[ 
https://issues.apache.org/jira/browse/SOLR-9278?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15552320#comment-15552320
 ] 

Mark Miller commented on SOLR-9278:
-----------------------------------

Commits are accidently logged in the dupe issue FYI: SOLR-9470

Commit ce22c2697c1342fd670e3bac460a53aef90d1d80 in lucene-solr's branch 
refs/heads/master from markrmiller
[ https://git-wip-us.apache.org/repos/asf?p=lucene-solr.git;h=ce22c26 ]
SOLR-9470: Index replication interactions with IndexWriter can cause deadlock. 

Commit 82440f307a211d8f05fe729ec1361bcd1abd0e4e in lucene-solr's branch 
refs/heads/branch_6x from markrmiller
[ https://git-wip-us.apache.org/repos/asf?p=lucene-solr.git;h=82440f3 ]
SOLR-9470: Index replication interactions with IndexWriter can cause deadlock.

> Possible deadlock in replication
> --------------------------------
>
>                 Key: SOLR-9278
>                 URL: https://issues.apache.org/jira/browse/SOLR-9278
>             Project: Solr
>          Issue Type: Bug
>      Security Level: Public(Default Security Level. Issues are Public) 
>          Components: Server
>    Affects Versions: 6.2
>         Environment: Linux
>            Reporter: Xunlong
>            Assignee: Mark Miller
>              Labels: replication
>             Fix For: 6.3, master (7.0)
>
>         Attachments: SOLR-9278.patch
>
>   Original Estimate: 48h
>  Remaining Estimate: 48h
>
> There is a bug in IndexFetcher for replication logic, it may cause deadlock 
> issue, and it's very easy to reproduce. If you change your solrconfig to keep 
> more than 1 commit points, this operation will causes 2 issues:
> 1. Slave has to download whole index directory of Master, instead of 
> incremental udpates only;
> 2. If you click "replicate now" button manually, this is cause deadlock, stop 
> both "indexFetcher" thread and "explicitFetcher" thread.
> The first issue is a design issue, can be worked around by keep only 1 commit 
> point. But the second issue can always happen if there is some file located 
> in slave's index directory, but can not be deleted by index delete policy 
> (due to permission issue etc), I have fixed this issue for my service, would 
> happy to contribute to Solr community to benefit others.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to