taskcluster/taskgraph/create.py

   1 # This Source Code Form is subject to the terms of the Mozilla Public
   2 # License, v. 2.0. If a copy of the MPL was not distributed with this
   3 # file, You can obtain one at http://mozilla.org/MPL/2.0/.
   4
   5 from __future__ import absolute_import, print_function, unicode_literals
   6
   7 import concurrent.futures as futures
   8 import requests
   9 import requests.adapters
  10 import json
  11 import os
  12 import sys
  13 import logging
  14
  15 from slugid import nice as slugid
  16 from taskgraph.util.parameterization import resolve_timestamps
  17 from taskgraph.util.time import current_json_time
  18 from taskgraph.util.taskcluster import get_session, CONCURRENCY
  19
  20 logger = logging.getLogger(__name__)
  21
  22 # this is set to true for `mach taskgraph action-callback --test`
  23 testing = False
  24
  25
  26 def create_tasks(taskgraph, label_to_taskid, params, decision_task_id=None):
  27     taskid_to_label = {t: l for l, t in label_to_taskid.iteritems()}
  28
  29
  30     decision_task_id = decision_task_id or os.environ.get('TASK_ID')
  31
  32     # when running as an actual decision task, we use the decision task's
  33     # taskId as the taskGroupId.  The process that created the decision task
  34     # helpfully placed it in this same taskGroup.  If there is no $TASK_ID,
  35     # fall back to a slugid
  36     task_group_id = decision_task_id or slugid()
  37     scheduler_id = 'gecko-level-{}'.format(params['level'])
  38
  39     # Add the taskGroupId, schedulerId and optionally the decision task
  40     # dependency
  41     for task_id in taskgraph.graph.nodes:
  42         task_def = taskgraph.tasks[task_id].task
  43
  44         # if this task has no dependencies *within* this taskgraph, make it
  45         # depend on this decision task. If it has another dependency within
  46         # the taskgraph, then it already implicitly depends on the decision
  47         # task.  The result is that tasks do not start immediately. if this
  48         # loop fails halfway through, none of the already-created tasks run.
  49         if decision_task_id:
  50             if not any(t in taskgraph.tasks for t in task_def.get('dependencies', [])):
  51                 task_def.setdefault('dependencies', []).append(decision_task_id)
  52
  53         task_def['taskGroupId'] = task_group_id
  54         task_def['schedulerId'] = scheduler_id
  55
  56     # If `testing` is True, then run without parallelization
  57     concurrency = CONCURRENCY if not testing else 1
  58     session = get_session()
  59     with futures.ThreadPoolExecutor(concurrency) as e:
  60         fs = {}
  61
  62         # We can't submit a task until its dependencies have been submitted.
  63         # So our strategy is to walk the graph and submit tasks once all
  64         # their dependencies have been submitted.
  65         tasklist = set(taskgraph.graph.visit_postorder())
  66         alltasks = tasklist.copy()
  67
  68         def schedule_tasks():
  69             # bail out early if any futures have failed
  70             if any(f.done() and f.exception() for f in fs.values()):
  71                 return
  72
  73             to_remove = set()
  74             new = set()
  75
  76             def submit(task_id, label, task_def):
  77                 fut = e.submit(create_task, session, task_id, label, task_def)
  78                 new.add(fut)
  79                 fs[task_id] = fut
  80
  81             for task_id in tasklist:
  82                 task_def = taskgraph.tasks[task_id].task
  83                 # If we haven't finished submitting all our dependencies yet,
  84                 # come back to this later.
  85                 # Some dependencies aren't in our graph, so make sure to filter
  86                 # those out
  87                 deps = set(task_def.get('dependencies', [])) & alltasks
  88                 if any((d not in fs or not fs[d].done()) for d in deps):
  89                     continue
  90
  91                 submit(task_id, taskid_to_label[task_id], task_def)
  92                 to_remove.add(task_id)
  93
  94                 # Schedule tasks as many times as task_duplicates indicates
  95                 attributes = taskgraph.tasks[task_id].attributes
  96                 for i in range(1, attributes.get('task_duplicates', 1)):
  97                     # We use slugid() since we want a distinct task id
  98                     submit(slugid(), taskid_to_label[task_id], task_def)
  99             tasklist.difference_update(to_remove)
 100
 101             # as each of those futures complete, try to schedule more tasks
 102             for f in futures.as_completed(new):
 103                 schedule_tasks()
 104
 105         # start scheduling tasks and run until everything is scheduled
 106         schedule_tasks()
 107
 108         # check the result of each future, raising an exception if it failed
 109         for f in futures.as_completed(fs.values()):
 110             f.result()
 111
 112
 113 def create_task(session, task_id, label, task_def):
 114     # create the task using 'http://taskcluster/queue', which is proxied to the queue service
 115     # with credentials appropriate to this job.
 116
 117     # Resolve timestamps
 118     now = current_json_time(datetime_format=True)
 119     task_def = resolve_timestamps(now, task_def)
 120
 121     if testing:
 122         json.dump([task_id, task_def], sys.stdout,
 123                   sort_keys=True, indent=4, separators=(',', ': '))
 124         # add a newline
 125         print("")
 126         return
 127
 128     logger.debug("Creating task with taskId {} for {}".format(task_id, label))
 129     res = session.put('http://taskcluster/queue/v1/task/{}'.format(task_id),
 130                       data=json.dumps(task_def))
 131     if res.status_code != 200:
 132         try:
 133             logger.error(res.json()['message'])
 134         except Exception:
 135             logger.error(res.text)
 136         res.raise_for_status()