[
https://issues.apache.org/jira/browse/SOLR-12607?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16565081#comment-16565081
]
Shalin Shekhar Mangar commented on SOLR-12607:
----------------------------------------------
Beasting passed for the tlog replicas but it uncovered yet another failure
(unrelated to tlog replicas) that happens once in 200 runs approximately.
The DistributedUpdateProcessor's doFinish() method relies on the failing node's
shard name to decide whether to return an exception to the client or to
continue. During shard split, we forward updates synchronously to sub shard
leaders and we'd like to return exceptions to clients so that they can retry.
However, due to using the parent shard name in the StdNode created for
sub-shard leader, the comparison between the current replica's shard and
StdNode's shard always passed and the errors were never returned. I've
committed a fix to the branch and I'll add a test to cover this particular
situation.
> Investigate ShardSplitTest failures
> -----------------------------------
>
> Key: SOLR-12607
> URL: https://issues.apache.org/jira/browse/SOLR-12607
> Project: Solr
> Issue Type: Task
> Security Level: Public(Default Security Level. Issues are Public)
> Components: SolrCloud
> Reporter: Shalin Shekhar Mangar
> Assignee: Shalin Shekhar Mangar
> Priority: Major
> Fix For: master (8.0), 7.5
>
>
> There have been many recent ShardSplitTest failures.
> According to http://fucit.org/solr-jenkins-reports/failure-report.html
> {code}
> Class: org.apache.solr.cloud.api.collections.ShardSplitTest
> Method: testSplitWithChaosMonkey
> Failures: 72.32% (81 / 112)
> Class: org.apache.solr.cloud.api.collections.ShardSplitTest
> Method: test
> Failures: 26.79% (30 / 112)
> {code}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org
For additional commands, e-mail: dev-h...@lucene.apache.org
