Mike Drob commented on HBASE-20095:

Yea, there is risk that we are overworking a solution here. Can you estimate 
the risk? Something like, if you set pool size to X and have Y files and each 
clean task takes Z seconds, then we get into bad situation. From there we can 
decide if we should place guardrails on the config, where we can warn that 
something bad is happening, or if we really do need to come up with a better 

Preventing this problem likely involves using {{ForkJoinPool.drainTasksTo}} a 
temporary collection, calling shutdown to block scheduling of new tasks, and 
then requeueing everything.

Taking a step back... why not interrupt immediately? So we have extra files 
that stay around for another cycle... It's a scheduled chore, so it will try 
again shortly, doesn't seem like a huge problem.

> Redesign single instance pool in CleanerChore
> ---------------------------------------------
>                 Key: HBASE-20095
>                 URL: https://issues.apache.org/jira/browse/HBASE-20095
>             Project: HBase
>          Issue Type: Improvement
>            Reporter: Reid Chan
>            Assignee: Reid Chan
>            Priority: Critical
>         Attachments: HBASE-20095.master.001.patch, 
> HBASE-20095.master.002.patch, HBASE-20095.master.003.patch, 
> HBASE-20095.master.004.patch, HBASE-20095.master.005.patch, 
> HBASE-20095.master.006.patch, HBASE-20095.master.007.patch, 
> HBASE-20095.master.008.patch

This message was sent by Atlassian JIRA

Reply via email to