Source code for galaxy.jobs.rule_helper

import hashlib
import logging
import random
from datetime import datetime

from sqlalchemy import func

from galaxy import (
    model,
    util
)
from galaxy.tools.deps.dependencies import ToolInfo

log = logging.getLogger(__name__)

VALID_JOB_HASH_STRATEGIES = ["job", "user", "history", "workflow_invocation"]


[docs]class RuleHelper(object):
    """ Utility to allow job rules to interface cleanly with the rest of
    Galaxy and shield them from low-level details of models, metrics, etc....

    Currently focus is on figuring out job statistics for a given user, but
    could interface with other stuff as well.
    """

[docs]    def __init__(self, app):
        self.app = app

[docs]    def supports_docker(self, job_or_tool):
        """ Job rules can pass this function a job, job_wrapper, or tool and
        determine if the underlying tool believes it can be containered.
        """
        # Not a ton of logic in this method - but the idea is to shield rule
        # developers from the details and they shouldn't have to know how to
        # interrogate tool or job to figure out if it can be run in a
        # container.
        if hasattr(job_or_tool, 'containers'):
            tool = job_or_tool
        elif hasattr(job_or_tool, 'tool'):
            # Have a JobWrapper-like
            tool = job_or_tool.tool
        else:
            # Have a Job object.
            tool = self.app.toolbox.get_tool(job_or_tool.tool_id, tool_version=job_or_tool.tool_version)
        tool_info = ToolInfo(tool.containers, tool.requirements, tool.requires_galaxy_python_environment, tool.docker_env_pass_through)
        container_description = self.app.container_finder.find_best_container_description(["docker"], tool_info)
        return container_description is not None

[docs]    def job_count(
        self,
        **kwds
    ):
        query = self.query(model.Job)
        return self._filter_job_query(query, **kwds).count()

[docs]    def sum_job_runtime(
        self,
        **kwds
    ):
        # TODO: Consider sum_core_hours or something that scales runtime by
        # by calculated cores per job.
        query = self.metric_query(
            select=func.sum(model.JobMetricNumeric.table.c.metric_value),
            metric_name="runtime_seconds",
            plugin="core",
        )
        query = query.join(model.Job)
        return float(self._filter_job_query(query, **kwds).first()[0])

[docs]    def metric_query(self, select, metric_name, plugin, numeric=True):
        metric_class = model.JobMetricNumeric if numeric else model.JobMetricText
        query = self.query(select)
        query = query.filter(metric_class.table.c.plugin == plugin)
        query = query.filter(metric_class.table.c.metric_name == metric_name)
        return query

[docs]    def query(self, select_expression):
        return self.app.model.context.query(select_expression)

    def _filter_job_query(
        self,
        query,
        for_user_email=None,
        for_destination=None,
        for_destinations=None,
        for_job_states=None,
        created_in_last=None,
        updated_in_last=None,
    ):
        if for_destination is not None:
            for_destinations = [for_destination]

        query = query.join(model.User)
        if for_user_email is not None:
            query = query.filter(model.User.table.c.email == for_user_email)

        if for_destinations is not None:
            if len(for_destinations) == 1:
                query = query.filter(model.Job.table.c.destination_id == for_destinations[0])
            else:
                query = query.filter(model.Job.table.c.destination_id.in_(for_destinations))

        if created_in_last is not None:
            end_date = datetime.now()
            start_date = end_date - created_in_last
            query = query.filter(model.Job.table.c.create_time >= start_date)

        if updated_in_last is not None:
            end_date = datetime.now()
            start_date = end_date - updated_in_last
            log.info(end_date)
            log.info(start_date)
            query = query.filter(model.Job.table.c.update_time >= start_date)

        if for_job_states is not None:
            # Optimize the singleton case - can be much more performant in my experience.
            if len(for_job_states) == 1:
                query = query.filter(model.Job.table.c.state == for_job_states[0])
            else:
                query = query.filter(model.Job.table.c.state.in_(for_job_states))

        return query

[docs]    def should_burst(self, destination_ids, num_jobs, job_states=None):
        """ Check if the specified destinations ``destination_ids`` have at
        least ``num_jobs`` assigned to it - send in ``job_state`` as ``queued``
        to limit this check to number of jobs queued.

        See stock_rules for an simple example of using this function - but to
        get the most out of it - it should probably be used with custom job
        rules that can respond to the bursting by allocating resources,
        launching cloud nodes, etc....
        """
        if job_states is None:
            job_states = "queued,running"
        from_destination_job_count = self.job_count(
            for_destinations=destination_ids,
            for_job_states=util.listify(job_states)
        )
        # Would this job push us over maximum job count before requiring
        # bursting (roughly... very roughly given many handler threads may be
        # scheduling jobs).
        return (from_destination_job_count + 1) > int(num_jobs)

[docs]    def choose_one(self, lst, hash_value=None):
        """ Choose a random value from supplied list. If hash_value is passed
        in then every request with that same hash_value would produce the same
        choice from the supplied list.
        """
        if hash_value is None:
            return random.choice(lst)

        if not isinstance(hash_value, int):
            # Convert hash_value string into index
            as_hex = hashlib.md5(util.smart_str(hash_value)).hexdigest()
            hash_value = int(as_hex, 16)
        # else assumed to be 'random' int from 0-~Inf
        random_index = hash_value % len(lst)
        return lst[random_index]

[docs]    def job_hash(self, job, hash_by=None):
        """ Produce a reproducible hash for the given job on various
        criteria - for instance if hash_by is "workflow_invocation,history" -
        all jobs within the same workflow invocation will receive the same
        hash - for jobs outside of workflows all jobs within the same history
        will receive the same hash, other jobs will be hashed on job's id
        randomly.

        Primarily intended for use with ``choose_one`` above - to consistent
        route or schedule related jobs.
        """
        if hash_by is None:
            hash_by = ["job"]
        hash_bys = util.listify(hash_by)
        for hash_by in hash_bys:
            job_hash = self._try_hash_for_job(job, hash_by)
            if job_hash:
                return job_hash

        # Fall back to just hashing by job id, should always return a value.
        return self._try_hash_for_job(job, "job")

    def _try_hash_for_job(self, job, hash_by):
        """ May return False or None if hash type is invalid for that job -
        e.g. attempting to hash by user for anonymous job or by workflow
        invocation for jobs outside of workflows.
        """
        if hash_by not in VALID_JOB_HASH_STRATEGIES:
            message = "Do not know how to hash jobs by %s, must be one of %s" % (hash_by, VALID_JOB_HASH_STRATEGIES)
            raise Exception(message)

        if hash_by == "workflow_invocation":
            return job.raw_param_dict().get("__workflow_invocation_uuid__", None)
        elif hash_by == "history":
            return job.history_id
        elif hash_by == "user":
            user = job.user
            return user and user.id
        elif hash_by == "job":
            return job.id