresentById): __tablename__ = "form_values" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) form_definition_id: Mapped[Optional[int]] = mapped_column(ForeignKey("form_definition.id"), index=True) content: Mapped[Optional[bytes]] = mapped_column(MutableJSONType) form_definition: Mapped[Optional["FormDefinition"]] = relationship( primaryjoin=(lambda: FormValues.form_definition_id == FormDefinition.id) ) def __init__(self, form_def=None, content=None): self.form_definition = form_def self.content = content
[docs]class UserAddress(Base, RepresentById): __tablename__ = "user_address" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) desc: Mapped[Optional[str]] = mapped_column(TrimmedString(255)) name: Mapped[str] = mapped_column(TrimmedString(255)) institution: Mapped[Optional[str]] = mapped_column(TrimmedString(255)) address: Mapped[str] = mapped_column(TrimmedString(255)) city: Mapped[str] = mapped_column(TrimmedString(255)) state: Mapped[str] = mapped_column(TrimmedString(255)) postal_code: Mapped[str] = mapped_column(TrimmedString(255)) country: Mapped[str] = mapped_column(TrimmedString(255)) phone: Mapped[Optional[str]] = mapped_column(TrimmedString(255)) deleted: Mapped[Optional[bool]] = mapped_column(index=True, default=False) purged: Mapped[Optional[bool]] = mapped_column(index=True, default=False) # `desc` needs to be fully qualified because it is shadowed by `desc` Column defined above # TODO: db migration to rename column, then use `desc` user: Mapped[Optional["User"]] = relationship(back_populates="addresses", order_by=sqlalchemy.desc("update_time"))
[docs] def to_dict(self, trans): return { "id": trans.security.encode_id(self.id), "name": sanitize_html(self.name), "desc": sanitize_html(self.desc), "institution": sanitize_html(self.institution), "address": sanitize_html(self.address), "city": sanitize_html(self.city), "state": sanitize_html(self.state), "postal_code": sanitize_html(self.postal_code), "country": sanitize_html(self.country), "phone": sanitize_html(self.phone), }
[docs]class PSAAssociation(Base, AssociationMixin, RepresentById): __tablename__ = "psa_association" id: Mapped[int] = mapped_column(primary_key=True) server_url: Mapped[Optional[str]] = mapped_column(VARCHAR(255)) handle: Mapped[Optional[str]] = mapped_column(VARCHAR(255)) secret: Mapped[Optional[str]] = mapped_column(VARCHAR(255)) issued: Mapped[Optional[int]] lifetime: Mapped[Optional[int]] assoc_type: Mapped[Optional[str]] = mapped_column(VARCHAR(64)) # This static property is set at: galaxy.authnz.psa_authnz.PSAAuthnz sa_session = None
[docs] def save(self): self.sa_session.add(self) with transaction(self.sa_session): self.sa_session.commit()
[docs] @classmethod def store(cls, server_url, association): """ Create an Association instance (Required by social_core.storage.AssociationMixin interface) """ def get_or_create(): stmt = select(PSAAssociation).filter_by(server_url=server_url, handle=association.handle).limit(1) assoc = cls.sa_session.scalars(stmt).first() return assoc or cls(server_url=server_url, handle=association.handle) assoc = get_or_create() assoc.secret = base64.encodebytes(association.secret).decode() assoc.issued = association.issued assoc.lifetime = association.lifetime assoc.assoc_type = association.assoc_type cls.sa_session.add(assoc) with transaction(cls.sa_session): cls.sa_session.commit()
[docs] @classmethod def get(cls, *args, **kwargs): """ Get an Association instance (Required by social_core.storage.AssociationMixin interface) """ stmt = select(PSAAssociation).filter_by(*args, **kwargs) return cls.sa_session.scalars(stmt).all()
[docs] @classmethod def remove(cls, ids_to_delete): """ Remove an Association instance (Required by social_core.storage.AssociationMixin interface) """ stmt = ( delete(PSAAssociation) .where(PSAAssociation.id.in_(ids_to_delete)) .execution_options(synchronize_session="fetch") ) PSAAssociation.sa_session.execute(stmt)
[docs]class PSACode(Base, CodeMixin, RepresentById): __tablename__ = "psa_code" __table_args__ = (UniqueConstraint("code", "email"),) id: Mapped[int] = mapped_column(primary_key=True) email: Mapped[Optional[str]] = mapped_column(VARCHAR(200)) code: Mapped[Optional[str]] = mapped_column(VARCHAR(32)) # This static property is set at: galaxy.authnz.psa_authnz.PSAAuthnz sa_session = None def __init__(self, email, code): self.email = email self.code = code
[docs] def save(self): self.sa_session.add(self) with transaction(self.sa_session): self.sa_session.commit()
[docs] @classmethod def get_code(cls, code): """ (Required by social_core.storage.CodeMixin interface) """ stmt = select(PSACode).where(PSACode.code == code).limit(1) return cls.sa_session.scalars(stmt).first()
[docs]class PSANonce(Base, NonceMixin, RepresentById): __tablename__ = "psa_nonce" id: Mapped[int] = mapped_column(primary_key=True) server_url: Mapped[Optional[str]] = mapped_column(VARCHAR(255)) timestamp: Mapped[Optional[int]] salt: Mapped[Optional[str]] = mapped_column(VARCHAR(40)) # This static property is set at: galaxy.authnz.psa_authnz.PSAAuthnz sa_session = None def __init__(self, server_url, timestamp, salt): self.server_url = server_url self.timestamp = timestamp self.salt = salt
[docs] def save(self): self.sa_session.add(self) with transaction(self.sa_session): self.sa_session.commit()
[docs] @classmethod def use(cls, server_url, timestamp, salt): """ Create a Nonce instance (Required by social_core.storage.NonceMixin interface) """ try: stmt = select(PSANonce).where(server_url=server_url, timestamp=timestamp, salt=salt).limit(1) return cls.sa_session.scalars(stmt).first() except IndexError: instance = cls(server_url=server_url, timestamp=timestamp, salt=salt) cls.sa_session.add(instance) with transaction(cls.sa_session): cls.sa_session.commit() return instance
[docs]class PSAPartial(Base, PartialMixin, RepresentById): __tablename__ = "psa_partial" id: Mapped[int] = mapped_column(primary_key=True) token: Mapped[Optional[str]] = mapped_column(VARCHAR(32)) data: Mapped[Optional[str]] = mapped_column(TEXT) next_step: Mapped[Optional[int]] backend: Mapped[Optional[str]] = mapped_column(VARCHAR(32)) # This static property is set at: galaxy.authnz.psa_authnz.PSAAuthnz sa_session = None def __init__(self, token, data, next_step, backend): self.token = token self.data = data self.next_step = next_step self.backend = backend
[docs] def save(self): self.sa_session.add(self) with transaction(self.sa_session): self.sa_session.commit()
[docs] @classmethod def load(cls, token): """ (Required by social_core.storage.PartialMixin interface) """ stmt = select(PSAPartial).where(PSAPartial.token == token).limit(1) return cls.sa_session.scalars(stmt).first()
[docs] @classmethod def destroy(cls, token): """ (Required by social_core.storage.PartialMixin interface) """ if partial := cls.load(token): session = cls.sa_session session.execute(delete(partial)) with transaction(session): session.commit()
[docs]class UserAuthnzToken(Base, UserMixin, RepresentById): __tablename__ = "oidc_user_authnz_tokens" __table_args__ = (UniqueConstraint("provider", "uid"),) id: Mapped[int] = mapped_column(primary_key=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) uid: Mapped[Optional[str]] = mapped_column(VARCHAR(255)) provider: Mapped[Optional[str]] = mapped_column(VARCHAR(32)) extra_data: Mapped[Optional[bytes]] = mapped_column(MutableJSONType) lifetime: Mapped[Optional[int]] assoc_type: Mapped[Optional[str]] = mapped_column(VARCHAR(64)) user: Mapped[Optional["User"]] = relationship(back_populates="social_auth") # This static property is set at: galaxy.authnz.psa_authnz.PSAAuthnz sa_session = None def __init__(self, provider, uid, extra_data=None, lifetime=None, assoc_type=None, user=None): self.provider = provider self.uid = uid self.user_id = user.id self.extra_data = extra_data self.lifetime = lifetime self.assoc_type = assoc_type
[docs] def get_id_token(self, strategy): if self.access_token_expired(): # Access and ID tokens have same expiration time; # hence, if one is expired, the other is expired too. self.refresh_token(strategy) return self.extra_data.get("id_token", None) if self.extra_data is not None else None
[docs] def set_extra_data(self, extra_data=None): if super().set_extra_data(extra_data): self.sa_session.add(self) with transaction(self.sa_session): self.sa_session.commit()
[docs] def save(self): self.sa_session.add(self) with transaction(self.sa_session): self.sa_session.commit()
[docs] @classmethod def changed(cls, user): """ The given user instance is ready to be saved (Required by social_core.storage.UserMixin interface) """ cls.sa_session.add(user) with transaction(cls.sa_session): cls.sa_session.commit()
[docs] @classmethod def get_username(cls, user): """ Return the username for given user (Required by social_core.storage.UserMixin interface) """ return getattr(user, "username", None)
[docs] @classmethod def user_model(cls): """ Return the user model (Required by social_core.storage.UserMixin interface) """ return User
