[
https://issues.apache.org/jira/browse/KUDU-3121?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17106880#comment-17106880
]
Andrew Wong commented on KUDU-3121:
-----------------------------------
Yeah, they're similar ideas, but with different motivations (I think). As I
understand it, the MM table prioritization in KUDU-2824 helps define
performance characteristics based on the table. A tool to pick the next best op
would be more of a one-time use tool, in case the scheduling algorithm is stuck
picking operations that have unintended results.
With KUDU-2929, KUDU-3002, and KUDU-1400 fixed, I don't know that there are
more cases like these, so I'm not sure how valuable such a tool would be. But
having such a tool would have mitigated those problems, and let users work
around the issues without having to upgrade to the fixed versions. That said,
_without_ such a tool, users are encouraged to upgrade to versions with the
fixes, so I don't imagine this tool would be very high priority :).
> Allow users to pick the next best op
> ------------------------------------
>
> Key: KUDU-3121
> URL: https://issues.apache.org/jira/browse/KUDU-3121
> Project: Kudu
> Issue Type: New Feature
> Components: ops-tooling
> Reporter: Andrew Wong
> Priority: Major
>
> Time and again, we'll see a case where the maintenance manager scheduler
> thread is, for whatever reason, scheduling an op that is actually not that
> helpful. KUDU-2929, KUDU-3002, and KUDU-1400 come to mind.
> It might be convenient in some cases to temporarily (maybe for a single round
> of scheduling) give a specific tablet or op priority.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
