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.webapps.reports.controllers.jobs
from __future__ import print_function
import calendar
import logging
import re
from collections import namedtuple
from datetime import (
date,
datetime,
timedelta
)
from math import ceil, floor
import sqlalchemy as sa
from markupsafe import escape
from sqlalchemy import and_, not_, or_
from galaxy import model, util
from galaxy.web.base.controller import BaseUIController, web
from galaxy.webapps.reports.controllers.query import ReportQueryBuilder
from galaxy.webapps.reports.framework import grids
log = logging.getLogger(__name__)
[docs]class Timer(object):
[docs] def stop(self):
try:
self.stop_time = datetime.now()
self.time_delta = self.stop_time - self.start_time
del(self.stop_time)
del(self.start_time)
except NameError:
print("You need to start before you can stop!")
[docs] def time_elapsed(self):
try:
return_time = self.time_delta - self.ERROR
except NameError:
print("You need to start and stop before there's an elapsed time!")
except AttributeError:
return_time = self.time_delta
return return_time
[docs]def sorter(default_sort_id, kwd):
"""
Initialize sorting variables
"""
SortSpec = namedtuple('SortSpec', ['sort_id', 'order', 'arrow', 'exc_order'])
sort_id = kwd.get('sort_id')
order = kwd.get('order')
# Parse the default value
if sort_id == "default":
sort_id = default_sort_id
# Create the sort
if order == "asc":
_order = sa.asc(sort_id)
elif order == "desc":
_order = sa.desc(sort_id)
else:
# In case of default
order = "desc"
_order = sa.desc(sort_id)
# Create an arrow icon to put beside the ordered column
up_arrow = "↑"
down_arrow = "↓"
arrow = " "
if order == "asc":
arrow += down_arrow
else:
arrow += up_arrow
return SortSpec(sort_id, order, arrow, _order)
[docs]def get_spark_time(time_period):
_time_period = 0
if time_period == "days":
_time_period = 1.0
elif time_period == "weeks":
_time_period = 7.0
elif time_period == "months":
_time_period = 30.0
elif time_period == "years":
_time_period = 365.0
else:
time_period = "days"
_time_period = 1.0
return time_period, _time_period
[docs]class SpecifiedDateListGrid(grids.Grid):
[docs] class StateColumn(grids.TextColumn):
[docs] def get_value(self, trans, grid, job):
return '<div class="count-box state-color-%s">%s</div>' % (job.state, job.state)
[docs] def filter(self, trans, user, query, column_filter):
if column_filter == 'Unfinished':
return query.filter(not_(or_(model.Job.table.c.state == model.Job.states.OK,
model.Job.table.c.state == model.Job.states.ERROR,
model.Job.table.c.state == model.Job.states.DELETED)))
return query
[docs] class ToolColumn(grids.TextColumn):
[docs] def filter(self, trans, user, query, column_filter):
if column_filter is not None:
query = query.filter(model.Job.table.c.tool_id == column_filter)
return query
[docs] class CreateTimeColumn(grids.DateTimeColumn):
[docs] def get_value(self, trans, grid, job):
return job.create_time.strftime("%b %d, %Y, %H:%M:%S")
[docs] class UserColumn(grids.GridColumn):
[docs] def get_value(self, trans, grid, job):
if job.user:
return escape(job.user.email)
return 'anonymous'
[docs] class EmailColumn(grids.GridColumn):
[docs] def filter(self, trans, user, query, column_filter):
if column_filter == 'All':
return query
return query.filter(and_(model.Job.table.c.user_id == model.User.table.c.id,
model.User.table.c.email == column_filter))
[docs] class SpecifiedDateColumn(grids.GridColumn):
[docs] def filter(self, trans, user, query, column_filter):
if column_filter == 'All':
return query
# We are either filtering on a date like YYYY-MM-DD or on a month like YYYY-MM,
# so we need to figure out which type of date we have
if column_filter.count('-') == 2: # We are filtering on a date like YYYY-MM-DD
year, month, day = map(int, column_filter.split("-"))
start_date = date(year, month, day)
end_date = start_date + timedelta(days=1)
if column_filter.count('-') == 1: # We are filtering on a month like YYYY-MM
year, month = map(int, column_filter.split("-"))
start_date = date(year, month, 1)
end_date = start_date + timedelta(days=calendar.monthrange(year, month)[1])
return query.filter(and_(self.model_class.table.c.create_time >= start_date,
self.model_class.table.c.create_time < end_date))
# Grid definition
use_async = False
model_class = model.Job
title = "Jobs"
default_sort_key = "id"
columns = [
JobIdColumn("Id",
key="id",
link=(lambda item: dict(operation="job_info", id=item.id, webapp="reports")),
attach_popup=False,
filterable="advanced"),
StateColumn("State",
key="state",
attach_popup=False),
ToolColumn("Tool Id",
key="tool_id",
link=(lambda item: dict(operation="tool_per_month", id=item.id, webapp="reports")),
attach_popup=False),
CreateTimeColumn("Creation Time",
key="create_time",
attach_popup=False),
UserColumn("User",
key="email",
model_class=model.User,
link=(lambda item: dict(operation="user_per_month", id=item.id, webapp="reports")),
