[
https://issues.apache.org/jira/browse/CASSANDRA-6042?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13774084#comment-13774084
]
Russell Alexander Spitzer commented on CASSANDRA-6042:
------------------------------------------------------
I've done a preliminary patch for this. I've added a variable to set at what
amount of tombstones should start being logged (DEBUG LEVEL). I want to add a
JMX handle for changing this on the fly but I'm not sure what a good place for
it would be (StorageProxy?) . I've also added a histogram per CF for tracking
tombstones in general. My current solution is not ideal as it only can check
the last row scanned. This should be sufficient for all queries which specify a
row but will miss out on queries which request multiple rows.
Also I'd love for some advice on how to write a Unit-test for the JMX
interaction.
> Add WARN when there are a lot of tombstones in a query
> ------------------------------------------------------
>
> Key: CASSANDRA-6042
> URL: https://issues.apache.org/jira/browse/CASSANDRA-6042
> Project: Cassandra
> Issue Type: Improvement
> Reporter: Jeremiah Jordan
> Assignee: Russell Alexander Spitzer
> Priority: Minor
> Fix For: 1.2.11
>
> Attachments: 0001-JMX-and-Debug-Messages-for-Max-Tombstone-Scans.patch
>
>
> Now that we count the number of tombstones hit (so it can go in tracing), can
> we pick some threshold (or make it configurable with 0 being don't warn), and
> spit out a warning saying "Just went through 10000 tombstones in partition
> XYZ".
> Right now if you are having GC problems because some row got a bunch of
> tombstones, you can turn on server side tracing, and hope the bad query gets
> in there, or you can keep making heap dumps, dig through them, and hope you
> catch the query in there.
> I have seen code problems at multiple places causing this same issue (some
> code causing way more tombstones than it should, for just one row). And it
> is a PITA+Luck to debug it right now.
--
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
