Merge pull request #3661 from open-craft/mtyaka/autorestart-workers

Always autorestart edxapp workers

playbooks/roles/edxapp/templates/workers.conf.j2

@@ -10,6 +10,11 @@ stderr_logfile={{ supervisor_log_dir }}/%(program_name)s-stderr.log
 command={{ edxapp_venv_dir + '/bin/newrelic-admin run-program ' if w.monitor and COMMON_ENABLE_NEWRELIC_APP else ''}}{{ edxapp_venv_bin }}/python {{ edxapp_code_dir }}/manage.py {{ w.service_variant }} --settings={{ worker_django_settings_module }} celery worker --loglevel=info --queues=edx.{{ w.service_variant }}.core.{{ w.queue }} --hostname=edx.{{ w.service_variant }}.core.{{ w.queue }}.%%h --concurrency={{ w.concurrency }} {{ '--maxtasksperchild ' + w.max_tasks_per_child|string if w.max_tasks_per_child is defined else '' }}
 killasgroup=true
 stopwaitsecs={{ w.stopwaitsecs | default(EDXAPP_WORKER_DEFAULT_STOPWAITSECS) }}
+; Set autorestart to `true`. The default value for autorestart is `unexpected`, but celery < 4.x will exit
+; with an exit code of zero for certain types of unrecoverable errors, so we must make sure that the workers
+; are auto restarted even when exiting with code 0.
+; The Celery bug was reported in https://github.com/celery/celery/issues/2024, and is fixed in Celery 4.0.0.
+autorestart=true
 
 {% endfor %}