attach_popup=False),
# Columns that are valid for filtering but are not visible.
SpecifiedDateColumn("Specified Date",
key="specified_date",
visible=False),
EmailColumn("Email",
key="email",
model_class=model.User,
visible=False),
grids.StateColumn("State",
key="state",
visible=False,
filterable="advanced")
]
columns.append(grids.MulticolFilterColumn("Search",
cols_to_filter=[columns[1], columns[2]],
key="free-text-search",
visible=False,
filterable="standard"))
standard_filters = []
default_filter = {'specified_date': 'All'}
num_rows_per_page = 50
use_paging = True
[docs] def build_initial_query(self, trans, **kwd):
params = util.Params(kwd)
monitor_email = params.get('monitor_email', 'monitor@bx.psu.edu')
monitor_user_id = get_monitor_id(trans, monitor_email)
return trans.sa_session.query(self.model_class) \
.join(model.User) \
.filter(model.Job.table.c.user_id != monitor_user_id)\
.enable_eagerloads(False)
[docs]class Jobs(BaseUIController, ReportQueryBuilder):
"""
Class contains functions for querying data requested by user via the webapp. It exposes the functions and
responds to requests with the filled .mako templates.
"""
specified_date_list_grid = SpecifiedDateListGrid()
[docs] @web.expose
def specified_date_handler(self, trans, **kwd):
# We add params to the keyword dict in this method in order to rename the param
# with an "f-" prefix, simulating filtering by clicking a search link. We have
# to take this approach because the "-" character is illegal in HTTP requests.
kwd['sort_id'] = 'default'
kwd['order'] = 'default'
if 'f-specified_date' in kwd and 'specified_date' not in kwd:
# The user clicked a State link in the Advanced Search box, so 'specified_date'
# will have been eliminated.
pass
elif 'specified_date' not in kwd:
kwd['f-specified_date'] = 'All'
else:
kwd['f-specified_date'] = kwd['specified_date']
if 'operation' in kwd:
operation = kwd['operation'].lower()
if operation == "job_info":
return trans.response.send_redirect(web.url_for(controller='jobs',
action='job_info',
**kwd))
elif operation == "tool_for_month":
kwd['f-tool_id'] = kwd['tool_id']
elif operation == "tool_per_month":
# The received id is the job id, so we need to get the job's tool_id.
job_id = kwd.get('id', None)
job = get_job(trans, job_id)
kwd['tool_id'] = job.tool_id
return trans.response.send_redirect(web.url_for(controller='jobs',
action='tool_per_month',
**kwd))
elif operation == "user_for_month":
kwd['f-email'] = util.restore_text(kwd['email'])
elif operation == "user_per_month":
