[jira] [Updated] (AURORA-1600) Job updates with large count of instance overrides halt scheduler perf

["Maxim Khutornenko (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/AURORA-1600?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Maxim Khutornenko updated AURORA-1600:
--------------------------------------
    Sprint: Twitter Aurora Q1'16 Sprint 18

> Job updates with large count of instance overrides halt scheduler perf
> ----------------------------------------------------------------------
>
>                 Key: AURORA-1600
>                 URL: https://issues.apache.org/jira/browse/AURORA-1600
>             Project: Aurora
>          Issue Type: Bug
>          Components: Scheduler
>            Reporter: Maxim Khutornenko
>            Assignee: Maxim Khutornenko
>            Priority: Critical
>             Fix For: 0.12.0
>
>
> We have observed a case when a user update with a large number of specified instance overrides (updateOnlyTheseInstances) results in significant performance deterioration to the extent of scheduler processing almost no offers and not scheduling any pending tasks for long periods (minutes to hours). 
> The culprit appears to be the {{selectInstructions}} query. It's unacceptably slow when number of instanceConfigs and/or instance overrides approaches 100. Since it's called inside a write lock to guide individual instance updates, nothing else can proceed including status updates and offer activities. 
> I was able to replicate this in jmh. Fix is incoming.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)