[
https://issues.apache.org/jira/browse/HBASE-3970?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13050587#comment-13050587
]
Subbu M Iyer commented on HBASE-3970:
-------------------------------------
Here is how this patch addresses this issue:
When a master dies during meta migration, we will have some rows of
META.CatalogFamily updated with new HRI, (i.e HRI with out HTD) and some still
hanging with legacy HRI. (i.e HRI with HTD). When the backup master/ or
fresh start of master attempts the migration again, it will encounter some rows
of META already updated with new HRI and some still legacy. This patch adds
additional logic during HRegionInfo deserialization to ignore the cases where
HRI is missing a HTD (i.e already migrated to new HRI) and attempts to migrate
rest of the rows.
> Address HMaster crash/failure half way through meta migration
> -------------------------------------------------------------
>
> Key: HBASE-3970
> URL: https://issues.apache.org/jira/browse/HBASE-3970
> Project: HBase
> Issue Type: Bug
> Components: client, master, regionserver
> Affects Versions: 0.92.0
> Reporter: Subbu M Iyer
> Assignee: Subbu M Iyer
> Priority: Blocker
> Attachments:
> HBASE-3970-First_Draft_-_Address_HMaster_crash_failure_half_way_through_meta_migration1.patch
>
> Original Estimate: 168h
> Remaining Estimate: 168h
>
> When HMaster tries to migrate (after HBASE-451 goes live) the old HRI (with
> HTD) to new HRI (with out HTD) and if the Master or the migration process
> crashes/fails midway, it will leave the .META. in a corrupt state and may not
> allow successful cluster startup.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