[docs] @classmethod def username_max_length(cls): """ Return the max length for username (Required by social_core.storage.UserMixin interface) """ # Note: This is the maximum field length set for the username column of the galaxy_user table. # A better alternative is to retrieve this number from the table, instead of this const value. return 255
[docs] @classmethod def user_exists(cls, *args, **kwargs): """ Return True/False if a User instance exists with the given arguments. Arguments are directly passed to filter() manager method. (Required by social_core.storage.UserMixin interface) """ stmt_user = select(User).filter_by(*args, **kwargs) stmt_count = select(func.count()).select_from(stmt_user) return cls.sa_session.scalar(stmt_count) > 0
[docs] @classmethod def create_user(cls, *args, **kwargs): """ This is used by PSA authnz, do not use directly. Prefer using the user manager. (Required by social_core.storage.UserMixin interface) """ model = cls.user_model() instance = model(*args, **kwargs) if cls.get_users_by_email(instance.email): raise Exception(f"User with this email '{instance.email}' already exists.") instance.set_random_password() cls.sa_session.add(instance) with transaction(cls.sa_session): cls.sa_session.commit() return instance
[docs] @classmethod def get_user(cls, pk): """ Return user instance for given id (Required by social_core.storage.UserMixin interface) """ return cls.sa_session.get(User, pk)
[docs] @classmethod def get_users_by_email(cls, email): """ Return users instances for given email address (Required by social_core.storage.UserMixin interface) """ stmt = select(User).where(func.lower(User.email) == email.lower()) return cls.sa_session.scalars(stmt).all()
[docs] @classmethod def get_social_auth(cls, provider, uid): """ Return UserSocialAuth for given provider and uid (Required by social_core.storage.UserMixin interface) """ uid = str(uid) stmt = select(cls).filter_by(provider=provider, uid=uid).limit(1) return cls.sa_session.scalars(stmt).first()
[docs] @classmethod def get_social_auth_for_user(cls, user, provider=None, id=None): """ Return all the UserSocialAuth instances for given user (Required by social_core.storage.UserMixin interface) """ stmt = select(cls).filter_by(user_id=user.id) if provider: stmt = stmt.filter_by(provider=provider) if id: stmt = stmt.filter_by(id=id) return cls.sa_session.scalars(stmt).all()
[docs] @classmethod def create_social_auth(cls, user, uid, provider): """ Create a UserSocialAuth instance for given user (Required by social_core.storage.UserMixin interface) """ uid = str(uid) instance = cls(user=user, uid=uid, provider=provider) cls.sa_session.add(instance) with transaction(cls.sa_session): cls.sa_session.commit() return instance
[docs]class CustosAuthnzToken(Base, RepresentById): __tablename__ = "custos_authnz_token" __table_args__ = ( UniqueConstraint("user_id", "external_user_id", "provider"), UniqueConstraint("external_user_id", "provider"), ) id: Mapped[int] = mapped_column(primary_key=True) user_id: Mapped[int] = mapped_column(ForeignKey("galaxy_user.id"), nullable=True) external_user_id: Mapped[Optional[str]] = mapped_column(String(255)) provider: Mapped[Optional[str]] = mapped_column(String(255)) access_token: Mapped[Optional[str]] = mapped_column(Text) id_token: Mapped[Optional[str]] = mapped_column(Text) refresh_token: Mapped[Optional[str]] = mapped_column(Text) expiration_time: Mapped[datetime] = mapped_column(nullable=True) refresh_expiration_time: Mapped[datetime] = mapped_column(nullable=True) user: Mapped["User"] = relationship("User", back_populates="custos_auth")
[docs]class Page(Base, HasTags, Dictifiable, RepresentById, UsesCreateAndUpdateTime): __tablename__ = "page" __table_args__ = (Index("ix_page_slug", "slug", mysql_length=200),) id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) user_id: Mapped[int] = mapped_column(ForeignKey("galaxy_user.id"), index=True) latest_revision_id: Mapped[Optional[int]] = mapped_column( ForeignKey("page_revision.id", use_alter=True, name="page_latest_revision_id_fk"), index=True ) title: Mapped[Optional[str]] = mapped_column(TEXT) deleted: Mapped[Optional[bool]] = mapped_column(index=True, default=False) importable: Mapped[Optional[bool]] = mapped_column(index=True, default=False) slug: Mapped[Optional[str]] = mapped_column(TEXT) published: Mapped[Optional[bool]] = mapped_column(index=True, default=False) user: Mapped["User"] = relationship() revisions: Mapped[List["PageRevision"]] = relationship( cascade="all, delete-orphan", primaryjoin=(lambda: Page.id == PageRevision.page_id), back_populates="page", ) latest_revision: Mapped[Optional["PageRevision"]] = relationship( post_update=True, primaryjoin=(lambda: Page.latest_revision_id == PageRevision.id), lazy=False, ) tags: Mapped[List["PageTagAssociation"]] = relationship( order_by=lambda: PageTagAssociation.id, back_populates="page" ) annotations: Mapped[List["PageAnnotationAssociation"]] = relationship( order_by=lambda: PageAnnotationAssociation.id, back_populates="page" ) ratings: Mapped[List["PageRatingAssociation"]] = relationship( order_by=lambda: PageRatingAssociation.id, back_populates="page", ) users_shared_with: Mapped[List["PageUserShareAssociation"]] = relationship(back_populates="page") # Set up proxy so that # Page.users_shared_with # returns a list of users that page is shared with. users_shared_with_dot_users = association_proxy("users_shared_with", "user") dict_element_visible_keys = [ "id", "title", "latest_revision_id", "slug", "published", "importable", "deleted", "username", "email_hash", "create_time", "update_time", ]
[docs] def to_dict(self, view="element"): rval = super().to_dict(view=view) if "importable" in rval and rval["importable"] is None: # pages created prior to 2011 might not have importable field # probably not worth creating a migration to fix that rval["importable"] = False rev = [] for a in self.revisions: rev.append(a.id) rval["revision_ids"] = rev return rval
# username needed for slug generation @property def username(self): return self.user.username # email needed for hash generation @property def email_hash(self): return md5_hash_str(self.user.email)
[docs]class PageRevision(Base, Dictifiable, RepresentById): __tablename__ = "page_revision" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) page_id: Mapped[int] = mapped_column(ForeignKey("page.id"), index=True) title: Mapped[Optional[str]] = mapped_column(TEXT) content: Mapped[Optional[str]] = mapped_column(TEXT) content_format: Mapped[Optional[str]] = mapped_column(TrimmedString(32)) page: Mapped["Page"] = relationship(primaryjoin=(lambda: Page.id == PageRevision.page_id)) DEFAULT_CONTENT_FORMAT = "html" dict_element_visible_keys = ["id", "page_id", "title", "content", "content_format"] def __init__(self): self.content_format = PageRevision.DEFAULT_CONTENT_FORMAT
[docs] def to_dict(self, view="element"): rval = super().to_dict(view=view) rval["create_time"] = self.create_time.isoformat() rval["update_time"] = self.update_time.isoformat() return rval
[docs]class PageUserShareAssociation(Base, UserShareAssociation): __tablename__ = "page_user_share_association" id: Mapped[int] = mapped_column(primary_key=True) page_id: Mapped[int] = mapped_column(ForeignKey("page.id"), index=True, nullable=True) user_id: Mapped[int] = mapped_column(ForeignKey("galaxy_user.id"), index=True, nullable=True) user: Mapped[User] = relationship() page: Mapped["Page"] = relationship(back_populates="users_shared_with")
[docs]class Visualization(Base, HasTags, Dictifiable, RepresentById, UsesCreateAndUpdateTime): __tablename__ = "visualization" __table_args__ = ( Index("ix_visualization_dbkey", "dbkey", mysql_length=200), Index("ix_visualization_slug", "slug", mysql_length=200), ) id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) user_id: Mapped[int] = mapped_column(ForeignKey("galaxy_user.id"), index=True) latest_revision_id: Mapped[Optional[int]] = mapped_column( ForeignKey("visualization_revision.id", use_alter=True, name="visualization_latest_revision_id_fk"), index=True, ) title: Mapped[Optional[str]] = mapped_column(TEXT) type: Mapped[Optional[str]] = mapped_column(TEXT) dbkey: Mapped[Optional[str]] = mapped_column(TEXT) deleted: Mapped[Optional[bool]] = mapped_column(default=False, index=True) importable: Mapped[Optional[bool]] = mapped_column(default=False, index=True) slug: Mapped[Optional[str]] = mapped_column(TEXT) published: Mapped[Optional[bool]] = mapped_column(default=False, index=True) user: Mapped["User"] = relationship() revisions: Mapped[List["VisualizationRevision"]] = relationship( back_populates="visualization", cascade="all, delete-orphan", primaryjoin=(lambda: Visualization.id == VisualizationRevision.visualization_id), cascade_backrefs=False, ) latest_revision: Mapped[Optional["VisualizationRevision"]] = relationship( post_update=True, primaryjoin=(lambda: Visualization.latest_revision_id == VisualizationRevision.id), lazy=False, ) tags: Mapped[List["VisualizationTagAssociation"]] = relationship( order_by=lambda: VisualizationTagAssociation.id, back_populates="visualization" ) annotations: Mapped[List["VisualizationAnnotationAssociation"]] = relationship( order_by=lambda: VisualizationAnnotationAssociation.id, back_populates="visualization", ) ratings: Mapped[List["VisualizationRatingAssociation"]] = relationship( order_by=lambda: VisualizationRatingAssociation.id, back_populates="visualization", ) users_shared_with: Mapped[List["VisualizationUserShareAssociation"]] = relationship(back_populates="visualization") average_rating = None # Set up proxy so that # Visualization.users_shared_with # returns a list of users that visualization is shared with. users_shared_with_dot_users = association_proxy("users_shared_with", "user") dict_element_visible_keys = [ "id", "annotation", "create_time", "db_key", "deleted", "importable", "published", "tags", "title", "type", "update_time", "username", ] def __init__(self, **kwd): super().__init__(**kwd) if self.latest_revision: self.revisions.append(self.latest_revision) @property def annotation(self): if len(self.annotations) == 1: return self.annotations[0].annotation return None
