cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Benedict (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (CASSANDRA-6655) Writing mostly deletes to a Memtable results in undercounting the table's occupancy so it may not flush
Date Wed, 05 Feb 2014 17:00:12 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-6655?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13892298#comment-13892298
] 

Benedict commented on CASSANDRA-6655:
-------------------------------------

bq. and it's currently somewhat breaking DataTacker.isEmpty()).

Good spot.

bq.  I'd change 'isModifiedBy' to something like 'mayModify' 

Sounds reasonable to me. I wanted to keep the logic the same as add() (to make sure it was
super trivial and I didn't screw up) which required something of the form testXBy, and maybeModifiedBy
sounded ugly at the time. But now it sounds no worse.

Want me to repatch, or do you want to ninja it in?

bq. For those following along at home, this only affects deletes through range tombstones
(though full partition deletions were not added to the memtable size, they were accounted
in currentOperations and wouldn't end up not flushing).

Regrettably this would affect them too, if not quite as severely, as it's not just DataTracker.isEmpty()
that's affected: the MeteredFlusher that would never select it for flushing, so it would only
flush on its normal cycle or when the CL ran out of room.

> Writing mostly deletes to a Memtable results in undercounting the table's occupancy so
it may not flush
> -------------------------------------------------------------------------------------------------------
>
>                 Key: CASSANDRA-6655
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-6655
>             Project: Cassandra
>          Issue Type: Improvement
>            Reporter: Benedict
>            Assignee: Benedict
>            Priority: Minor
>             Fix For: 2.0.5, 2.0.6
>
>         Attachments: tmp-2.1.patch, tmp.patch
>
>
> In the extreme case of only deletes the memtable will never flush, and we will OOM.



--
This message was sent by Atlassian JIRA
(v6.1.5#6160)

Mime
View raw message