[
https://issues.apache.org/jira/browse/SOLR-4191?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13544756#comment-13544756
]
Shawn Heisey commented on SOLR-4191:
------------------------------------
Added -XX:+CMSIncrementalMode. It didn't help. I suppose that's not really
surprising, since it's a ParNew collection. The correlation between these DOWN
events and a very long GC pause is too close to think it's anything else.
{noformat}
haproxy log:
Jan 5 08:20:10 localhost haproxy[12443]: Server idx/idxa1 is DOWN, reason:
Layer7 timeout, check duration: 5001ms. 0 active and 3 backup servers left.
Running on backup. 3 sessions active, 0 requeued, 0 remaining in queue.
jetty exception:
2013-01-05 08:20:14.181:WARN::/solr/ncmain/admin/ping
gc log:
2013-01-05T08:20:01.418-0700: 28482.701: [GC 28482.701: [ParNew:
1887482K->111621K(1887488K), 12.6964980 secs] 5062679K->3492727K(5430212K)
icms_dc=15 , 12.6966660 secs] [Times: user=0.00 sys=18.41, real=12.69 secs]
{noformat}
I'm going to try increasing NewRatio to 3. My max heap is 8GB, NewRatio is
currently 1, but watching the heap graph with jconsole suggests I can get by
with a smaller young generation, which would reduce the cost of ParNew
collections. It'll result in more CMS collections, but perhaps with
incremental mode on, that won't be a huge problem.
On my dev server with java7 and Solr 4.1, I have turned on the G1 collector's
max pause target at 250 milliseconds.
> Exceptions thrown when /admin/mbeans or /admin/ping is accessed during
> update/commit
> ------------------------------------------------------------------------------------
>
> Key: SOLR-4191
> URL: https://issues.apache.org/jira/browse/SOLR-4191
> Project: Solr
> Issue Type: Bug
> Affects Versions: 3.5, 4.1
> Environment: solr-impl 4.1-SNAPSHOT 1421496 - ncindex - 2012-12-13
> 14:56:25
> Reporter: Shawn Heisey
> Fix For: 4.2, 5.0
>
> Attachments: solr-2012-12-14[1].log
>
>
> I am getting the following exceptions in quick succession in the solr log
> when /admin/mbeans is accessed at the moment that an update/commit is
> happening:
> ERROR - 2012-12-13 18:17:01.930; org.apache.solr.common.SolrException;
> null:org.eclipse.jetty.io.EofException
> ERROR - 2012-12-13 18:17:01.982; org.apache.solr.common.SolrException;
> null:org.eclipse.jetty.io.EofException
> WARN - 2012-12-13 18:17:01.984; org.eclipse.jetty.server.Response; Committed
> before 500 {msg=Broken pipe,trace=org.eclipse.jetty.io.EofException
> WARN - 2012-12-13 18:17:01.984; org.eclipse.jetty.servlet.ServletHandler;
> /solr/s0live/admin/mbeans
> java.lang.IllegalStateException: Committed
> I will attach the full solr log. Before SOLR-4135 was fixed, I got a *lot*
> of those exceptions, but these were far less common. Now these appear to be
> the only thing I am getting in my logs, which log4j is logging at WARN.
--
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: [email protected]
For additional commands, e-mail: [email protected]