[docs] def copy(self, user=None, title=None): """ Provide copy of visualization with only its latest revision. """ # NOTE: a shallow copy is done: the config is copied as is but datasets # are not copied nor are the dataset ids changed. This means that the # user does not have a copy of the data in his/her history and the # user who owns the datasets may delete them, making them inaccessible # for the current user. # TODO: a deep copy option is needed. if not user: user = self.user if not title: title = self.title copy_viz = Visualization(user=user, type=self.type, title=title, dbkey=self.dbkey) copy_revision = self.latest_revision.copy(visualization=copy_viz) copy_viz.latest_revision = copy_revision return copy_viz
[docs] def to_dict(self, view="element"): rval = super().to_dict(view=view) return rval
# username needed for slug generation @property def username(self): return self.user.username
[docs]class VisualizationRevision(Base, RepresentById): __tablename__ = "visualization_revision" __table_args__ = (Index("ix_visualization_revision_dbkey", "dbkey", mysql_length=200),) id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True) visualization_id: Mapped[int] = mapped_column(ForeignKey("visualization.id"), index=True) title: Mapped[Optional[str]] = mapped_column(TEXT) dbkey: Mapped[Optional[str]] = mapped_column(TEXT) config: Mapped[Optional[bytes]] = mapped_column(MutableJSONType) visualization: Mapped["Visualization"] = relationship( back_populates="revisions", primaryjoin=(lambda: Visualization.id == VisualizationRevision.visualization_id), )
[docs] def copy(self, visualization=None): """ Returns a copy of this object. """ if not visualization: visualization = self.visualization return VisualizationRevision( visualization=visualization, title=self.title, dbkey=self.dbkey, config=self.config )
[docs]class VisualizationUserShareAssociation(Base, UserShareAssociation): __tablename__ = "visualization_user_share_association" id: Mapped[int] = mapped_column(primary_key=True) visualization_id: Mapped[int] = mapped_column(ForeignKey("visualization.id"), index=True, nullable=True) user_id: Mapped[int] = mapped_column(ForeignKey("galaxy_user.id"), index=True, nullable=True) user: Mapped[User] = relationship() visualization: Mapped["Visualization"] = relationship(back_populates="users_shared_with")
[docs]class Tag(Base, RepresentById): __tablename__ = "tag" __table_args__ = (UniqueConstraint("name"),) id: Mapped[int] = mapped_column(primary_key=True) type: Mapped[Optional[int]] parent_id: Mapped[Optional[int]] = mapped_column(ForeignKey("tag.id")) name: Mapped[Optional[str]] = mapped_column(TrimmedString(255)) children: Mapped[List["Tag"]] = relationship(back_populates="parent") parent: Mapped[Optional["Tag"]] = relationship(back_populates="children", remote_side=[id]) def __str__(self): return "Tag(id=%s, type=%i, parent_id=%s, name=%s)" % (self.id, self.type or -1, self.parent_id, self.name)
[docs]class ItemTagAssociation(Dictifiable): dict_collection_visible_keys = ["id", "user_tname", "user_value"] dict_element_visible_keys = dict_collection_visible_keys user_tname: Mapped[Optional[str]] user_value: Mapped[Optional[str]] = mapped_column(TrimmedString(255), index=True) def __init_subclass__(cls, **kwargs): super().__init_subclass__(**kwargs)
[docs] def copy(self, cls=None): if cls: new_ta = cls() else: new_ta = type(self)() new_ta.tag_id = self.tag_id new_ta.user_tname = self.user_tname new_ta.value = self.value new_ta.user_value = self.user_value return new_ta
[docs]class HistoryTagAssociation(Base, ItemTagAssociation, RepresentById): __tablename__ = "history_tag_association" id: Mapped[int] = mapped_column(primary_key=True) history_id: Mapped[int] = mapped_column(ForeignKey("history.id"), index=True, nullable=True) tag_id: Mapped[int] = mapped_column(ForeignKey("tag.id"), index=True, nullable=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) user_tname: Mapped[Optional[str]] = mapped_column(TrimmedString(255), index=True) value: Mapped[Optional[str]] = mapped_column(TrimmedString(255), index=True) history: Mapped["History"] = relationship(back_populates="tags") tag: Mapped["Tag"] = relationship() user: Mapped[Optional["User"]] = relationship()
[docs]class HistoryDatasetAssociationTagAssociation(Base, ItemTagAssociation, RepresentById): __tablename__ = "history_dataset_association_tag_association" id: Mapped[int] = mapped_column(primary_key=True) history_dataset_association_id: Mapped[int] = mapped_column( ForeignKey("history_dataset_association.id"), index=True, nullable=True ) tag_id: Mapped[int] = mapped_column(ForeignKey("tag.id"), index=True, nullable=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) user_tname: Mapped[Optional[str]] = mapped_column(TrimmedString(255), index=True) value: Mapped[Optional[str]] = mapped_column(TrimmedString(255), index=True) history_dataset_association: Mapped["HistoryDatasetAssociation"] = relationship(back_populates="tags") tag: Mapped["Tag"] = relationship() user: Mapped[Optional["User"]] = relationship()
[docs]class LibraryDatasetDatasetAssociationTagAssociation(Base, ItemTagAssociation, RepresentById): __tablename__ = "library_dataset_dataset_association_tag_association" id: Mapped[int] = mapped_column(primary_key=True) library_dataset_dataset_association_id: Mapped[int] = mapped_column( ForeignKey("library_dataset_dataset_association.id"), index=True, nullable=True ) tag_id: Mapped[int] = mapped_column(ForeignKey("tag.id"), index=True, nullable=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) user_tname: Mapped[Optional[str]] = mapped_column(TrimmedString(255), index=True) value: Mapped[Optional[str]] = mapped_column(TrimmedString(255), index=True) library_dataset_dataset_association: Mapped["LibraryDatasetDatasetAssociation"] = relationship( back_populates="tags" ) tag: Mapped["Tag"] = relationship() user: Mapped[Optional["User"]] = relationship()
[docs]class PageTagAssociation(Base, ItemTagAssociation, RepresentById): __tablename__ = "page_tag_association" id: Mapped[int] = mapped_column(primary_key=True) page_id: Mapped[int] = mapped_column(ForeignKey("page.id"), index=True, nullable=True) tag_id: Mapped[int] = mapped_column(ForeignKey("tag.id"), index=True, nullable=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) user_tname: Mapped[Optional[str]] = mapped_column(TrimmedString(255), index=True) value: Mapped[Optional[str]] = mapped_column(TrimmedString(255), index=True) page: Mapped["Page"] = relationship(back_populates="tags") tag: Mapped["Tag"] = relationship() user: Mapped[Optional["User"]] = relationship()
[docs]class WorkflowStepTagAssociation(Base, ItemTagAssociation, RepresentById): __tablename__ = "workflow_step_tag_association" id: Mapped[int] = mapped_column(primary_key=True) workflow_step_id: Mapped[int] = mapped_column(ForeignKey("workflow_step.id"), index=True, nullable=True) tag_id: Mapped[int] = mapped_column(ForeignKey("tag.id"), index=True, nullable=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) user_tname: Mapped[Optional[str]] = mapped_column(TrimmedString(255), index=True) value: Mapped[Optional[str]] = mapped_column(TrimmedString(255), index=True) workflow_step: Mapped["WorkflowStep"] = relationship(back_populates="tags") tag: Mapped["Tag"] = relationship() user: Mapped[Optional["User"]] = relationship()
