[
https://issues.apache.org/jira/browse/SOLR-9554?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15525428#comment-15525428
]
Alan Woodward commented on SOLR-9554:
-------------------------------------
Another failure caused by this:
https://jenkins.thetaphi.de/job/Lucene-Solr-master-Solaris/869/consoleFull
In this case, the race is as follows:
* core 1 finds that there's no managed-schema
* core 2 finds that there's no managed-schema
* core 1 loads schema.xml, upgrades to managed-schema, and moves schema.xml to
schema.xml.bak
* core 2 tries to load schema.xml and crashes with a NullPointerException
I've tried writing a test case to hammer on this, but it refuses to fail -
possibly it needs to be running on a slow enough machine? In any case, I think
we need some kind of distributed lock when doing the fallback-and-upgrade
manoeuvre.
> Multiple ManagedIndexSchemaFactory upgrades running simultaneously can clash,
> causing cores not to load
> -------------------------------------------------------------------------------------------------------
>
> Key: SOLR-9554
> URL: https://issues.apache.org/jira/browse/SOLR-9554
> Project: Solr
> Issue Type: Bug
> Security Level: Public(Default Security Level. Issues are Public)
> Reporter: Alan Woodward
>
> If a collection is created using a configset with a ManagedSchemaFactory but
> no managed-schema file, then multiple cores may try and convert the schema
> file simultaneously.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org
For additional commands, e-mail: dev-h...@lucene.apache.org
