[
https://issues.apache.org/jira/browse/SOLR-6678?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14191590#comment-14191590
]
Ramkumar Aiyengar commented on SOLR-6678:
-----------------------------------------
But if this indeed is the problem, then that just exposes the issue that these
hooks are not being GC'd? Either the unloaded core is hanging around or the
hooks are being accumulated by the reloaded core?
> Collection/core reload is causing a memory leak
> -----------------------------------------------
>
> Key: SOLR-6678
> URL: https://issues.apache.org/jira/browse/SOLR-6678
> Project: Solr
> Issue Type: Bug
> Affects Versions: 4.10
> Reporter: Alexey Serba
> Attachments: ReloadMemoryLeak.png
>
>
> I have a use case where I need to periodically
> [reload|https://cwiki.apache.org/confluence/display/solr/Collections+API#CollectionsAPI-api2]
> a SolrCloud collection. Recently I did ~1k reload operations and noticed
> that the cluster was running slower and slower, so I connected to it with
> jconsole and noticed that heap was growing with every reload operation,
> forcing GC wasn't helping.
> So I took a heap dump and noticed that I have too many SolrCore-s hanging
> around.
> It's hard for me to grok the root cause of this, but maybe someone more
> knowledgable in Solr internals can figure it out by looking into this GC root
> path (see attached image)? If I interpret this correctly, it looks like one
> SolrCore is referencing another SolrCore through SolrSuggester. Maybe close
> hook for SolrSuggester component doesn't release everything that it should be
> releasing (like SolrSuggester.dictionary)?
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