[docs]class StoredWorkflowTagAssociation(Base, ItemTagAssociation, RepresentById): __tablename__ = "stored_workflow_tag_association" id: Mapped[int] = mapped_column(primary_key=True) stored_workflow_id: Mapped[int] = mapped_column(ForeignKey("stored_workflow.id"), index=True, nullable=True) tag_id: Mapped[int] = mapped_column(ForeignKey("tag.id"), index=True, nullable=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) user_tname: Mapped[Optional[str]] = mapped_column(TrimmedString(255), index=True) value: Mapped[Optional[str]] = mapped_column(TrimmedString(255), index=True) stored_workflow: Mapped["StoredWorkflow"] = relationship(back_populates="tags") tag: Mapped["Tag"] = relationship() user: Mapped[Optional["User"]] = relationship()
[docs]class VisualizationTagAssociation(Base, ItemTagAssociation, RepresentById): __tablename__ = "visualization_tag_association" id: Mapped[int] = mapped_column(primary_key=True) visualization_id: Mapped[int] = mapped_column(ForeignKey("visualization.id"), index=True, nullable=True) tag_id: Mapped[int] = mapped_column(ForeignKey("tag.id"), index=True, nullable=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) user_tname: Mapped[Optional[str]] = mapped_column(TrimmedString(255), index=True) value: Mapped[Optional[str]] = mapped_column(TrimmedString(255), index=True) visualization: Mapped["Visualization"] = relationship(back_populates="tags") tag: Mapped["Tag"] = relationship() user: Mapped[Optional["User"]] = relationship()
[docs]class HistoryDatasetCollectionTagAssociation(Base, ItemTagAssociation, RepresentById): __tablename__ = "history_dataset_collection_tag_association" id: Mapped[int] = mapped_column(primary_key=True) history_dataset_collection_id: Mapped[int] = mapped_column( ForeignKey("history_dataset_collection_association.id"), index=True, nullable=True ) tag_id: Mapped[int] = mapped_column(ForeignKey("tag.id"), index=True, nullable=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) user_tname: Mapped[Optional[str]] = mapped_column(TrimmedString(255), index=True) value: Mapped[Optional[str]] = mapped_column(TrimmedString(255), index=True) dataset_collection: Mapped["HistoryDatasetCollectionAssociation"] = relationship(back_populates="tags") tag: Mapped["Tag"] = relationship() user: Mapped[Optional["User"]] = relationship()
[docs]class LibraryDatasetCollectionTagAssociation(Base, ItemTagAssociation, RepresentById): __tablename__ = "library_dataset_collection_tag_association" id: Mapped[int] = mapped_column(primary_key=True) library_dataset_collection_id: Mapped[int] = mapped_column( ForeignKey("library_dataset_collection_association.id"), index=True, nullable=True ) tag_id: Mapped[int] = mapped_column(ForeignKey("tag.id"), index=True, nullable=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) user_tname: Mapped[Optional[str]] = mapped_column(TrimmedString(255), index=True) value: Mapped[Optional[str]] = mapped_column(TrimmedString(255), index=True) dataset_collection: Mapped["LibraryDatasetCollectionAssociation"] = relationship(back_populates="tags") tag: Mapped["Tag"] = relationship() user: Mapped[Optional["User"]] = relationship()
[docs]class ToolTagAssociation(Base, ItemTagAssociation, RepresentById): __tablename__ = "tool_tag_association" id: Mapped[int] = mapped_column(primary_key=True) tool_id: Mapped[str] = mapped_column(TrimmedString(255), index=True, nullable=True) tag_id: Mapped[int] = mapped_column(ForeignKey("tag.id"), index=True, nullable=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) user_tname: Mapped[Optional[str]] = mapped_column(TrimmedString(255), index=True) value: Mapped[Optional[str]] = mapped_column(TrimmedString(255), index=True) tag: Mapped["Tag"] = relationship() user: Mapped[Optional["User"]] = relationship()
# Item annotation classes.
[docs]class HistoryAnnotationAssociation(Base, RepresentById): __tablename__ = "history_annotation_association" __table_args__ = (Index("ix_history_anno_assoc_annotation", "annotation", mysql_length=200),) id: Mapped[int] = mapped_column(primary_key=True) history_id: Mapped[int] = mapped_column(ForeignKey("history.id"), index=True, nullable=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) annotation: Mapped[str] = mapped_column(TEXT, nullable=True) history: Mapped["History"] = relationship(back_populates="annotations") user: Mapped["User"] = relationship()
[docs]class HistoryDatasetAssociationAnnotationAssociation(Base, RepresentById): __tablename__ = "history_dataset_association_annotation_association" __table_args__ = (Index("ix_history_dataset_anno_assoc_annotation", "annotation", mysql_length=200),) id: Mapped[int] = mapped_column(primary_key=True) history_dataset_association_id: Mapped[int] = mapped_column( ForeignKey("history_dataset_association.id"), index=True, nullable=True ) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) annotation: Mapped[str] = mapped_column(TEXT, nullable=True) hda: Mapped["HistoryDatasetAssociation"] = relationship(back_populates="annotations") user: Mapped[Optional["User"]] = relationship()
[docs]class StoredWorkflowAnnotationAssociation(Base, RepresentById): __tablename__ = "stored_workflow_annotation_association" __table_args__ = (Index("ix_stored_workflow_ann_assoc_annotation", "annotation", mysql_length=200),) id: Mapped[int] = mapped_column(primary_key=True) stored_workflow_id: Mapped[int] = mapped_column(ForeignKey("stored_workflow.id"), index=True, nullable=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) annotation: Mapped[str] = mapped_column(TEXT, nullable=True) stored_workflow: Mapped["StoredWorkflow"] = relationship(back_populates="annotations") user: Mapped[Optional["User"]] = relationship()
[docs]class WorkflowStepAnnotationAssociation(Base, RepresentById): __tablename__ = "workflow_step_annotation_association" __table_args__ = (Index("ix_workflow_step_ann_assoc_annotation", "annotation", mysql_length=200),) id: Mapped[int] = mapped_column(primary_key=True) workflow_step_id: Mapped[int] = mapped_column(ForeignKey("workflow_step.id"), index=True, nullable=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) annotation: Mapped[str] = mapped_column(TEXT, nullable=True) workflow_step: Mapped["WorkflowStep"] = relationship(back_populates="annotations") user: Mapped[Optional["User"]] = relationship()
[docs]class PageAnnotationAssociation(Base, RepresentById): __tablename__ = "page_annotation_association" __table_args__ = (Index("ix_page_annotation_association_annotation", "annotation", mysql_length=200),) id: Mapped[int] = mapped_column(primary_key=True) page_id: Mapped[int] = mapped_column(ForeignKey("page.id"), index=True, nullable=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) annotation: Mapped[str] = mapped_column(TEXT, nullable=True) page: Mapped["Page"] = relationship(back_populates="annotations") user: Mapped[Optional["User"]] = relationship()
[docs]class VisualizationAnnotationAssociation(Base, RepresentById): __tablename__ = "visualization_annotation_association" __table_args__ = (Index("ix_visualization_annotation_association_annotation", "annotation", mysql_length=200),) id: Mapped[int] = mapped_column(primary_key=True) visualization_id: Mapped[int] = mapped_column(ForeignKey("visualization.id"), index=True, nullable=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) annotation: Mapped[str] = mapped_column(TEXT, nullable=True) visualization: Mapped["Visualization"] = relationship(back_populates="annotations") user: Mapped[Optional["User"]] = relationship()
[docs]class HistoryDatasetCollectionAssociationAnnotationAssociation(Base, RepresentById): __tablename__ = "history_dataset_collection_annotation_association" id: Mapped[int] = mapped_column(primary_key=True) history_dataset_collection_id: Mapped[int] = mapped_column( ForeignKey("history_dataset_collection_association.id"), index=True, nullable=True ) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) annotation: Mapped[str] = mapped_column(TEXT, nullable=True) history_dataset_collection: Mapped["HistoryDatasetCollectionAssociation"] = relationship( back_populates="annotations" ) user: Mapped[Optional["User"]] = relationship()
[docs]class LibraryDatasetCollectionAnnotationAssociation(Base, RepresentById): __tablename__ = "library_dataset_collection_annotation_association" id: Mapped[int] = mapped_column(primary_key=True) library_dataset_collection_id: Mapped[int] = mapped_column( ForeignKey("library_dataset_collection_association.id"), index=True, nullable=True ) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) annotation: Mapped[str] = mapped_column(TEXT, nullable=True) dataset_collection: Mapped["LibraryDatasetCollectionAssociation"] = relationship(back_populates="annotations") user: Mapped[Optional["User"]] = relationship()
[docs]class Vault(Base): __tablename__ = "vault" key: Mapped[str] = mapped_column(Text, primary_key=True) parent_key: Mapped[Optional[str]] = mapped_column(Text, ForeignKey(key), index=True) children: Mapped[List["Vault"]] = relationship(back_populates="parent") parent: Mapped[Optional["Vault"]] = relationship(back_populates="children", remote_side=[key]) value: Mapped[Optional[str]] = mapped_column(Text) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, nullable=True)
