airflow-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Bolke de Bruin (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (AIRFLOW-1166) _change_state_for_tis_without_dagrun can take very long and end up in deadlocks
Date Tue, 02 May 2017 20:23:04 GMT

    [ https://issues.apache.org/jira/browse/AIRFLOW-1166?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15993659#comment-15993659
] 

Bolke de Bruin commented on AIRFLOW-1166:
-----------------------------------------

[~saguziel] [~aoen] you probably want to be aware of this as it might turn up in single scheduler
setups as well and besides the delay is significant.

> _change_state_for_tis_without_dagrun can take very long and end up in deadlocks
> -------------------------------------------------------------------------------
>
>                 Key: AIRFLOW-1166
>                 URL: https://issues.apache.org/jira/browse/AIRFLOW-1166
>             Project: Apache Airflow
>          Issue Type: Bug
>    Affects Versions: 1.8.1
>            Reporter: Bolke de Bruin
>            Assignee: Bolke de Bruin
>              Labels: scheduler
>             Fix For: 1.8.2
>
>
> While testing multi-scheduler setups we've seen some serious delays (>8s, with just
one dag running) in _change_state_for_tis_without_dagrun. This can also result in deadlocks
in the db due to the fact the query in _change_state_for_tis_without_dagrun is locking a significant
amount of task instances. This is unnecessary.



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

Mime
View raw message