cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Robert Coli (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (CASSANDRA-2958) Flush memtables on shutdown when durable writes are disabled
Date Fri, 31 Aug 2012 21:50:08 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-2958?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13446416#comment-13446416
] 

Robert Coli commented on CASSANDRA-2958:
----------------------------------------

What I "want" is to be able to explain Cassandra to customers in simple sentences like "a
memtable holds changes until it is full enough to hit a flush condition or you explicitly
flush it" without having to pepper these sentences with caveats like "except if you restart
your node or if you stop it with durable_writes off, triggering an unexpected flush." Patches
such as this one, where vaguely defined ends appear to justify whatever ad-hoc inconsistent
means, do not appear to further this goal.

Let me phrase my objection to this patch in another way...

"What does this patch gain us, and at what cost?"

Your stated goal is to not lose more than "necessary" when stopping a node. It seems your
goal can be achieved without patching, by simply advising "durable_writes off" operators to
run "nodetool drain" when stopping a node. They are, after all, the ones stopping their node
and are perfectly capable of draining it if they do not want to lose the explicitly non-durable
non-durable_writes contents of memtables.

>From what I can tell, the only thing this patch gains us is "people who are running with
durable_writes off don't have to run 'nodetool drain' before stopping nodes."

What we trade for that is the until-now universal expectation that stopping a Cassandra node
never triggers a flush.

Is "the very small group of operators who run with non-durable writes don't have to run 'nodetool
drain'" such a compelling win that we should change a fundamental behavior of Cassandra, making
it less predictable, in order to obtain it? My answer is no.
                
> Flush memtables on shutdown when durable writes are disabled
> ------------------------------------------------------------
>
>                 Key: CASSANDRA-2958
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-2958
>             Project: Cassandra
>          Issue Type: Bug
>    Affects Versions: 0.8.1
>            Reporter: David Phillips
>            Assignee: Jonathan Ellis
>             Fix For: 0.8.3
>
>         Attachments: 2958.txt
>
>
>  

--
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

Mime
View raw message