[jira] [Commented] (SOLR-14278) data loss during live shard split if leader dies

["Yonik Seeley (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SOLR-14278?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17043780#comment-17043780 ] 

Yonik Seeley commented on SOLR-14278:
-------------------------------------

Testing update: I let the test loop overnight with split shard commented out.  There were no failures.  With the split in the test, the failure rate looks somewhere between 30-50% on my hardware. 

> data loss during live shard split if leader dies
> ------------------------------------------------
>
>                 Key: SOLR-14278
>                 URL: https://issues.apache.org/jira/browse/SOLR-14278
>             Project: Solr
>          Issue Type: Bug
>      Security Level: Public(Default Security Level. Issues are Public) 
>            Reporter: Yonik Seeley
>            Priority: Major
>          Time Spent: 10m
>  Remaining Estimate: 0h
>
> While trying to develop better tests for shared storage (SOLR-13101), I ran across another failure for normal replica types as well (one of the first things I do when a test fails for shared storage is to try and validate that normal NRT replicas succeed.)  The PR I'll open has a test adapted from the one in SOLR-13813 for master.
> Scenario:
>   - indexing is happening during shard split
>   - leader is killed shortly after (before the split has finished) and never brought back up
>   - there are often some missing documents at the end
> While it's possible that the simulated killing of the node in the unit test is imperfect, I haven't reproduced a failure if I comment out the split command and just kill the leader.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@lucene.apache.org
For additional commands, e-mail: issues-help@lucene.apache.org