# The received id is the job id, so we need to get the id of the user
# that submitted the job.
job_id = kwd.get('id', None)
job = get_job(trans, job_id)
if job.user:
kwd['email'] = job.user.email
else:
kwd['email'] = None # For anonymous users
return trans.response.send_redirect(web.url_for(controller='jobs',
action='user_per_month',
**kwd))
elif operation == "specified_date_in_error":
kwd['f-state'] = 'error'
elif operation == "unfinished":
kwd['f-state'] = 'Unfinished'
elif operation == "specified_tool_in_error":
kwd['f-state'] = 'error'
kwd['f-tool_id'] = kwd['tool_id']
return self.specified_date_list_grid(trans, **kwd)
def _calculate_trends_for_jobs(self, jobs_query):
trends = dict()
for job in jobs_query.execute():
job_day = int(job.date.strftime("%-d")) - 1
job_month = int(job.date.strftime("%-m"))
job_month_name = job.date.strftime("%B")
job_year = job.date.strftime("%Y")
key = str(job_month_name + job_year)
try:
trends[key][job_day] += 1
except KeyError:
job_year = int(job_year)
wday, day_range = calendar.monthrange(job_year, job_month)
trends[key] = [0] * day_range
trends[key][job_day] += 1
return trends
def _calculate_job_table(self, jobs_query):
jobs = []
for row in jobs_query.execute():
month_name = row.date.strftime("%B")
year = int(row.date.strftime("%Y"))
jobs.append((
row.date.strftime("%Y-%m"),
row.total_jobs,
month_name,
year
))
return jobs
[docs] @web.expose
def specified_month_all(self, trans, **kwd):
"""
Queries the DB for all jobs in given month, defaults to current month.
"""
message = ''
PageSpec = namedtuple('PageSpec', ['entries', 'offset', 'page', 'pages_found'])
params = util.Params(kwd)
monitor_email = params.get('monitor_email', 'monitor@bx.psu.edu')
specs = sorter('date', kwd)
offset = 0
limit = 10
sort_id = specs.sort_id
order = specs.order
arrow = specs.arrow
_order = specs.exc_order
if "entries" in kwd:
entries = int(kwd.get('entries'))
else:
entries = 10
limit = entries * 4
if "offset" in kwd:
offset = int(kwd.get('offset'))
else:
offset = 0
if "page" in kwd:
page = int(kwd.get('page'))
else:
page = 1
# In case we don't know which is the monitor user we will query for all jobs
monitor_user_id = get_monitor_id(trans, monitor_email)
# If specified_date is not received, we'll default to the current month
specified_date = kwd.get('specified_date', datetime.utcnow().strftime("%Y-%m-%d"))
specified_month = specified_date[:7]
year, month = map(int, specified_month.split("-"))
start_date = date(year, month, 1)
end_date = start_date + timedelta(days=calendar.monthrange(year, month)[1])
month_label = start_date.strftime("%B")
year_label = start_date.strftime("%Y")
# Use to make the page table
month_jobs = sa.select((sa.func.date(model.Job.table.c.create_time).label('date'),
sa.func.count(model.Job.table.c.id).label('total_jobs')),
whereclause=sa.and_(model.Job.table.c.user_id != monitor_user_id,
model.Job.table.c.create_time >= start_date,
model.Job.table.c.create_time < end_date),
from_obj=[model.Job.table],
group_by=['date'],
order_by=[_order],
offset=offset,
limit=limit)
# Use to make trendline
all_jobs = sa.select((model.Job.table.c.create_time.label('date'), model.Job.table.c.id.label('id')),
whereclause=sa.and_(model.Job.table.c.user_id != monitor_user_id,
model.Job.table.c.create_time >= start_date,
model.Job.table.c.create_time < end_date))
trends = dict()
for job in all_jobs.execute():
job_hour = int(job.date.strftime("%-H"))
job_day = job.date.strftime("%d")
try:
trends[job_day][job_hour] += 1
except KeyError:
trends[job_day] = [0] * 24
trends[job_day][job_hour] += 1
jobs = []
for row in month_jobs.execute():
row_dayname = row.date.strftime("%A")
row_day = row.date.strftime("%d")
jobs.append((row_dayname,
row_day,
row.total_jobs,
row.date))
pages_found = ceil(len(jobs) / float(entries))
page_specs = PageSpec(entries, offset, page, pages_found)
return trans.fill_template('/webapps/reports/jobs_specified_month_all.mako',
order=order,
arrow=arrow,
sort_id=sort_id,
month_label=month_label,
year_label=year_label,
month=month,
page_specs=page_specs,
jobs=jobs,
trends=trends,
is_user_jobs_only=monitor_user_id,
message=message)
[docs] @web.expose
def specified_month_in_error(self, trans, **kwd):
"""
Queries the DB for the user jobs in error.
