[
https://issues.apache.org/jira/browse/AIRFLOW-2535?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Tylar Murray updated AIRFLOW-2535:
----------------------------------
Description:
`TriggerDagRunOperator` does not respect the `max_active_runs` setting. This in
itself is not a huge issue, but the scheduler breaks down if `max_active_runs`
is exceeded. Task scheduling becomes absurdly slow, quickly leading to
perpetual DAG buildup.
`TriggerDagRunOperator` could throw exception if `max_active_runs` will be
exceeded, or maybe dags could be put into a `queued` state rather than directly
into `running`?
Alternatively: the scheduler probably shouldn't slow task scheduling when
`max_active_runs` is exceeded.
was:
`TriggerDagRunOperator` does not respect the `max_active_runs` setting. This in
itself is not a huge issue, but the scheduler breaks down if `max_active_runs`
is exceeded. Task scheduling becomes absurdly slow quickly leading to perpetual
DAG buildup.
`TriggerDagRunOperator` could throw exception if `max_active_runs` will be
exceeded, or maybe dags could be put into a `queued` state rather than directly
into `running`?
Alternatively: the scheduler probably shouldn't slow task scheduling when
`max_active_runs` is exceeded.
> TriggerDagRunOperator ignores max_active_runs, leading to scheduler breakdown
> -----------------------------------------------------------------------------
>
> Key: AIRFLOW-2535
> URL: https://issues.apache.org/jira/browse/AIRFLOW-2535
> Project: Apache Airflow
> Issue Type: Bug
> Components: operators, scheduler
> Affects Versions: 1.9.0
> Environment: CeleryExec, Rabbitmq, mysql
> Reporter: Tylar Murray
> Priority: Minor
>
> `TriggerDagRunOperator` does not respect the `max_active_runs` setting. This
> in itself is not a huge issue, but the scheduler breaks down if
> `max_active_runs` is exceeded. Task scheduling becomes absurdly slow, quickly
> leading to perpetual DAG buildup.
> `TriggerDagRunOperator` could throw exception if `max_active_runs` will be
> exceeded, or maybe dags could be put into a `queued` state rather than
> directly into `running`?
> Alternatively: the scheduler probably shouldn't slow task scheduling when
> `max_active_runs` is exceeded.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
