[
https://issues.apache.org/jira/browse/SOLR-16046?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17534378#comment-17534378
]
Michael Gibney commented on SOLR-16046:
---------------------------------------
I can say that my PR
([apache/solr#842|https://github.com/apache/solr/pull/842]) addresses an issue
that definitely causes thread leaks that are detected and throw errors, and are
not addressed by either Houston's or Kevin's PRs.
I think [apache/solr#841|https://github.com/apache/solr/pull/841] (Kevin's PR)
is good cleanup, and orthogonal to
[apache/solr#842|https://github.com/apache/solr/pull/842]. I can't see what
impact it would have on any thread leak issues though, except perhaps the
reorganization of try/finally? (paradoxically, unless I'm mistaken, moving the
{{killer}} thread into an Executor may actually make that particular aspect
_more_ prone to thread leaks, by making it impossible to explicitly {{join()}}
on the thread! -- and as I pointed out in a comment on the PR, I'm not sure
about the practical impact of running the {{killer}} thread actions in a
separate thread at all, given that we {{start()}} and {{join()}} the thread
before proceeding).
I'm curious about Houston's PR -- I don't understand the implications of
avoiding the Zk reconnect loop in ConnectionManager.
> Thread leak in TestLeaderElectionZkExpiry
> -----------------------------------------
>
> Key: SOLR-16046
> URL: https://issues.apache.org/jira/browse/SOLR-16046
> Project: Solr
> Issue Type: Bug
> Security Level: Public(Default Security Level. Issues are Public)
> Reporter: Houston Putman
> Priority: Major
> Time Spent: 1.5h
> Remaining Estimate: 0h
>
> Currently the {{TestLeaderElectionZkExpiry.testLeaderElectionWithZkExpiry}}
> test fails roughly 11% of the time due to zombie threads.
> These are {{zkConnectionManagerCallback}} threads used by the SolrZkClient.
> In {{ConnectionManager.process()}}, the method being called by the thread, it
> is currently swallowing up an interrupted exception and always looping until
> a valid connection can be made to zookeeper.
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
