[
https://issues.apache.org/jira/browse/SOLR-5364?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13805172#comment-13805172
]
Chris commented on SOLR-5364:
-----------------------------
Mark, thanks for taking the time to test this.
It turns out it is a garbage collection issue, I was running the default java
args too, however my machines have 128GB RAM, and had been allocated 20GB heap.
It seems this really needs tuning with Java 7. Since updating my GC collection
settings, I have not had any issues. For reference these are the GC settings
I'm running with:
JAVA_OPTS="$JAVA_OPTS -server -XX:NewRatio=1 -XX:SurvivorRatio=6 \
-XX:+UseConcMarkSweepGC -XX:+CMSIncrementalMode
-XX:CMSIncrementalDutyCycleMin=0 \
-XX:CMSIncrementalDutyCycle=10 -XX:+CMSIncrementalPacing \
-XX:+CMSClassUnloadingEnabled -XX:+DisableExplicitGC \
-XX:ConcGCThreads=10 \
-XX:ParallelGCThreads=10 \
-XX:MaxGCPauseMillis=30000"
I've also set heap to 12g, eden to 5g: -Xmx12g -Xms12g -Xmn5g
> SolrCloud stops accepting updates
> ---------------------------------
>
> Key: SOLR-5364
> URL: https://issues.apache.org/jira/browse/SOLR-5364
> Project: Solr
> Issue Type: Bug
> Components: SolrCloud
> Affects Versions: 4.4, 4.5, 4.6
> Reporter: Chris
> Priority: Blocker
>
> I'm attempting to import data into a SolrCloud cluster. After a certain
> amount of time, the cluster stops accepting updates.
> I have tried numerous suggestions in IRC from Elyorag and others without
> resolve.
> I have had this issue with 4.4, and understood there was a deadlock issue
> fixed in 4.5, which hasn't resolved the issue, neither have the 4.6 snapshots.
> I've tried with Tomcat, various tomcat configuration changes to threading,
> and with Jetty. Tried with various index merging configurations as I
> initially thought there was a deadlock with concurrent merg scheduler,
> however same issue with SerialMergeScheduler.
> The cluster stops accepting updates after some amount of time, this seems to
> vary and is inconsistent. Sometimes I manage to index 400k docs, other times
> ~1million . Querying the cluster continues to work. I can reproduce the
> issue consistently, and is currently blocking our transition to Solr.
> I can provide stack traces, thread dumps, jstack dumps as required.
> Here are two jstacks thus far:
> http://pastebin.com/1ktjBYbf
> http://pastebin.com/8JiQc3rb
> I have got these jstacks from the latest 4.6 snapshot, also running solrj
> snapshot. The issue is also consistently reproducable with BinaryRequest
> writer.
--
This message was sent by Atlassian JIRA
(v6.1#6144)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
