[
https://issues.apache.org/jira/browse/FLINK-34906?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Rui Fan updated FLINK-34906:
----------------------------
Description:
Currently, the autoscaler will scale a job when the JobStatus is RUNNING. But
the JobStatus will be RUNNING once job starts schedule, so it doesn't mean all
tasks are running. Especially, when the resource isn't enough or job recovers
from large state.
The autoscaler will throw exception and generate the AutoscalerError event when
tasks are not ready, such as:
!image-2024-03-21-17-40-23-523.png!
Also, we don't need to scale it when some tasks are not ready.
Solution: we only scale job that all tasks are running(some of tasks may be
finished).
was:
Currently, the autoscaler will scale a job when the JobStatus is RUNNING. But
the JobStatus will be RUNNING once job starts schedule, so it doesn't mean all
tasks are running. Especially, when the resource isn't enough or job recovers
from large state.
The autoscaler will throw exception and generate the AutoscalerError event when
tasks are not ready, such as:
!image-2024-03-21-17-40-23-523.png!
Solution: we only scale job that all tasks are running(some of tasks may be
finished).
> Don't start autoscaling when some tasks are not running
> -------------------------------------------------------
>
> Key: FLINK-34906
> URL: https://issues.apache.org/jira/browse/FLINK-34906
> Project: Flink
> Issue Type: Improvement
> Components: Autoscaler
> Reporter: Rui Fan
> Assignee: Rui Fan
> Priority: Major
> Fix For: 1.9.0
>
> Attachments: image-2024-03-21-17-40-23-523.png
>
>
> Currently, the autoscaler will scale a job when the JobStatus is RUNNING. But
> the JobStatus will be RUNNING once job starts schedule, so it doesn't mean
> all tasks are running. Especially, when the resource isn't enough or job
> recovers from large state.
> The autoscaler will throw exception and generate the AutoscalerError event
> when tasks are not ready, such as:
> !image-2024-03-21-17-40-23-523.png!
> Also, we don't need to scale it when some tasks are not ready.
> Solution: we only scale job that all tasks are running(some of tasks may be
> finished).
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
