mirror of
https://github.com/ansible/awx.git
synced 2026-05-17 22:37:41 -02:30
Make canceling view non-atomic to fix 500 errors with job bursts (#13072)
* Make canceling view non-atomic to fix 500 errors with job bursts * Update test calls for cancel method changes
This commit is contained in:
@@ -3,6 +3,7 @@ import uuid
|
||||
import json
|
||||
|
||||
from django.conf import settings
|
||||
from django.db import connection
|
||||
import redis
|
||||
|
||||
from awx.main.dispatch import get_local_queuename
|
||||
@@ -49,7 +50,10 @@ class Control(object):
|
||||
reply_queue = Control.generate_reply_queue_name()
|
||||
self.result = None
|
||||
|
||||
with pg_bus_conn(new_connection=True) as conn:
|
||||
if not connection.get_autocommit():
|
||||
raise RuntimeError('Control-with-reply messages can only be done in autocommit mode')
|
||||
|
||||
with pg_bus_conn() as conn:
|
||||
conn.listen(reply_queue)
|
||||
send_data = {'control': command, 'reply_to': reply_queue}
|
||||
if extra_data:
|
||||
|
||||
@@ -1465,23 +1465,23 @@ class UnifiedJob(
|
||||
self.job_explanation = job_explanation
|
||||
cancel_fields.append('job_explanation')
|
||||
|
||||
# Important to save here before sending cancel signal to dispatcher to cancel because
|
||||
# the job control process will use the cancel_flag to distinguish a shutdown from a cancel
|
||||
self.save(update_fields=cancel_fields)
|
||||
|
||||
controller_notified = False
|
||||
if self.celery_task_id:
|
||||
controller_notified = self.cancel_dispatcher_process()
|
||||
|
||||
else:
|
||||
# Avoid race condition where we have stale model from pending state but job has already started,
|
||||
# its checking signal but not cancel_flag, so re-send signal after this database commit
|
||||
connection.on_commit(self.fallback_cancel)
|
||||
|
||||
# If a SIGTERM signal was sent to the control process, and acked by the dispatcher
|
||||
# then we want to let its own cleanup change status, otherwise change status now
|
||||
if not controller_notified:
|
||||
if self.status != 'canceled':
|
||||
self.status = 'canceled'
|
||||
cancel_fields.append('status')
|
||||
|
||||
self.save(update_fields=cancel_fields)
|
||||
self.save(update_fields=['status'])
|
||||
# Avoid race condition where we have stale model from pending state but job has already started,
|
||||
# its checking signal but not cancel_flag, so re-send signal after updating cancel fields
|
||||
self.fallback_cancel()
|
||||
|
||||
return self.cancel_flag
|
||||
|
||||
|
||||
@@ -50,7 +50,10 @@ def test_cancel(unified_job):
|
||||
# Some more thought may want to go into only emitting canceled if/when the job record
|
||||
# status is changed to canceled. Unlike, currently, where it's emitted unconditionally.
|
||||
unified_job.websocket_emit_status.assert_called_with("canceled")
|
||||
unified_job.save.assert_called_with(update_fields=['cancel_flag', 'start_args', 'status'])
|
||||
assert [(args, kwargs) for args, kwargs in unified_job.save.call_args_list] == [
|
||||
((), {'update_fields': ['cancel_flag', 'start_args']}),
|
||||
((), {'update_fields': ['status']}),
|
||||
]
|
||||
|
||||
|
||||
def test_cancel_job_explanation(unified_job):
|
||||
@@ -60,7 +63,10 @@ def test_cancel_job_explanation(unified_job):
|
||||
unified_job.cancel(job_explanation=job_explanation)
|
||||
|
||||
assert unified_job.job_explanation == job_explanation
|
||||
unified_job.save.assert_called_with(update_fields=['cancel_flag', 'start_args', 'job_explanation', 'status'])
|
||||
assert [(args, kwargs) for args, kwargs in unified_job.save.call_args_list] == [
|
||||
((), {'update_fields': ['cancel_flag', 'start_args', 'job_explanation']}),
|
||||
((), {'update_fields': ['status']}),
|
||||
]
|
||||
|
||||
|
||||
def test_organization_copy_to_jobs():
|
||||
|
||||
Reference in New Issue
Block a user