[
https://issues.apache.org/jira/browse/HBASE-15055?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15115140#comment-15115140
]
Anoop Sam John commented on HBASE-15055:
----------------------------------------
As ExploringCompactionPolicy is the default policy now we have to fix
HBASE-14409 to get this fix visible.
{code}
// Compute the total size of files that will
// have to be read if this set of files is compacted.
long size = getTotalStoreSize(potentialMatchFiles);
// Store the smallest set of files. This stored set of files will be
used
// if it looks like the algorithm is stuck.
if (mightBeStuck && size < smallestSize) {
smallest = potentialMatchFiles;
smallestSize = size;
}
if (size > comConf.getMaxCompactSize(mayUseOffPeak)) {
continue;
}
{code}
Here the sum of file's size is checked against the max compaction size.
When the system decides it is periodic major compaction (after the specified
time elapsed), we should do the compaction of all files. Or else when it will
get compacted?
For new policy like, DateBasedPolicy, this might not stand. But for the current
policies.... Comments?
Ping [[email protected]] as HBASE-14409 is with him.
> Major compaction is not triggered when both of TTL and
> hbase.hstore.compaction.max.size are set
> -----------------------------------------------------------------------------------------------
>
> Key: HBASE-15055
> URL: https://issues.apache.org/jira/browse/HBASE-15055
> Project: HBase
> Issue Type: Bug
> Reporter: Eungsop Yoo
> Assignee: Eungsop Yoo
> Priority: Minor
> Attachments: HBASE-15055-v1.patch, HBASE-15055-v10.patch,
> HBASE-15055-v11.patch, HBASE-15055-v2.patch, HBASE-15055-v3.patch,
> HBASE-15055-v4.patch, HBASE-15055-v5.patch, HBASE-15055-v6.patch,
> HBASE-15055-v7.patch, HBASE-15055-v8.patch, HBASE-15055-v9.patch,
> HBASE-15055.patch
>
>
> Some large files may be skipped by hbase.hstore.compaction.max.size in
> candidate selection. It causes skipping of major compaction. So the TTL
> expired records are still remained in the disks and keep consuming disks.
> To resolve this issue, I suggest that to skip large files only if there is no
> TTL expired record.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
