cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jason Brown (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (CASSANDRA-14194) Chain commit log marker potential performance regression in batch commit mode
Date Sat, 27 Jan 2018 18:15:00 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-14194?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16342247#comment-16342247
] 

Jason Brown commented on CASSANDRA-14194:
-----------------------------------------

I've run the head of cassandra-3.11 (currently sha {{b8c12fba064fb0b6a3b6306b2670497434471920}}),
and it completed in the same time that cassandra-3.0 did (no timeout). I ran trunk (sha {{6fc3699adb4f609f78a40888af9797bb205216b0}}),
and it *did* timeout.

I'm running 3.0 and 3.11 in circleci to see if they timeout. If not, then I think the regression
is on trunk alone, and I'll focus my effort there. Obviously, if the source of the source
looks like it'll affect earlier versions, i'll dig into that,as well.

> Chain commit log marker potential performance regression in batch commit mode
> -----------------------------------------------------------------------------
>
>                 Key: CASSANDRA-14194
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-14194
>             Project: Cassandra
>          Issue Type: Bug
>          Components: Core, Testing
>            Reporter: Ariel Weisberg
>            Assignee: Jason Brown
>            Priority: Major
>         Attachments: jstack.txt
>
>
> CASSANDRA-13987 modified how the commit log sync thread works. I noticed that cql3.ViewTest
and ViewBuilderTaskTest have been timing out, but only in CircleCI. When I jstack in CircleCI
what I see is that the commit log writer thread is parked and the threads trying to append
to the commit log are blocked waiting on it.
> I tested the commit before 13987 and it passed in CircleCI and then I tested with 13987
and it timed out. I suspect this may be a general performance regression and not just a CircleCI
issue.
> And this is with write barriers disabled (sync thread doesn't actually sync) so it wasn't
blocked in the filesystem.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@cassandra.apache.org
For additional commands, e-mail: commits-help@cassandra.apache.org


Mime
View raw message