[
https://issues.apache.org/jira/browse/HBASE-8234?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13956776#comment-13956776
]
Jeffrey Zhong commented on HBASE-8234:
--------------------------------------
This is nice to have thing. We won't loss data when recovering a disabling
table because recovering logic automatically switch back to recovered edits
files for regions of a disabling table.
Introducing the new state will complicate AM logic and still can't get rid of
recovered edit implementations as it's also used in snapshot. Therefore, the
benefits to have a new state isn't that much.
Thanks.
> Introducing "recovering" region state in AM to mark a region in recovering
> status used in distributedLogReplay
> --------------------------------------------------------------------------------------------------------------
>
> Key: HBASE-8234
> URL: https://issues.apache.org/jira/browse/HBASE-8234
> Project: HBase
> Issue Type: Sub-task
> Components: MTTR
> Reporter: Jeffrey Zhong
>
> There are two advantages to have this new "recovering" state in Assignment
> Manager for a region:
> 1) Instead of mark a region "recovering" in ZK, we can consolidate all region
> states in one place and be aware by assignment manager
> 2) When handing disabling table, we have to have this new state so that
> regions of a disabling table can be transitioned into this state for
> recovering.
> Notes:
> In the initial release of distributed log replay, we may not do this subtask
> for simplifications. Without the new state, we still need to create recovered
> edits files for regions of a disabling table. With the new state, we can
> retire recover edits files creation business totally.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
