[
https://issues.apache.org/jira/browse/HBASE-14636?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14964472#comment-14964472
]
Anoop Sam John commented on HBASE-14636:
----------------------------------------
bq.What about closeCheckInterval? We are already on an interval checking for a
close. Could we do the shipped in here on this same interval? Or at least unite
the two time-based checks?
closeCheckInterval -> This is not really time based check. This is size written
based check. Default is 10 MB. After this much sized cells written, we check
whether store is still writable. Use this same for shipped call also? 2
concerns
1. This config can be 0 and then we won't be doing the shipped call at all. We
don't want this to happen.
2. The def 10 MB seems too high.
I first tried with a size based check only. Then just changed to time based as
it looked simple wrt #ops also.
If we go with size based I would suggest the size limit for check to be some
thing like
HConstants.DEFAULT_BLOCKSIZE * conf.get(min_files_for_compact) -> So by def 64
KB * 3 so that there is chance for at least one block will get cleared from
prevBlocks list.
This make sense?
We will need another long variable and add cell sizes to that taking some more
ops to be done. Not a big deal any way. I can do this if all agree. And size
based check is the best way IMO
> Clear HFileScannerImpl#prevBlocks in between Compaction flow
> ------------------------------------------------------------
>
> Key: HBASE-14636
> URL: https://issues.apache.org/jira/browse/HBASE-14636
> Project: HBase
> Issue Type: Sub-task
> Components: regionserver, Scanners
> Reporter: Anoop Sam John
> Assignee: Anoop Sam John
> Priority: Blocker
> Fix For: 2.0.0
>
> Attachments: HBASE-14636.patch, HBASE-14636.patch
>
>
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