# Item rating classes.
[docs]class ItemRatingAssociation(Base): __abstract__ = True
[docs] def __init__(self, user, item, rating=0): self.user = user self.rating = rating self._set_item(item)
def _set_item(self, item): """Set association's item.""" raise NotImplementedError()
[docs]class HistoryRatingAssociation(ItemRatingAssociation, RepresentById): __tablename__ = "history_rating_association" id: Mapped[int] = mapped_column(primary_key=True) history_id: Mapped[int] = mapped_column(ForeignKey("history.id"), index=True, nullable=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) rating: Mapped[int] = mapped_column(index=True, nullable=True) history: Mapped["History"] = relationship(back_populates="ratings") user: Mapped[Optional["User"]] = relationship() def _set_item(self, history): add_object_to_object_session(self, history) self.history = history
[docs]class HistoryDatasetAssociationRatingAssociation(ItemRatingAssociation, RepresentById): __tablename__ = "history_dataset_association_rating_association" id: Mapped[int] = mapped_column(primary_key=True) history_dataset_association_id: Mapped[int] = mapped_column( ForeignKey("history_dataset_association.id"), index=True, nullable=True ) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) rating: Mapped[int] = mapped_column(index=True, nullable=True) history_dataset_association: Mapped["HistoryDatasetAssociation"] = relationship(back_populates="ratings") user: Mapped[Optional["User"]] = relationship() def _set_item(self, history_dataset_association): add_object_to_object_session(self, history_dataset_association) self.history_dataset_association = history_dataset_association
[docs]class StoredWorkflowRatingAssociation(ItemRatingAssociation, RepresentById): __tablename__ = "stored_workflow_rating_association" id: Mapped[int] = mapped_column(primary_key=True) stored_workflow_id: Mapped[int] = mapped_column(ForeignKey("stored_workflow.id"), index=True, nullable=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) rating: Mapped[int] = mapped_column(index=True, nullable=True) stored_workflow: Mapped["StoredWorkflow"] = relationship(back_populates="ratings") user: Mapped[Optional["User"]] = relationship() def _set_item(self, stored_workflow): add_object_to_object_session(self, stored_workflow) self.stored_workflow = stored_workflow
[docs]class PageRatingAssociation(ItemRatingAssociation, RepresentById): __tablename__ = "page_rating_association" id: Mapped[int] = mapped_column(primary_key=True) page_id: Mapped[int] = mapped_column(ForeignKey("page.id"), index=True, nullable=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) rating: Mapped[int] = mapped_column(index=True, nullable=True) page: Mapped["Page"] = relationship(back_populates="ratings") user: Mapped[Optional["User"]] = relationship() def _set_item(self, page): add_object_to_object_session(self, page) self.page = page
[docs]class VisualizationRatingAssociation(ItemRatingAssociation, RepresentById): __tablename__ = "visualization_rating_association" id: Mapped[int] = mapped_column(primary_key=True) visualization_id: Mapped[int] = mapped_column(ForeignKey("visualization.id"), index=True, nullable=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) rating: Mapped[int] = mapped_column(index=True, nullable=True) visualization: Mapped["Visualization"] = relationship(back_populates="ratings") user: Mapped[Optional["User"]] = relationship() def _set_item(self, visualization): add_object_to_object_session(self, visualization) self.visualization = visualization
[docs]class HistoryDatasetCollectionRatingAssociation(ItemRatingAssociation, RepresentById): __tablename__ = "history_dataset_collection_rating_association" id: Mapped[int] = mapped_column(primary_key=True) history_dataset_collection_id: Mapped[int] = mapped_column( ForeignKey("history_dataset_collection_association.id"), index=True, nullable=True ) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) rating: Mapped[int] = mapped_column(index=True, nullable=True) dataset_collection: Mapped["HistoryDatasetCollectionAssociation"] = relationship(back_populates="ratings") user: Mapped[Optional["User"]] = relationship() def _set_item(self, dataset_collection): add_object_to_object_session(self, dataset_collection) self.dataset_collection = dataset_collection
[docs]class LibraryDatasetCollectionRatingAssociation(ItemRatingAssociation, RepresentById): __tablename__ = "library_dataset_collection_rating_association" id: Mapped[int] = mapped_column(primary_key=True) library_dataset_collection_id: Mapped[int] = mapped_column( ForeignKey("library_dataset_collection_association.id"), index=True, nullable=True ) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) rating: Mapped[int] = mapped_column(index=True, nullable=True) dataset_collection: Mapped["LibraryDatasetCollectionAssociation"] = relationship(back_populates="ratings") user: Mapped[Optional["User"]] = relationship() def _set_item(self, dataset_collection): add_object_to_object_session(self, dataset_collection) self.dataset_collection = dataset_collection
# Data manager classes.
[docs]class DataManagerHistoryAssociation(Base, RepresentById): __tablename__ = "data_manager_history_association" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(index=True, default=now, onupdate=now, nullable=True) history_id: Mapped[Optional[int]] = mapped_column(ForeignKey("history.id"), index=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) history: Mapped[Optional["History"]] = relationship() user: Mapped[Optional["User"]] = relationship(back_populates="data_manager_histories")
[docs]class DataManagerJobAssociation(Base, RepresentById): __tablename__ = "data_manager_job_association" __table_args__ = (Index("ix_data_manager_job_association_data_manager_id", "data_manager_id", mysql_length=200),) id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[datetime] = mapped_column(index=True, default=now, onupdate=now, nullable=True) job_id: Mapped[Optional[int]] = mapped_column(ForeignKey("job.id"), index=True) data_manager_id: Mapped[Optional[str]] = mapped_column(TEXT) job: Mapped[Optional["Job"]] = relationship(back_populates="data_manager_association", uselist=False)
[docs]class UserPreference(Base, RepresentById): __tablename__ = "user_preference" id: Mapped[int] = mapped_column(primary_key=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) name: Mapped[Optional[str]] = mapped_column(Unicode(255), index=True) value: Mapped[Optional[str]] = mapped_column(Text) def __init__(self, name=None, value=None): # Do not remove this constructor: it is set as the creator for the User.preferences # AssociationProxy to which 2 args are passed. self.name = name self.value = value
[docs]class UsesTemplatesAppConfig(Protocol): # TODO: delete this config protocol def and uses in dev pass
[docs]class HasConfigSecrets(RepresentById): secret_config_type: str template_secrets: Mapped[Optional[CONFIGURATION_TEMPLATE_CONFIGURATION_SECRET_NAMES_TYPE]] uuid: Mapped[Union[UUID, str]] user: Mapped["User"]
[docs] @classmethod def vault_key_from_uuid(clazz, uuid: Union[str, UUID], secret: str, app_config: UsesTemplatesAppConfig) -> str: return f"{clazz.secret_config_type}/{str(get_uuid(uuid))}/{secret}"
[docs] def vault_id_prefix(self, app_config: UsesTemplatesAppConfig) -> str: id_str = str(self.uuid) user_vault_id_prefix = f"{self.secret_config_type}/{id_str}" return user_vault_id_prefix
[docs] def vault_key(self, secret: str, app_config: UsesTemplatesAppConfig) -> str: user_vault_id_prefix = self.vault_id_prefix(app_config) key = f"{user_vault_id_prefix}/{secret}" return key
[docs]class HasConfigEnvironment(RepresentById): template_definition: Mapped[Optional[CONFIGURATION_TEMPLATE_DEFINITION_TYPE]] @property def template_environment(self) -> TemplateEnvironment: definition = self.template_definition if not definition: return TemplateEnvironment.model_validate([]) environment = definition.get("environment") if not environment: return TemplateEnvironment.model_validate([]) else: return TemplateEnvironment.model_validate(environment)
T = TypeVar("T", bound=ConfigTemplate, covariant=True)
[docs]class HasConfigTemplate(HasConfigSecrets, HasConfigEnvironment, RepresentById, Generic[T]): name: Mapped[str] description: Mapped[Optional[str]] template_id: Mapped[str] template_version: Mapped[int] template_variables: Mapped[Optional[CONFIGURATION_TEMPLATE_CONFIGURATION_VARIABLES_TYPE]] hidden: Mapped[bool] active: Mapped[bool] purged: Mapped[bool] @property def template(self) -> T: raise NotImplementedError()