"""
message = ''
PageSpec = namedtuple('PageSpec', ['entries', 'offset', 'page', 'pages_found'])
params = util.Params(kwd)
monitor_email = params.get('monitor_email', 'monitor@bx.psu.edu')
specs = sorter('date', kwd)
sort_id = specs.sort_id
order = specs.order
arrow = specs.arrow
_order = specs.exc_order
offset = 0
limit = 10
if "entries" in kwd:
entries = int(kwd.get('entries'))
else:
entries = 10
limit = entries * 4
if "offset" in kwd:
offset = int(kwd.get('offset'))
else:
offset = 0
if "page" in kwd:
page = int(kwd.get('page'))
else:
page = 1
# In case we don't know which is the monitor user we will query for all jobs instead
monitor_user_id = get_monitor_id(trans, monitor_email)
# If specified_date is not received, we'll default to the current month
specified_date = kwd.get('specified_date', datetime.utcnow().strftime("%Y-%m-%d"))
specified_month = specified_date[:7]
year, month = map(int, specified_month.split("-"))
start_date = date(year, month, 1)
end_date = start_date + timedelta(days=calendar.monthrange(year, month)[1])
month_label = start_date.strftime("%B")
year_label = start_date.strftime("%Y")
month_jobs_in_error = sa.select((sa.func.date(model.Job.table.c.create_time).label('date'),
sa.func.count(model.Job.table.c.id).label('total_jobs')),
whereclause=sa.and_(model.Job.table.c.user_id != monitor_user_id,
model.Job.table.c.state == 'error',
model.Job.table.c.create_time >= start_date,
model.Job.table.c.create_time < end_date),
from_obj=[model.Job.table],
group_by=['date'],
order_by=[_order],
offset=offset,
limit=limit)
# Use to make trendline
all_jobs_in_error = sa.select((model.Job.table.c.create_time.label('date'), model.Job.table.c.id.label('id')),
whereclause=sa.and_(model.Job.table.c.user_id != monitor_user_id,
model.Job.table.c.state == 'error',
model.Job.table.c.create_time >= start_date,
model.Job.table.c.create_time < end_date))
trends = dict()
for job in all_jobs_in_error.execute():
job_hour = int(job.date.strftime("%-H"))
job_day = job.date.strftime("%d")
try:
trends[job_day][job_hour] += 1
except KeyError:
trends[job_day] = [0] * 24
trends[job_day][job_hour] += 1
jobs = []
for row in month_jobs_in_error.execute():
row_dayname = row.date.strftime("%A")
row_day = row.date.strftime("%d")
jobs.append((row_dayname,
row_day,
row.total_jobs,
row.date))
pages_found = ceil(len(jobs) / float(entries))
page_specs = PageSpec(entries, offset, page, pages_found)
return trans.fill_template('/webapps/reports/jobs_specified_month_in_error.mako',
order=order,
arrow=arrow,
sort_id=sort_id,
month_label=month_label,
year_label=year_label,
month=month,
jobs=jobs,
trends=trends,
message=message,
is_user_jobs_only=monitor_user_id,
page_specs=page_specs)
[docs] @web.expose
def per_month_all(self, trans, **kwd):
"""
Queries the DB for all jobs. Avoids monitor jobs.
"""
message = ''
PageSpec = namedtuple('PageSpec', ['entries', 'offset', 'page', 'pages_found'])
params = util.Params(kwd)
monitor_email = params.get('monitor_email', 'monitor@bx.psu.edu')
specs = sorter('date', kwd)
sort_id = specs.sort_id
order = specs.order
arrow = specs.arrow
_order = specs.exc_order
offset = 0
limit = 10
if "entries" in kwd:
entries = int(kwd.get('entries'))
else:
entries = 10
limit = entries * 4
if "offset" in kwd:
offset = int(kwd.get('offset'))
else:
offset = 0
if "page" in kwd:
page = int(kwd.get('page'))
else:
page = 1
# In case we don't know which is the monitor user we will query for all jobs
monitor_user_id = get_monitor_id(trans, monitor_email)
# Use to make the page table
jobs_by_month = sa.select((self.select_month(model.Job.table.c.create_time).label('date'),
sa.func.count(model.Job.table.c.id).label('total_jobs')),
whereclause=model.Job.table.c.user_id != monitor_user_id,
from_obj=[model.Job.table],
group_by=self.group_by_month(model.Job.table.c.create_time),
order_by=[_order],
offset=offset,
limit=limit)
# Use to make sparkline
all_jobs = sa.select((self.select_day(model.Job.table.c.create_time).label('date'),
model.Job.table.c.id.label('id')))
trends = self._calculate_trends_for_jobs(all_jobs)
jobs = self._calculate_job_table(jobs_by_month)
pages_found = ceil(len(jobs) / float(entries))
page_specs = PageSpec(entries, offset, page, pages_found)
return trans.fill_template('/webapps/reports/jobs_per_month_all.mako',
order=order,
arrow=arrow,
sort_id=sort_id,
trends=trends,
jobs=jobs,
is_user_jobs_only=monitor_user_id,
message=message,
page_specs=page_specs)
[docs] @web.expose
def per_month_in_error(self, trans, **kwd):
"""
Queries the DB for user jobs in error. Filters out monitor jobs.
