[
https://issues.apache.org/jira/browse/HBASE-616?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12650079#action_12650079
]
stack commented on HBASE-616:
-----------------------------
Would be interesting to figure if a long-running GC was going on concurrently?
If so, perhaps as suggested by Rong-en a while back, the concurrent low-pause
collector might work better when under load:
http://java.sun.com/javase/technologies/hotspot/gc/gc_tuning_6.html#icms.available_options.
> " We slept XXXXXX ms, ten times longer than scheduled: 3000" happens
> frequently.
> --------------------------------------------------------------------------------
>
> Key: HBASE-616
> URL: https://issues.apache.org/jira/browse/HBASE-616
> Project: Hadoop HBase
> Issue Type: Bug
> Reporter: stack
>
> Just saw the below in a log... all in a row on the one server.
> {code}
> 4493 2008-05-05 18:08:17,512 WARN org.apache.hadoop.hbase.util.Sleeper: We
> slept 34557ms, ten times longer than scheduled: 3000
> 4494 2008-05-05 18:11:08,879 WARN org.apache.hadoop.hbase.util.Sleeper: We
> slept 30576ms, ten times longer than scheduled: 3000
> 4495 2008-05-05 18:30:45,056 WARN org.apache.hadoop.hbase.util.Sleeper: We
> slept 1091720ms, ten times longer than scheduled: 3000
> 4496 2008-05-05 18:30:45,056 WARN org.apache.hadoop.hbase.util.Sleeper: We
> slept 1094209ms, ten times longer than scheduled: 10000
> 4497 2008-05-05 18:30:45,429 FATAL org.apache.hadoop.hbase.HRegionServer:
> unable to report to master for 1092093 milliseconds - aborting server
> {code}
> We're seeing these kinda outages pretty frequently. In the case above, it
> was small cluster that was using TableReduce to insert. The MR, HDFS and
> HBase were all running on same nodes.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