[docs]class UserObjectStore(Base, HasConfigTemplate): __tablename__ = "user_object_store" secret_config_type = "object_store_config" id: Mapped[int] = mapped_column(Integer, primary_key=True) user_id: Mapped[Optional[int]] = mapped_column(Integer, ForeignKey("galaxy_user.id"), index=True) uuid: Mapped[Union[UUID, str]] = mapped_column(UUIDType(), index=True) create_time: Mapped[datetime] = mapped_column(DateTime, default=now) update_time: Mapped[datetime] = mapped_column(DateTime, default=now, onupdate=now, index=True) # user specified name of the instance they've created name: Mapped[str] = mapped_column(String(255), index=True) # user specified description of the instance they've created description: Mapped[Optional[str]] = mapped_column(Text) # active but doesn't appear in user selection hidden: Mapped[bool] = mapped_column(default=False) # set to False to deactive the source active: Mapped[bool] = mapped_column(default=True) # a purged file source cannot be re-activated because secrets have been purged purged: Mapped[bool] = mapped_column(default=False) # the template store id template_id: Mapped[str] = mapped_column(String(255), index=True) # the template store version (0, 1, ...) template_version: Mapped[int] = mapped_column(Integer, index=True) # Full template from object_store_templates.yml catalog. # For tools we just store references, so here we could easily just use # the id/version and not record the definition... as the templates change # over time this choice has some big consequences despite being easy to swap # implementations. template_definition: Mapped[Optional[CONFIGURATION_TEMPLATE_DEFINITION_TYPE]] = mapped_column(JSONType) # Big JSON blob of the variable name -> value mapping defined for the store's # variables by the user. template_variables: Mapped[Optional[CONFIGURATION_TEMPLATE_CONFIGURATION_VARIABLES_TYPE]] = mapped_column(JSONType) # Track a list of secrets that were defined for this object store at creation template_secrets: Mapped[Optional[CONFIGURATION_TEMPLATE_CONFIGURATION_SECRET_NAMES_TYPE]] = mapped_column(JSONType) user: Mapped["User"] = relationship("User", back_populates="object_stores") @property def template(self) -> ObjectStoreTemplate: return ObjectStoreTemplate(**self.template_definition or {})
[docs] def object_store_configuration( self, secrets: SecretsDict, environment: EnvironmentDict, templates: Optional[List[ObjectStoreTemplate]] = None ) -> ObjectStoreConfiguration: if templates is None: templates = [self.template] user_details = self.user.config_template_details() variables: CONFIGURATION_TEMPLATE_CONFIGURATION_VARIABLES_TYPE = self.template_variables or {} first_exception = None for template in templates: try: return object_store_template_to_configuration( template, variables=variables, secrets=secrets, user_details=user_details, environment=environment, ) except Exception as e: if first_exception is None: first_exception = e continue if first_exception: raise first_exception raise ValueError("No template sent to object_store_configuration")
[docs]class UserFileSource(Base, HasConfigTemplate): __tablename__ = "user_file_source" secret_config_type = "file_source_config" id: Mapped[int] = mapped_column(primary_key=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) uuid: Mapped[Union[UUID, str]] = mapped_column(UUIDType(), index=True) create_time: Mapped[datetime] = mapped_column(default=now) update_time: Mapped[datetime] = mapped_column(default=now, onupdate=now, index=True) # user specified name of the instance they've created name: Mapped[str] = mapped_column(String(255), index=True) # user specified description of the instance they've created description: Mapped[Optional[str]] = mapped_column(Text) # active but doesn't appear in user selection hidden: Mapped[bool] = mapped_column(default=False) # set to False to deactive the source active: Mapped[bool] = mapped_column(default=True) # a purged file source cannot be re-activated because secrets have been purged purged: Mapped[bool] = mapped_column(default=False) # the template store id template_id: Mapped[str] = mapped_column(String(255), index=True) # the template store version (0, 1, ...) template_version: Mapped[int] = mapped_column(index=True) # Full template from file_sources_templates.yml catalog. # For tools we just store references, so here we could easily just use # the id/version and not record the definition... as the templates change # over time this choice has some big consequences despite being easy to swap # implementations. template_definition: Mapped[Optional[CONFIGURATION_TEMPLATE_DEFINITION_TYPE]] = mapped_column(JSONType) # Big JSON blob of the variable name -> value mapping defined for the store's # variables by the user. template_variables: Mapped[Optional[CONFIGURATION_TEMPLATE_CONFIGURATION_VARIABLES_TYPE]] = mapped_column(JSONType) # Track a list of secrets that were defined for this object store at creation template_secrets: Mapped[Optional[CONFIGURATION_TEMPLATE_CONFIGURATION_SECRET_NAMES_TYPE]] = mapped_column(JSONType) user: Mapped["User"] = relationship("User", back_populates="file_sources") @property def template(self) -> FileSourceTemplate: return FileSourceTemplate(**self.template_definition or {})
[docs] def file_source_configuration( self, secrets: SecretsDict, environment: EnvironmentDict, implicit: ImplicitConfigurationParameters, templates: Optional[List[FileSourceTemplate]] = None, ) -> FileSourceConfiguration: if templates is None: templates = [self.template] user_details = self.user.config_template_details() variables: CONFIGURATION_TEMPLATE_CONFIGURATION_VARIABLES_TYPE = self.template_variables or {} first_exception = None for template in templates: try: return file_source_template_to_configuration( template, variables=variables, secrets=secrets, user_details=user_details, environment=environment, implicit=implicit, ) except Exception as e: if first_exception is None: first_exception = e continue if first_exception: raise first_exception raise ValueError("No template sent to file_source_configuration")
# TODO: add link from tool_request to this
[docs]class ToolLandingRequest(Base): __tablename__ = "tool_landing_request" id: Mapped[int] = mapped_column(primary_key=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[Optional[datetime]] = mapped_column(index=True, default=now, onupdate=now, nullable=True) uuid: Mapped[Union[UUID, str]] = mapped_column(UUIDType(), index=True) tool_id: Mapped[str] = mapped_column(String(255)) tool_version: Mapped[Optional[str]] = mapped_column(String(255), default=None) request_state: Mapped[Optional[Dict]] = mapped_column(JSONType) client_secret: Mapped[Optional[str]] = mapped_column(String(255), default=None) public: Mapped[bool] = mapped_column(Boolean) user: Mapped[Optional["User"]] = relationship()
# TODO: add link from workflow_invocation to this
[docs]class WorkflowLandingRequest(Base): __tablename__ = "workflow_landing_request" id: Mapped[int] = mapped_column(primary_key=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) workflow_id: Mapped[Optional[int]] = mapped_column(ForeignKey("stored_workflow.id"), nullable=True) stored_workflow_id: Mapped[Optional[int]] = mapped_column(ForeignKey("workflow.id"), nullable=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) update_time: Mapped[Optional[datetime]] = mapped_column(index=True, default=now, onupdate=now, nullable=True) uuid: Mapped[Union[UUID, str]] = mapped_column(UUIDType(), index=True) request_state: Mapped[Optional[Dict]] = mapped_column(JSONType) client_secret: Mapped[Optional[str]] = mapped_column(String(255), default=None) workflow_source: Mapped[Optional[str]] = mapped_column(String(255), default=None) workflow_source_type: Mapped[Optional[str]] = mapped_column(String(255), default=None) public: Mapped[bool] = mapped_column(Boolean) user: Mapped[Optional["User"]] = relationship() stored_workflow: Mapped[Optional["StoredWorkflow"]] = relationship() workflow: Mapped[Optional["Workflow"]] = relationship()
[docs]class UserAction(Base, RepresentById): __tablename__ = "user_action" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) session_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_session.id"), index=True) action: Mapped[Optional[str]] = mapped_column(Unicode(255)) context: Mapped[Optional[str]] = mapped_column(Unicode(512)) params: Mapped[Optional[str]] = mapped_column(Unicode(1024)) user: Mapped[Optional["User"]] = relationship()
[docs]class APIKeys(Base, RepresentById): __tablename__ = "api_keys" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) user_id: Mapped[Optional[int]] = mapped_column(ForeignKey("galaxy_user.id"), index=True) key: Mapped[Optional[str]] = mapped_column(TrimmedString(32), index=True, unique=True) user: Mapped[Optional["User"]] = relationship(back_populates="api_keys") deleted: Mapped[bool] = mapped_column(index=True, server_default=false())
[docs]def copy_list(lst, *args, **kwds): if lst is None: return lst else: return [el.copy(*args, **kwds) for el in lst]
def _prepare_metadata_for_serialization(id_encoder, serialization_options, metadata): """Prepare metatdata for exporting.""" processed_metadata = {} for name, value in metadata.items(): # Metadata files are not needed for export because they can be # regenerated. if isinstance(value, MetadataFile): if serialization_options.strip_metadata_files: continue else: value = value.serialize(id_encoder, serialization_options) processed_metadata[name] = value return processed_metadata # The following CleanupEvent* models could be defined as tables only; # however making them models keeps things simple and consistent.
