[ 
https://issues.apache.org/jira/browse/SOLR-4471?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13583528#comment-13583528
 ] 

Andre Charton commented on SOLR-4471:
-------------------------------------

Today after debugging in a bigger environment I found slave marked the index as 
stale (method isStale() in SnapPuller). This is because segment the file, which 
comes from master and is different in size than the file on slave. This only 
happens if there are enough data and also 'delta-date'. With few data file 
looks same (exists + same size). 
So again: problem is that slave generates this segment file after replication 
(and also a new version). In the end the version number is may a "subproblem" 
of this new segment file. I'm note sure how lucene work, but that looks for me 
as the general problem. 
I also checked solr 3.6.1, the version is same. Also 3.6 didn't create index 
dir with time stamp (eg. index.12345) and segment files are same slave <-> 
master.
                
> slave index version is different form master (full index copy needed)
> ---------------------------------------------------------------------
>
>                 Key: SOLR-4471
>                 URL: https://issues.apache.org/jira/browse/SOLR-4471
>             Project: Solr
>          Issue Type: Bug
>          Components: replication (java)
>    Affects Versions: 4.1
>            Reporter: Andre Charton
>            Assignee: Mark Miller
>              Labels: master, replication, slave, version
>             Fix For: 4.2, 5.0
>
>         Attachments: SOLR-4471.patch
>
>
> Scenario: master/slave replication, master delta index runs every 10 minutes, 
> slave poll interval is 10 sec.
> There was an issue SOLR-4413 - slave reads index from wrong directory, so 
> slave is full copy index from master every time, which is fixed after 
> applying this patch from 4413 (see script below).
> Now on replication the slave downloads only updated files, but slave is 
> create a new segement file and also a new version of index (generation is 
> identical with master). On next polling the slave is download the full index 
> again, because the new version slave is force a full copy.
> Problem is the new version of index on the slave after first replication.
> {noformat:apply patch SOLR-4413 script, please copy patch into patches 
> directory before useage.}
> mkdir work
> cd work
> svn co http://svn.apache.org/repos/asf/lucene/dev/branches/lucene_solr_4_1/
> cd lucene_solr_4_1
> patch -p0 < ../../patches/SOLR-4413.patch
> cd solr
> ant dist
> {noformat}

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org
For additional commands, e-mail: dev-h...@lucene.apache.org

Reply via email to