"""
message = ''
PageSpec = namedtuple('PageSpec', ['entries', 'offset', 'page', 'pages_found'])
params = util.Params(kwd)
monitor_email = params.get('monitor_email', 'monitor@bx.psu.edu')
specs = sorter('date', kwd)
sort_id = specs.sort_id
order = specs.order
arrow = specs.arrow
_order = specs.exc_order
offset = 0
limit = 10
if "entries" in kwd:
entries = int(kwd.get('entries'))
else:
entries = 10
limit = entries * 4
if "offset" in kwd:
offset = int(kwd.get('offset'))
else:
offset = 0
if "page" in kwd:
page = int(kwd.get('page'))
else:
page = 1
# In case we don't know which is the monitor user we will query for all jobs
monitor_user_id = get_monitor_id(trans, monitor_email)
# Use to make the page table
jobs_in_error_by_month = sa.select((self.select_month(model.Job.table.c.create_time).label('date'),
sa.func.count(model.Job.table.c.id).label('total_jobs')),
whereclause=sa.and_(model.Job.table.c.state == 'error',
model.Job.table.c.user_id != monitor_user_id),
from_obj=[model.Job.table],
group_by=self.group_by_month(model.Job.table.c.create_time),
order_by=[_order],
offset=offset,
limit=limit)
# Use to make trendline
all_jobs = sa.select((self.select_day(model.Job.table.c.create_time).label('date'),
model.Job.table.c.id.label('id')),
whereclause=sa.and_(model.Job.table.c.state == 'error',
model.Job.table.c.user_id != monitor_user_id))
trends = self._calculate_trends_for_jobs(all_jobs)
jobs = self._calculate_job_table(jobs_in_error_by_month)
pages_found = ceil(len(jobs) / float(entries))
page_specs = PageSpec(entries, offset, page, pages_found)
return trans.fill_template('/webapps/reports/jobs_per_month_in_error.mako',
order=order,
arrow=arrow,
sort_id=sort_id,
trends=trends,
jobs=jobs,
message=message,
is_user_jobs_only=monitor_user_id,
page_specs=page_specs,
offset=offset,
limit=limit)
[docs] @web.expose
def per_user(self, trans, **kwd):
total_time = Timer()
q_time = Timer()
total_time.start()
params = util.Params(kwd)
message = ''
PageSpec = namedtuple('PageSpec', ['entries', 'offset', 'page', 'pages_found'])
monitor_email = params.get('monitor_email', 'monitor@bx.psu.edu')
specs = sorter('user_email', kwd)
sort_id = specs.sort_id
order = specs.order
arrow = specs.arrow
_order = specs.exc_order
time_period = kwd.get('spark_time')
time_period, _time_period = get_spark_time(time_period)
spark_limit = 30
offset = 0
limit = 10
if "entries" in kwd:
entries = int(kwd.get('entries'))
else:
entries = 10
limit = entries * 4
if "offset" in kwd:
offset = int(kwd.get('offset'))
else:
offset = 0
if "page" in kwd:
page = int(kwd.get('page'))
else:
page = 1
jobs = []
jobs_per_user = sa.select((model.User.table.c.email.label('user_email'),
sa.func.count(model.Job.table.c.id).label('total_jobs')),
from_obj=[sa.outerjoin(model.Job.table, model.User.table)],
group_by=['user_email'],
order_by=[_order],
offset=offset,
limit=limit)
q_time.start()
for row in jobs_per_user.execute():
if (row.user_email is None):
jobs.append(('Anonymous',
row.total_jobs))
elif (row.user_email == monitor_email):
continue
else:
jobs.append((row.user_email,
row.total_jobs))
q_time.stop()
query1time = q_time.time_elapsed()
users = sa.select([model.User.table.c.email],
from_obj=[model.User.table])
all_jobs_per_user = sa.select((model.Job.table.c.id.label('id'),
model.Job.table.c.create_time.label('date'),
model.User.table.c.email.label('user_email')),
from_obj=[sa.outerjoin(model.Job.table, model.User.table)],
whereclause=model.User.table.c.email.in_(users))