[docs]class CleanupEvent(Base): __tablename__ = "cleanup_event" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) message: Mapped[Optional[str]] = mapped_column(TrimmedString(1024))
[docs]class CleanupEventDatasetAssociation(Base): __tablename__ = "cleanup_event_dataset_association" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) cleanup_event_id: Mapped[int] = mapped_column(ForeignKey("cleanup_event.id"), index=True, nullable=True) dataset_id: Mapped[int] = mapped_column(ForeignKey("dataset.id"), index=True, nullable=True)
[docs]class CleanupEventMetadataFileAssociation(Base): __tablename__ = "cleanup_event_metadata_file_association" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) cleanup_event_id: Mapped[int] = mapped_column(ForeignKey("cleanup_event.id"), index=True, nullable=True) metadata_file_id: Mapped[int] = mapped_column(ForeignKey("metadata_file.id"), index=True, nullable=True)
[docs]class CleanupEventHistoryAssociation(Base): __tablename__ = "cleanup_event_history_association" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) cleanup_event_id: Mapped[int] = mapped_column(ForeignKey("cleanup_event.id"), index=True, nullable=True) history_id: Mapped[int] = mapped_column(ForeignKey("history.id"), index=True, nullable=True)
[docs]class CleanupEventHistoryDatasetAssociationAssociation(Base): __tablename__ = "cleanup_event_hda_association" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) cleanup_event_id: Mapped[int] = mapped_column(ForeignKey("cleanup_event.id"), index=True, nullable=True) hda_id: Mapped[int] = mapped_column(ForeignKey("history_dataset_association.id"), index=True, nullable=True)
[docs]class CleanupEventLibraryAssociation(Base): __tablename__ = "cleanup_event_library_association" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) cleanup_event_id: Mapped[int] = mapped_column(ForeignKey("cleanup_event.id"), index=True, nullable=True) library_id: Mapped[int] = mapped_column(ForeignKey("library.id"), index=True, nullable=True)
[docs]class CleanupEventLibraryFolderAssociation(Base): __tablename__ = "cleanup_event_library_folder_association" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) cleanup_event_id: Mapped[int] = mapped_column(ForeignKey("cleanup_event.id"), index=True, nullable=True) library_folder_id: Mapped[int] = mapped_column(ForeignKey("library_folder.id"), index=True, nullable=True)
[docs]class CleanupEventLibraryDatasetAssociation(Base): __tablename__ = "cleanup_event_library_dataset_association" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) cleanup_event_id: Mapped[int] = mapped_column(ForeignKey("cleanup_event.id"), index=True, nullable=True) library_dataset_id: Mapped[int] = mapped_column(ForeignKey("library_dataset.id"), index=True, nullable=True)
[docs]class CleanupEventLibraryDatasetDatasetAssociationAssociation(Base): __tablename__ = "cleanup_event_ldda_association" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) cleanup_event_id: Mapped[int] = mapped_column(ForeignKey("cleanup_event.id"), index=True, nullable=True) ldda_id: Mapped[int] = mapped_column( ForeignKey("library_dataset_dataset_association.id"), index=True, nullable=True )
[docs]class CleanupEventImplicitlyConvertedDatasetAssociationAssociation(Base): __tablename__ = "cleanup_event_icda_association" id: Mapped[int] = mapped_column(primary_key=True) create_time: Mapped[datetime] = mapped_column(default=now, nullable=True) cleanup_event_id: Mapped[int] = mapped_column(ForeignKey("cleanup_event.id"), index=True, nullable=True) icda_id: Mapped[int] = mapped_column( ForeignKey("implicitly_converted_dataset_association.id"), index=True, nullable=True )
[docs]class CeleryUserRateLimit(Base): """ For each user stores the last time a task was scheduled for execution. Used to limit the number of tasks allowed per user per second. """ __tablename__ = "celery_user_rate_limit" user_id: Mapped[int] = mapped_column(ForeignKey("galaxy_user.id", ondelete="CASCADE"), primary_key=True) last_scheduled_time: Mapped[datetime] def __repr__(self): return ( f"CeleryUserRateLimit(id_type={self.id_type!r}, " f"id={self.id!r}, last_scheduled_time={self.last_scheduled_time!r})" )
# The following models (HDA, LDDA) are mapped imperatively (for details see discussion in PR #12064) # TLDR: there are issues ('metadata' property, Galaxy object wrapping) that need to be addressed separately # before these models can be mapped declaratively. Keeping them in the mapping module breaks the auth package # tests (which import model directly bypassing the mapping module); fixing that is possible by importing # mapping into the test; however, having all models mapped in the same module is cleaner. HistoryDatasetAssociation.table = Table( "history_dataset_association", mapper_registry.metadata, Column("id", Integer, primary_key=True), Column("history_id", Integer, ForeignKey("history.id"), index=True), Column("dataset_id", Integer, ForeignKey("dataset.id"), index=True), Column("create_time", DateTime, default=now), Column("update_time", DateTime, default=now, onupdate=now, index=True), Column("state", TrimmedString(64), index=True, key="_state"), Column( "copied_from_history_dataset_association_id", Integer, ForeignKey("history_dataset_association.id"), nullable=True, ), Column( "copied_from_library_dataset_dataset_association_id", Integer, ForeignKey("library_dataset_dataset_association.id"), nullable=True, ), Column("name", TrimmedString(255)), Column("info", TrimmedString(255)), Column("blurb", TrimmedString(255)), Column("peek", TEXT, key="_peek"), Column("tool_version", TEXT), Column("extension", TrimmedString(64), index=True), Column("metadata", MetadataType, key="_metadata"), Column("metadata_deferred", Boolean, key="metadata_deferred"), Column("parent_id", Integer, ForeignKey("history_dataset_association.id"), nullable=True), Column("designation", TrimmedString(255)), Column("deleted", Boolean, index=True, default=False), Column("visible", Boolean), Column("extended_metadata_id", Integer, ForeignKey("extended_metadata.id"), index=True), Column("version", Integer, default=1, nullable=True, index=True), Column("hid", Integer), Column("purged", Boolean, index=True, default=False), Column("validated_state", TrimmedString(64), default="unvalidated", nullable=False), Column("validated_state_message", TEXT), Column( "hidden_beneath_collection_instance_id", ForeignKey("history_dataset_collection_association.id"), nullable=True ), ) LibraryDatasetDatasetAssociation.table = Table( "library_dataset_dataset_association", mapper_registry.metadata, Column("id", Integer, primary_key=True), Column("library_dataset_id", Integer, ForeignKey("library_dataset.id"), index=True), Column("dataset_id", Integer, ForeignKey("dataset.id"), index=True), Column("create_time", DateTime, default=now), Column("update_time", DateTime, default=now, onupdate=now, index=True), Column("state", TrimmedString(64), index=True, key="_state"), Column( "copied_from_history_dataset_association_id", Integer, ForeignKey( "history_dataset_association.id", use_alter=True, name="history_dataset_association_dataset_id_fkey" ), nullable=True, ), Column( "copied_from_library_dataset_dataset_association_id", Integer, ForeignKey( "library_dataset_dataset_association.id", use_alter=True, name="library_dataset_dataset_association_id_fkey" ), nullable=True, ), Column("name", TrimmedString(255), index=True), Column("info", TrimmedString(255)), Column("blurb", TrimmedString(255)), Column("peek", TEXT, key="_peek"), Column("tool_version", TEXT), Column("extension", TrimmedString(64)), Column("metadata", MetadataType, key="_metadata"), Column("metadata_deferred", Boolean, key="metadata_deferred"), Column("parent_id", Integer, ForeignKey("library_dataset_dataset_association.id"), nullable=True), Column("designation", TrimmedString(255)), Column("deleted", Boolean, index=True, default=False), Column("validated_state", TrimmedString(64), default="unvalidated", nullable=False), Column("validated_state_message", TEXT), Column("visible", Boolean), Column("extended_metadata_id", Integer, ForeignKey("extended_metadata.id"), index=True), Column("user_id", Integer, ForeignKey("galaxy_user.id"), index=True), Column("message", TrimmedString(255)), ) mapper_registry.map_imperatively( HistoryDatasetAssociation, HistoryDatasetAssociation.table, properties=dict( dataset=relationship( Dataset, primaryjoin=(lambda: Dataset.id == HistoryDatasetAssociation.table.c.dataset_id), lazy="joined", back_populates="history_associations", ), copied_from_history_dataset_association=relationship( HistoryDatasetAssociation, primaryjoin=( HistoryDatasetAssociation.table.c.copied_from_history_dataset_association_id == HistoryDatasetAssociation.table.c.id ), remote_side=[HistoryDatasetAssociation.table.c.id], uselist=False, back_populates="copied_to_history_dataset_associations", ), copied_from_library_dataset_dataset_association=relationship( LibraryDatasetDatasetAssociation, primaryjoin=( LibraryDatasetDatasetAssociation.table.c.id == HistoryDatasetAssociation.table.c.copied_from_library_dataset_dataset_association_id ), back_populates="copied_to_history_dataset_associations", ), copied_to_history_dataset_associations=relationship( HistoryDatasetAssociation, primaryjoin=( HistoryDatasetAssociation.table.c.copied_from_history_dataset_association_id == HistoryDatasetAssociation.table.c.id ), back_populates="copied_from_history_dataset_association", ), copied_to_library_dataset_dataset_associations=relationship( LibraryDatasetDatasetAssociation, primaryjoin=( HistoryDatasetAssociation.table.c.id == LibraryDatasetDatasetAssociation.table.c.copied_from_history_dataset_association_id ), back_populates="copied_from_history_dataset_association", ), tags=relationship( HistoryDatasetAssociationTagAssociation, order_by=HistoryDatasetAssociationTagAssociation.id, back_populates="history_dataset_association", ), annotations=relationship( HistoryDatasetAssociationAnnotationAssociation, order_by=HistoryDatasetAssociationAnnotationAssociation.id, back_populates="hda", ), ratings=relationship( HistoryDatasetAssociationRatingAssociation, order_by=HistoryDatasetAssociationRatingAssociation.id, back_populates="history_dataset_association", ), extended_metadata=relationship( ExtendedMetadata, primaryjoin=(HistoryDatasetAssociation.table.c.extended_metadata_id == ExtendedMetadata.id), ), hidden_beneath_collection_instance=relationship( HistoryDatasetCollectionAssociation, primaryjoin=( HistoryDatasetAssociation.table.c.hidden_beneath_collection_instance_id == HistoryDatasetCollectionAssociation.id ), uselist=False, ), _metadata=deferred(HistoryDatasetAssociation.table.c._metadata), dependent_jobs=relationship(JobToInputDatasetAssociation, back_populates="dataset"), creating_job_associations=relationship(JobToOutputDatasetAssociation, back_populates="dataset"), history=relationship(History, back_populates="datasets", cascade_backrefs=False), implicitly_converted_datasets=relationship( ImplicitlyConvertedDatasetAssociation, primaryjoin=(lambda: ImplicitlyConvertedDatasetAssociation.hda_parent_id == HistoryDatasetAssociation.id), back_populates="parent_hda", ), implicitly_converted_parent_datasets=relationship( ImplicitlyConvertedDatasetAssociation, primaryjoin=(lambda: ImplicitlyConvertedDatasetAssociation.hda_id == HistoryDatasetAssociation.id), back_populates="dataset", ), ), ) mapper_registry.map_imperatively( LibraryDatasetDatasetAssociation, LibraryDatasetDatasetAssociation.table, properties=dict( dataset=relationship( Dataset, primaryjoin=(lambda: LibraryDatasetDatasetAssociation.table.c.dataset_id == Dataset.id), back_populates="library_associations", ), library_dataset=relationship( LibraryDataset, foreign_keys=LibraryDatasetDatasetAssociation.table.c.library_dataset_id ), user=relationship(User), copied_from_library_dataset_dataset_association=relationship( LibraryDatasetDatasetAssociation, primaryjoin=( LibraryDatasetDatasetAssociation.table.c.copied_from_library_dataset_dataset_association_id == LibraryDatasetDatasetAssociation.table.c.id ), remote_side=[LibraryDatasetDatasetAssociation.table.c.id], uselist=False, back_populates="copied_to_library_dataset_dataset_associations", ), copied_to_library_dataset_dataset_associations=relationship( LibraryDatasetDatasetAssociation, primaryjoin=( LibraryDatasetDatasetAssociation.table.c.copied_from_library_dataset_dataset_association_id == LibraryDatasetDatasetAssociation.table.c.id ), back_populates="copied_from_library_dataset_dataset_association", ), copied_to_history_dataset_associations=relationship( HistoryDatasetAssociation, primaryjoin=( LibraryDatasetDatasetAssociation.table.c.id == HistoryDatasetAssociation.table.c.copied_from_library_dataset_dataset_association_id ), back_populates="copied_from_library_dataset_dataset_association", ), implicitly_converted_datasets=relationship( ImplicitlyConvertedDatasetAssociation, primaryjoin=( ImplicitlyConvertedDatasetAssociation.ldda_parent_id == LibraryDatasetDatasetAssociation.table.c.id ), back_populates="parent_ldda", ), tags=relationship( LibraryDatasetDatasetAssociationTagAssociation, order_by=LibraryDatasetDatasetAssociationTagAssociation.id, back_populates="library_dataset_dataset_association", ), extended_metadata=relationship( ExtendedMetadata, primaryjoin=(LibraryDatasetDatasetAssociation.table.c.extended_metadata_id == ExtendedMetadata.id), ), _metadata=deferred(LibraryDatasetDatasetAssociation.table.c._metadata), actions=relationship( LibraryDatasetDatasetAssociationPermissions, back_populates="library_dataset_dataset_association" ), dependent_jobs=relationship(JobToInputLibraryDatasetAssociation, back_populates="dataset"), creating_job_associations=relationship(JobToOutputLibraryDatasetAssociation, back_populates="dataset"), implicitly_converted_parent_datasets=relationship( ImplicitlyConvertedDatasetAssociation, primaryjoin=(lambda: ImplicitlyConvertedDatasetAssociation.ldda_id == LibraryDatasetDatasetAssociation.id), back_populates="dataset_ldda", ), copied_from_history_dataset_association=relationship( HistoryDatasetAssociation, primaryjoin=( HistoryDatasetAssociation.table.c.id == LibraryDatasetDatasetAssociation.table.c.copied_from_history_dataset_association_id ), back_populates="copied_to_library_dataset_dataset_associations", ), ), ) # ---------------------------------------------------------------------------------------- # The following statements must not precede the mapped models defined above. Job.any_output_dataset_collection_instances_deleted = deferred( column_property( # type:ignore[assignment] exists(HistoryDatasetCollectionAssociation.id).where( and_( Job.id == JobToOutputDatasetCollectionAssociation.job_id, HistoryDatasetCollectionAssociation.id == JobToOutputDatasetCollectionAssociation.dataset_collection_id, HistoryDatasetCollectionAssociation.deleted == true(), ) ), ) ) Job.any_output_dataset_deleted = deferred( column_property( # type:ignore[assignment] exists(HistoryDatasetAssociation.id).where( and_( Job.id == JobToOutputDatasetAssociation.job_id, HistoryDatasetAssociation.table.c.id == JobToOutputDatasetAssociation.dataset_id, HistoryDatasetAssociation.table.c.deleted == true(), ) ), ) ) History.average_rating = column_property( # type:ignore[assignment] select(func.avg(HistoryRatingAssociation.rating)) .where(HistoryRatingAssociation.history_id == History.id) .scalar_subquery(), deferred=True, ) History.users_shared_with_count = column_property( # type:ignore[assignment] select(func.count(HistoryUserShareAssociation.id)) .where(History.id == HistoryUserShareAssociation.history_id) .scalar_subquery(), deferred=True, ) Page.average_rating = column_property( select(func.avg(PageRatingAssociation.rating)).where(PageRatingAssociation.page_id == Page.id).scalar_subquery(), deferred=True, ) StoredWorkflow.average_rating = column_property( # type:ignore[assignment] select(func.avg(StoredWorkflowRatingAssociation.rating)) .where(StoredWorkflowRatingAssociation.stored_workflow_id == StoredWorkflow.id) .scalar_subquery(), deferred=True, ) Visualization.average_rating = column_property( # type:ignore[assignment] select(func.avg(VisualizationRatingAssociation.rating)) .where(VisualizationRatingAssociation.visualization_id == Visualization.id) .scalar_subquery(), deferred=True, ) Workflow.step_count = column_property( # type:ignore[assignment] select(func.count(WorkflowStep.id)).where(Workflow.id == WorkflowStep.workflow_id).scalar_subquery(), deferred=True ) WorkflowInvocationStep.subworkflow_invocation_id = column_property( select(WorkflowInvocationToSubworkflowInvocationAssociation.subworkflow_invocation_id) .where( and_( WorkflowInvocationToSubworkflowInvocationAssociation.workflow_invocation_id == WorkflowInvocationStep.workflow_invocation_id, WorkflowInvocationToSubworkflowInvocationAssociation.workflow_step_id == WorkflowInvocationStep.workflow_step_id, ) ) .scalar_subquery(), ) # Set up proxy so that this syntax is possible: # <user_obj>.preferences[pref_name] = pref_value User.preferences = association_proxy("_preferences", "value", creator=UserPreference) # Optimized version of getting the current Galaxy session. # See https://github.com/sqlalchemy/sqlalchemy/discussions/7638 for approach session_partition = select( GalaxySession, func.row_number() .over(order_by=GalaxySession.update_time.desc(), partition_by=GalaxySession.user_id) .label("index"), ).alias() partitioned_session = aliased(GalaxySession, session_partition) User.current_galaxy_session = relationship( partitioned_session, primaryjoin=and_(partitioned_session.user_id == User.id, session_partition.c.index < 2), uselist=False, viewonly=True, )
[docs]@event.listens_for(HistoryDatasetCollectionAssociation, "init") def receive_init(target, args, kwargs): """ Listens for the 'init' event. This is not called when 'target' is loaded from the database. https://docs.sqlalchemy.org/en/14/orm/events.html#sqlalchemy.orm.InstanceEvents.init Addresses SQLAlchemy 2.0 compatibility issue: see inline documentation for `add_object_to_object_session` in galaxy.model.orm.util. """ for key in ("history", "copied_from_history_dataset_collection_association"): obj = kwargs.get(key) if obj: add_object_to_object_session(target, obj) return # Once is enough.
JobStateSummary = NamedTuple("JobStateSummary", [(value, int) for value in enum_values(Job.states)] + [("all_jobs", int)]) # type: ignore[misc] # Ref https://github.com/python/mypy/issues/848#issuecomment-255237167