[
https://issues.apache.org/jira/browse/HBASE-12405?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14351321#comment-14351321
]
zhangduo commented on HBASE-12405:
----------------------------------
[~stack] I think the test plan is good.
And I'm a little worried about 'HBASE-11569 Flush / Compaction handling from
secondary region replicas'. HRegion.replayWALFlushCommitMarker also plays with
maxFlushedSeqId, and they just use flush sequence number as maxFlushedSeqId.
Perhaps this is not right when per column family flush is turned on since we
may not always flush all stores.
Thanks~
> WAL accounting by Store
> -----------------------
>
> Key: HBASE-12405
> URL: https://issues.apache.org/jira/browse/HBASE-12405
> Project: HBase
> Issue Type: Improvement
> Components: wal
> Affects Versions: 2.0.0, 1.1.0
> Reporter: zhangduo
> Assignee: zhangduo
> Fix For: 2.0.0, 1.1.0
>
> Attachments: HBASE-12405.patch, HBASE-12405_1.patch,
> HBASE-12405_2.patch, HBASE-12405_3.patch, HBASE-12405_4.patch
>
>
> HBASE-10201 has made flush decisions per Store, but has not done enough work
> on HLog, so there are two problems:
> 1. We record minSeqId both in HRegion and FSHLog, which is a duplication.
> 2. There maybe holes in WAL accounting.
> For example, assume family A with sequence id 1 and 3, family B with
> seqId 2. If we flush family A, we can only record that WAL before sequence id
> 1 can be removed safely. If we do a replay at this point, sequence id 3 will
> also be replayed which is unnecessary.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
