[
https://issues.apache.org/jira/browse/CASSANDRA-2170?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13107233#comment-13107233
]
Peter Schuller commented on CASSANDRA-2170:
-------------------------------------------
Wow, interesting. Are you sure it's 0.8.5 though? The stack trace is not
matching what I see in the 0.8.5 tag (mismatched line number for
MessagingService.addCallback()).
We've been seeing load spikes on 0.7, but havent reported it because it's such
an old version. However we were never able to grab stacks because no JMX query
would ever succeed during this condition.
The stack trace indicates it's stuck doing resize operations on the NBHM where
each thread is trying to help the resizing operation along by performing
potentially duplicate (for forward progress producing) work.
Do you have a list of all stacks? Do you find any thread (should be 0 or 1)
that are executing in ExpiringMap.CacheMonitor.run() at the time of the load
spikes?
I guess we're seeing some kind of fallen-and-cant-get-up senario having to do
with the resize. Maybe dogpiling the resize is making it overall slow enough
that it never gets unstuck without a temporary stop in incoming requests. Or
some such. That's gut feely speculation without having actually looked at it
carefully, so take it with a grain of salt :)
> Load spikes
> -----------
>
> Key: CASSANDRA-2170
> URL: https://issues.apache.org/jira/browse/CASSANDRA-2170
> Project: Cassandra
> Issue Type: Bug
> Affects Versions: 0.6.11
> Reporter: Jonathan Ellis
> Assignee: Brandon Williams
>
> as reported on CASSANDRA-2058, some users are still seeing load spikes on
> 0.6.11, even with fairly low-volume read workloads.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
