cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Ariel Weisberg (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (CASSANDRA-13090) Coalescing strategy sleep too much
Date Fri, 06 Jan 2017 17:06:58 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-13090?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15805011#comment-15805011
] 

Ariel Weisberg commented on CASSANDRA-13090:
--------------------------------------------

* I don't think {{otc_coalescing_max_coalesced_messages}} is useful although I'm interested
in hearing the argument for it. When is pulling more than 128 items out of the LBQ going to
improve performance?
* I think rather for the growing backlog scenario we should skip coalescing entirely and rely
on back pressure from the socket to block the thread. So right now if coalescing says "write
stuff" we should skip coalescing until the backlog has been drained into the socket. The strategy
will need to know about the messages so it can log the arrival rate.
* Granted since the batch size is 128 it will still only add the undesired coalescing every
128 messages and only if the backlog is < {{otc_coaslescing_enough_coalesced_messages}}
* {{otc_coaslescing_enough_coalesced_messages}} in the yaml has a typo
* {{otc_coalescing_window_us}} has a typo in the comment
* Is there an extra drain to in {{coalesceInternal}}? One in the condition and another immediately
after before checking for enough?

> Coalescing strategy sleep too much
> ----------------------------------
>
>                 Key: CASSANDRA-13090
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-13090
>             Project: Cassandra
>          Issue Type: Bug
>          Components: Streaming and Messaging
>            Reporter: Corentin Chary
>             Fix For: 3.x
>
>         Attachments: 0001-Fix-wait-time-coalescing-CASSANDRA-13090.patch
>
>
> With the current code maybeSleep is called even if we managed to take maxItems out of
the backlog. In this case we should really avoid sleeping because it means that backlog is
building up.
> I'll send a patch shortly.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message