currday = datetime.today()
trends = dict()
q_time.start()
for job in all_jobs_per_user.execute():
if job.user_email is None:
curr_user = 'Anonymous'
else:
curr_user = re.sub(r'\W+', '', job.user_email)
try:
day = currday - job.date
except TypeError:
day = currday - datetime.date(job.date)
day = day.days
container = floor(day / _time_period)
container = int(container)
try:
if container < spark_limit:
trends[curr_user][container] += 1
except KeyError:
trends[curr_user] = [0] * spark_limit
if container < spark_limit:
trends[curr_user][container] += 1
q_time.stop()
query2time = q_time.time_elapsed()
pages_found = ceil(len(jobs) / float(entries))
page_specs = PageSpec(entries, offset, page, pages_found)
total_time.stop()
ttime = total_time.time_elapsed()
return trans.fill_template('/webapps/reports/jobs_per_user.mako',
order=order,
arrow=arrow,
sort_id=sort_id,
spark_limit=spark_limit,
time_period=time_period,
q1time=query1time,
q2time=query2time,
ttime=ttime,
trends=trends,
jobs=jobs,
message=message,
page_specs=page_specs)
[docs] @web.expose
def user_per_month(self, trans, **kwd):
params = util.Params(kwd)
message = ''
email = util.restore_text(params.get('email', ''))
specs = sorter('date', kwd)
sort_id = specs.sort_id
order = specs.order
arrow = specs.arrow
_order = specs.exc_order
q = sa.select((self.select_month(model.Job.table.c.create_time).label('date'),
sa.func.count(model.Job.table.c.id).label('total_jobs')),
whereclause=model.User.table.c.email == email,
from_obj=[sa.join(model.Job.table, model.User.table)],
group_by=self.group_by_month(model.Job.table.c.create_time),
order_by=[_order])
all_jobs_per_user = sa.select((model.Job.table.c.create_time.label('date'),
model.Job.table.c.id.label('job_id')),
whereclause=sa.and_(model.User.table.c.email == email),
from_obj=[sa.join(model.Job.table, model.User.table)])
trends = dict()
for job in all_jobs_per_user.execute():
job_day = int(job.date.strftime("%-d")) - 1
job_month = int(job.date.strftime("%-m"))
job_month_name = job.date.strftime("%B")
job_year = job.date.strftime("%Y")
key = str(job_month_name + job_year)
try:
trends[key][job_day] += 1
except KeyError:
job_year = int(job_year)
wday, day_range = calendar.monthrange(job_year, job_month)
trends[key] = [0] * day_range
trends[key][job_day] += 1
jobs = []
for row in q.execute():
jobs.append((row.date.strftime("%Y-%m"),
row.total_jobs,
row.date.strftime("%B"),
row.date.strftime("%Y")))
return trans.fill_template('/webapps/reports/jobs_user_per_month.mako',
order=order,
arrow=arrow,
sort_id=sort_id,
id=kwd.get('id'),
trends=trends,
email=util.sanitize_text(email),
jobs=jobs, message=message)
[docs] @web.expose
def per_tool(self, trans, **kwd):
message = ''
PageSpec = namedtuple('PageSpec', ['entries', 'offset', 'page', 'pages_found'])
params = util.Params(kwd)
monitor_email = params.get('monitor_email', 'monitor@bx.psu.edu')
specs = sorter('tool_id', kwd)
sort_id = specs.sort_id
order = specs.order
arrow = specs.arrow
_order = specs.exc_order
time_period = kwd.get('spark_time')
time_period, _time_period = get_spark_time(time_period)
spark_limit = 30
offset = 0
limit = 10
if "entries" in kwd:
entries = int(kwd.get('entries'))
else:
entries = 10
limit = entries * 4
if "offset" in kwd:
offset = int(kwd.get('offset'))
else:
offset = 0
if "page" in kwd:
page = int(kwd.get('page'))
else:
page = 1
# In case we don't know which is the monitor user we will query for all jobs
monitor_user_id = get_monitor_id(trans, monitor_email)
