[
https://issues.apache.org/jira/browse/SOLR-6683?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14249826#comment-14249826
]
Forest Soup commented on SOLR-6683:
-----------------------------------
I applied the patch for SOLR-6359 on 4.7 and did some test. It does not work as
expected.
When I set below config, it still go into SnapPuller code even if I only newly
added 800 doc.
<updateLog>
<str name="dir">${solr.ulog.dir:}</str>
<int name="numRecordsToKeep">10000</int>
<int name="maxNumLogsToKeep">100</int>
</updateLog>
After my reading code, it seems that lines in
org.apache.solr.update.PeerSync.handleVersions(ShardResponse srsp) cause the
issue:
if (ourHighThreshold < otherLow) {
// Small overlap between version windows and ours is older
// This means that we might miss updates if we attempted to use this
method.
// Since there exists just one replica that is so much newer, we must
// fail the sync.
log.info(msg() + " Our versions are too old.
ourHighThreshold="+ourHighThreshold + " otherLowThreshold="+otherLow);
return false;
}
Could you please comment? Thanks!
> Need a configurable parameter to control the doc number between peersync and
> the snapshot pull recovery
> -------------------------------------------------------------------------------------------------------
>
> Key: SOLR-6683
> URL: https://issues.apache.org/jira/browse/SOLR-6683
> Project: Solr
> Issue Type: Bug
> Components: replication (java)
> Affects Versions: 4.7
> Environment: Redhat Linux 64bit
> Reporter: Forest Soup
> Priority: Critical
> Labels: performance
>
> If there are >100 docs gap between the recovering node and the good node, the
> solr will do snap pull recovery instead of peersync.
> Can the 100 docs be configurable? For example, there can be 10000, 1000, or
> 10 docs gap between the good node and the node to recover.
> For 100 doc, a regular restart of a solr node will trigger a full recovery,
> which is a huge impact to the performance of the running systems
> Thanks!
--
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
