[
https://issues.apache.org/jira/browse/SOLR-5436?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13828404#comment-13828404
]
Mark Miller commented on SOLR-5436:
-----------------------------------
bq. The Overseer went into a GC pause or something and others nodes assume it
is down and re-elected another OverSeer.
This will only happen if our ephemeral overseer leader node goes down - which
means we lost our connection to zookeeper. If we lose our connection to
zookeeper, the queue thread will exit trying to talk to zk. If we reconnect to
zookeeper before the queue loop has a chance to fail (some kind of wicked quick
flap), we will stop the overseer threads on getting in line to be the Overseer
again. Also, if zk tells us the overseer leader went down, we stop any overseer
threads we might have.
I think its about as strong as the poll.
> Eliminate the 1500ms wait in overseer loop
> ------------------------------------------
>
> Key: SOLR-5436
> URL: https://issues.apache.org/jira/browse/SOLR-5436
> Project: Solr
> Issue Type: Improvement
> Reporter: Noble Paul
> Assignee: Noble Paul
> Priority: Minor
> Attachments: SOLR-5436.patch, SOLR-5436.patch, SOLR-5436.patch,
> SOLR-5436.patch, SOLR-5436.patch, SOLR-5436.patch, SOLR-5436.patch
>
>
> The Overseer thread waits 1500 ms before it polls for new events. The wait
> should be eliminated and it should just wait for new events till they come
> the way it is done in OverseerCollectionProcessor
--
This message was sent by Atlassian JIRA
(v6.1#6144)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
