[
https://issues.apache.org/jira/browse/HBASE-1238?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12679504#action_12679504
]
stack commented on HBASE-1238:
------------------------------
+1 on patch.
When committing, please add profuse comments that this is a hack and a better
way needs to be figured. That as is, compactions may be run prematurely.
We need to figure some means of bubbling up the result of HRegion#flushCache up
out of MemcacheFlusher#flushRegion in emergencies. If true, compaction is
wanted. If this came up out of MemcacheFlusher#flushRegion call, then we add
to the flush-later queue you just added.
Good stuff J-D.
> Under upload, region servers are unable to compact when loaded with hundreds
> of regions
> ---------------------------------------------------------------------------------------
>
> Key: HBASE-1238
> URL: https://issues.apache.org/jira/browse/HBASE-1238
> Project: Hadoop HBase
> Issue Type: Bug
> Components: regionserver
> Affects Versions: 0.19.0
> Reporter: Jean-Daniel Cryans
> Priority: Blocker
> Fix For: 0.19.1, 0.20.0
>
> Attachments: hbase-1238.patch
>
>
> We have a situation where each region server is loaded with 100+ regions,
> most of them in the same table. During a long upload of webpages, each
> memcache gets filled near equally fast so that the global memcache limit is
> usually triggered before the max memcache size. Since that emergency flush
> does not trigger compactions, the number of store files just keeps growing
> until it fails on all kinds of errors.
> We need a better story for this as this is a "normal" situation.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
