[jira] [Commented] (CASSANDRA-6117) Avoid death-by-tombstone by default

["Sylvain Lebresne (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/CASSANDRA-6117?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13782776#comment-13782776 ] 

Sylvain Lebresne commented on CASSANDRA-6117:
---------------------------------------------

I think I don't really like that idea. It's perfectly possible to reach the tombstone threshold in production without meaning to: maybe some users has a temporary surge of delete activity, or maybe compaction has got a bit behind temporarily and hasn't GCed tombstones as fast as it should. And it's perfectly possible for that to happen without having a real a problem. And chances are, that kind of situation won't happen while testing and users won't have particular protected their code against that situation. Failing in those cases seems plain wrong to me.  

I understand that many users fall into the trap of models that inserts a lot of tombstones, and I agree that we should help preventing getting there as much as we can. But it seems to me that issuing warning when we suspect such workload is as far as we should go. Note that I wouldn't be against lowering the default warning threshold for instance, if we want to make sure people that do create models that uses lot of tombstone get an early warning and can change their model.

At the very least, I don't think we should do that in 2.0 (but really I just think we should keep it to the warning).


> Avoid death-by-tombstone by default
> -----------------------------------
>
>                 Key: CASSANDRA-6117
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-6117
>             Project: Cassandra
>          Issue Type: New Feature
>          Components: Core
>            Reporter: Jonathan Ellis
>            Assignee: Jonathan Ellis
>             Fix For: 2.0.2
>
>         Attachments: 6117.txt
>
>
> We added warnings to 1.2 (CASSANDRA-6042); for 2.0 we should go farther and drop requests (with an error logged) that exceed the threshold.  Users who want to tread dangerously are free to crank the threshold up, but "I queried a lot of tombstones and Cassandra fell over" is possibly the number one way of killing Cassandra nodes right now.



--
This message was sent by Atlassian JIRA
(v6.1#6144)