You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@airflow.apache.org by "Teresa Martyny (JIRA)" <ji...@apache.org> on 2019/05/07 19:58:00 UTC
[jira] [Commented] (AIRFLOW-2761) Parallelize Celery Executor
enqueuing
[ https://issues.apache.org/jira/browse/AIRFLOW-2761?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16835061#comment-16835061 ]
Teresa Martyny commented on AIRFLOW-2761:
-----------------------------------------
Hey there,
We're seeing a quite a few process time outs raising during our core pipeline run. We traced it back to CeleryExecutor#sync calling #fetch_celery_task_state and timing out here:
```
with timeout(seconds=2):
# Accessing state property of celery task will make actual network request
# to get the current state of the task.
res = (celery_task[0], celery_task[1].state)
```
along with an AirflowTaskTimeout raising here:
```
if isinstance(result, ExceptionWithTraceback):
self.log.error(
CELERY_SEND_ERR_MSG_HEADER + ":{}\n{}\n".format(
result.exception, result.traceback))
```
Any suggestions as to what might be causing these timeouts in syncing the task state that we could fix on our end?
Thanks!!
> Parallelize Celery Executor enqueuing
> -------------------------------------
>
> Key: AIRFLOW-2761
> URL: https://issues.apache.org/jira/browse/AIRFLOW-2761
> Project: Apache Airflow
> Issue Type: Improvement
> Components: celery
> Reporter: Kevin Yang
> Priority: Major
> Fix For: 1.10.3
>
>
> Currently celery executor enqueues in an async fashion but still doing that in a single process loop. This can slows down scheduler loop and creates scheduling delay if we have large # of task to schedule in a short time, e.g. UTC midnight we need to schedule large # of sensors in a short period.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)