jobs = []
q = sa.select((model.Job.table.c.tool_id.label('tool_id'),
sa.func.count(model.Job.table.c.id).label('total_jobs')),
whereclause=model.Job.table.c.user_id != monitor_user_id,
from_obj=[model.Job.table],
group_by=['tool_id'],
order_by=[_order],
offset=offset,
limit=limit)
all_jobs_per_tool = sa.select((model.Job.table.c.tool_id.label('tool_id'),
model.Job.table.c.id.label('id'),
self.select_day(model.Job.table.c.create_time).label('date')),
whereclause=model.Job.table.c.user_id != monitor_user_id,
from_obj=[model.Job.table])
currday = date.today()
trends = dict()
for job in all_jobs_per_tool.execute():
curr_tool = re.sub(r'\W+', '', str(job.tool_id))
try:
day = currday - job.date
except TypeError:
day = currday - datetime.date(job.date)
day = day.days
container = floor(day / _time_period)
container = int(container)
try:
if container < spark_limit:
trends[curr_tool][container] += 1
except KeyError:
trends[curr_tool] = [0] * spark_limit
if container < spark_limit:
trends[curr_tool][container] += 1
for row in q.execute():
jobs.append((row.tool_id,
row.total_jobs))
pages_found = ceil(len(jobs) / float(entries))
page_specs = PageSpec(entries, offset, page, pages_found)
return trans.fill_template('/webapps/reports/jobs_per_tool.mako',
order=order,
arrow=arrow,
sort_id=sort_id,
spark_limit=spark_limit,
time_period=time_period,
trends=trends,
jobs=jobs,
message=message,
is_user_jobs_only=monitor_user_id,
page_specs=page_specs)
[docs] @web.expose
def errors_per_tool(self, trans, **kwd):
"""
Queries the DB for user jobs in error. Filters out monitor jobs.
"""
message = ''
PageSpec = namedtuple('PageSpec', ['entries', 'offset', 'page', 'pages_found'])
params = util.Params(kwd)
monitor_email = params.get('monitor_email', 'monitor@bx.psu.edu')
specs = sorter('tool_id', kwd)
sort_id = specs.sort_id
order = specs.order
arrow = specs.arrow
_order = specs.exc_order
time_period = kwd.get('spark_time')
time_period, _time_period = get_spark_time(time_period)
spark_limit = 30
offset = 0
limit = 10
if "entries" in kwd:
entries = int(kwd.get('entries'))
else:
entries = 10
limit = entries * 4
if "offset" in kwd:
offset = int(kwd.get('offset'))
else:
offset = 0
if "page" in kwd:
page = int(kwd.get('page'))
else:
page = 1
# In case we don't know which is the monitor user we will query for all jobs
monitor_user_id = get_monitor_id(trans, monitor_email)
jobs_in_error_per_tool = sa.select((model.Job.table.c.tool_id.label('tool_id'),
sa.func.count(model.Job.table.c.id).label('total_jobs')),
whereclause=sa.and_(model.Job.table.c.state == 'error',
model.Job.table.c.user_id != monitor_user_id),
from_obj=[model.Job.table],
group_by=['tool_id'],
order_by=[_order],
offset=offset,
limit=limit)
all_jobs_per_tool_errors = sa.select((self.select_day(model.Job.table.c.create_time).label('date'),
model.Job.table.c.id.label('id'),
model.Job.table.c.tool_id.label('tool_id')),
whereclause=sa.and_(model.Job.table.c.state == 'error',
model.Job.table.c.user_id != monitor_user_id),
from_obj=[model.Job.table])
currday = date.today()
trends = dict()
for job in all_jobs_per_tool_errors.execute():
curr_tool = re.sub(r'\W+', '', str(job.tool_id))
try:
day = currday - job.date
except TypeError:
day = currday - datetime.date(job.date)
# convert day into days/weeks/months/years
day = day.days
container = floor(day / _time_period)
container = int(container)
try:
if container < spark_limit:
