[
https://issues.apache.org/jira/browse/CASSANDRA-5033?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13611981#comment-13611981
]
Jonathan Ellis commented on CASSANDRA-5033:
-------------------------------------------
I'm not sure I want to handicap the average case (by adding serialize-off-heap)
overhead, for a minor benefit in the worst case (you're so overloaded you're
running out of memory to hold all the callbacks).
I say "minor" because you're still overloaded, which is the real problem.
I suppose maybe you could add code on the write path like CASSANDRA-4705 and
only serialize if it's taking more than Xth percentile?
> Keep intransit mutation tracking off heap
> -----------------------------------------
>
> Key: CASSANDRA-5033
> URL: https://issues.apache.org/jira/browse/CASSANDRA-5033
> Project: Cassandra
> Issue Type: New Feature
> Components: Core
> Affects Versions: 1.1.0
> Reporter: T Jake Luciani
>
> When a node is down and writes are timing out to that node the coordinator
> keeps the mutation in memory and writes out the hint on timeout.
> The problem is under heavy write load these mutations pile up quite a bit
> (when CL is low). It causes a lot of heap pressure and causes a lot of
> problems (which shouldn't happen just because a node is down)
> If we serialized the mutation and put it in a off heap cache vs on head we
> could avoid this problem.
--
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
