[
https://issues.apache.org/jira/browse/HBASE-30043?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
ASF GitHub Bot updated HBASE-30043:
-----------------------------------
Labels: pull-request-available (was: )
> Truncate procedure with recovery enabled could process non-existent regions
> if run concurrently with merge procedure
> --------------------------------------------------------------------------------------------------------------------
>
> Key: HBASE-30043
> URL: https://issues.apache.org/jira/browse/HBASE-30043
> Project: HBase
> Issue Type: Bug
> Reporter: Dimas Shidqi Parikesit
> Priority: Critical
> Labels: pull-request-available
>
> During truncate recovery, the procedure can release the region lock between
> states. If that region gets merged away in that gap, the truncate procedure
> could still try to unassign/reassign or touch filesystem state for a region
> that no longer exists.
> The proposed fix adds an existence check before MAKE_OFFLINE, centralizes
> region-node lookup in a helper that throws UnknownRegionException when the
> region is gone, and makes rollback only reassign if the region still exists
> and is in an offline/closed state.
> The tests verify two things: truncate fails cleanly instead of recreating
> state or scheduling child procedures, and rollback also does not recreate the
> removed region or queue assigned work.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