trends[curr_tool][container] += 1
except KeyError:
trends[curr_tool] = [0] * spark_limit
if day < spark_limit:
trends[curr_tool][container] += 1
jobs = []
for row in jobs_in_error_per_tool.execute():
jobs.append((row.total_jobs, row.tool_id))
pages_found = ceil(len(jobs) / float(entries))
page_specs = PageSpec(entries, offset, page, pages_found)
return trans.fill_template('/webapps/reports/jobs_errors_per_tool.mako',
order=order,
arrow=arrow,
sort_id=sort_id,
spark_limit=spark_limit,
time_period=time_period,
trends=trends,
jobs=jobs,
message=message,
is_user_jobs_only=monitor_user_id,
page_specs=page_specs)
[docs] @web.expose
def tool_per_month(self, trans, **kwd):
message = ''
params = util.Params(kwd)
monitor_email = params.get('monitor_email', 'monitor@bx.psu.edu')
specs = sorter('date', kwd)
sort_id = specs.sort_id
order = specs.order
arrow = specs.arrow
_order = specs.exc_order
# In case we don't know which is the monitor user we will query for all jobs
monitor_user_id = get_monitor_id(trans, monitor_email)
tool_id = params.get('tool_id', 'Add a column1')
specified_date = params.get('specified_date', datetime.utcnow().strftime("%Y-%m-%d"))
q = sa.select((self.select_month(model.Job.table.c.create_time).label('date'),
sa.func.count(model.Job.table.c.id).label('total_jobs')),
whereclause=sa.and_(model.Job.table.c.tool_id == tool_id,
model.Job.table.c.user_id != monitor_user_id),
from_obj=[model.Job.table],
group_by=self.group_by_month(model.Job.table.c.create_time),
order_by=[_order])
# Use to make sparkline
all_jobs_for_tool = sa.select((self.select_month(model.Job.table.c.create_time).label('month'),
self.select_day(model.Job.table.c.create_time).label('day'),
model.Job.table.c.id.label('id')),
whereclause=sa.and_(model.Job.table.c.tool_id == tool_id,
model.Job.table.c.user_id != monitor_user_id),
from_obj=[model.Job.table])
trends = dict()
for job in all_jobs_for_tool.execute():
job_day = int(job.day.strftime("%-d")) - 1
job_month = int(job.month.strftime("%-m"))
job_month_name = job.month.strftime("%B")
job_year = job.month.strftime("%Y")
key = str(job_month_name + job_year)
try:
trends[key][job_day] += 1
except KeyError:
job_year = int(job_year)
wday, day_range = calendar.monthrange(job_year, job_month)
trends[key] = [0] * day_range
trends[key][job_day] += 1
jobs = []
for row in q.execute():
jobs.append((row.date.strftime("%Y-%m"),
row.total_jobs,
row.date.strftime("%B"),
row.date.strftime("%Y")))
return trans.fill_template('/webapps/reports/jobs_tool_per_month.mako',
order=order,
arrow=arrow,
sort_id=sort_id,
specified_date=specified_date,
tool_id=tool_id,
trends=trends,
jobs=jobs,
message=message,
is_user_jobs_only=monitor_user_id)
[docs] @web.expose
def job_info(self, trans, **kwd):
message = ''
job = trans.sa_session.query(model.Job) \
.get(trans.security.decode_id(kwd.get('id', '')))
return trans.fill_template('/webapps/reports/job_info.mako',
job=job,
message=message)
# ---- Utility methods -------------------------------------------------------
[docs]def get_job(trans, id):
return trans.sa_session.query(trans.model.Job).get(trans.security.decode_id(id))
[docs]def get_monitor_id(trans, monitor_email):
"""
A convenience method to obtain the monitor job id.
"""
monitor_user_id = None
monitor_row = trans.sa_session.query(trans.model.User.table.c.id) \
.filter(trans.model.User.table.c.email == monitor_email) \
.first()
if monitor_row is not None:
monitor_user_id = monitor_row[0]
return monitor_user_id