[
https://issues.apache.org/jira/browse/CASSANDRA-2327?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13006784#comment-13006784
]
Jonathan Ellis edited comment on CASSANDRA-2327 at 3/15/11 3:48 AM:
--------------------------------------------------------------------
>From the comments to Table.maybeSwitchMemtable:
{noformat}
/*
* If we can get the writelock, that means no new updates can come in
and
* all ongoing updates to memtables have completed. We can get the tail
* of the log and use it as the starting position for log replay on
recovery.
*
* This is why we Table.flusherLock needs to be global instead of
per-Table:
* we need to schedule discardCompletedSegments calls in the same order
as their
* contexts (commitlog position) were read, even though the flush
executor
* is multithreaded.
*/
{noformat}
was (Author: jbellis):
From the comments to Table.maybeSwitchMemtable:
{noformat}
/*
* If we can get the writelock, that means no new updates can come in
and
* all ongoing updates to memtables have completed. We can get the tail
* of the log and use it as the starting position for log replay on
recovery.
*
* This is why we Table.flusherLock needs to be global instead of
per-Table:
* we need to schedule discardCompletedSegments calls in the same order
as their
* contexts (commitlog position) were read, even though the flush
executor
* is multithreaded.
*/
{noforma
> Table.flusherLock is static final.. remove static.
> --------------------------------------------------
>
> Key: CASSANDRA-2327
> URL: https://issues.apache.org/jira/browse/CASSANDRA-2327
> Project: Cassandra
> Issue Type: Bug
> Components: Core
> Affects Versions: 0.7.4, 0.7.5
> Environment: JVM
> Reporter: Vijay
> Assignee: Vijay
> Priority: Minor
> Fix For: 0.7.5
>
>
> I see read and write latency spike when the system tables are
> flushing(according to opscentral)....
> Only reason which i can come-up with is probably because of the
> Table.flusherLock is static final... i think it should not be static because
> the flush is per keyspace and this lock will lock all the read operations
> because one table is going through a flush.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
