[
https://issues.apache.org/jira/browse/AURORA-1240?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Bill Farner updated AURORA-1240:
--------------------------------
Component/s: (was: Client)
Description:
The UpdateConfig {{restart_theshold}} \[1\] setting does not appear to deliver
much user value as it's highly sensitive to scheduling performance and may
result in aborted/rolled back job updates when set too low.
Some background: This timeout controls task transition from {{PENDING}} to
{{RUNNING}} during the job update. In the event of cluster capacity shortage,
assigning a task to a host may take considerably longer thus expiring the
timeout and depending on the failure settings causing an unnecessary job update
abort or rollback. It was meant to give users some protection against
unsatisfiable resource/constraint requirements. In reality though, it proved to
be rather an annoyance to users when an update is interrupted due to unexpected
delay in task assignment.
Consider deprecating and subsequently removing this setting.
This ticket tracks a first step to ignore this value in the scheduler updater.
See linked tickets for follow-up work.
\[1\] -
https://github.com/apache/aurora/blob/master/docs/configuration-reference.md#updateconfig-objects
was:
The UpdateConfig {{restart_theshold}} \[1\] setting does not appear to deliver
much user value as it's highly sensitive to scheduling performance and may
result in aborted/rolled back job updates when set too low.
Some background: This timeout controls task transition from {{PENDING}} to
{{RUNNING}} during the job update. In the event of cluster capacity shortage,
assigning a task to a host may take considerably longer thus expiring the
timeout and depending on the failure settings causing an unnecessary job update
abort or rollback. It was meant to give users some protection against
unsatisfiable resource/constraint requirements. In reality though, it proved to
be rather an annoyance to users when an update is interrupted due to unexpected
delay in task assignment.
Consider deprecating and subsequently removing this setting.
\[1\] -
https://github.com/apache/aurora/blob/master/docs/configuration-reference.md#updateconfig-objects
Summary: Ignore JobUpdateSettings.maxWaitToInstanceRunningMs in the
scheduler (was: Deprecate UpdateConfig "restart_threshold" setting)
> Ignore JobUpdateSettings.maxWaitToInstanceRunningMs in the scheduler
> --------------------------------------------------------------------
>
> Key: AURORA-1240
> URL: https://issues.apache.org/jira/browse/AURORA-1240
> Project: Aurora
> Issue Type: Task
> Components: Scheduler
> Reporter: Maxim Khutornenko
> Assignee: Bill Farner
>
> The UpdateConfig {{restart_theshold}} \[1\] setting does not appear to
> deliver much user value as it's highly sensitive to scheduling performance
> and may result in aborted/rolled back job updates when set too low.
> Some background: This timeout controls task transition from {{PENDING}} to
> {{RUNNING}} during the job update. In the event of cluster capacity shortage,
> assigning a task to a host may take considerably longer thus expiring the
> timeout and depending on the failure settings causing an unnecessary job
> update abort or rollback. It was meant to give users some protection against
> unsatisfiable resource/constraint requirements. In reality though, it proved
> to be rather an annoyance to users when an update is interrupted due to
> unexpected delay in task assignment.
> Consider deprecating and subsequently removing this setting.
> This ticket tracks a first step to ignore this value in the scheduler
> updater. See linked tickets for follow-up work.
> \[1\] -
> https://github.com/apache/aurora/blob/master/docs/configuration-reference.md#updateconfig-objects
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
