[
https://issues.apache.org/jira/browse/HBASE-30501?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Umesh Kumar Kumawat resolved HBASE-30501.
-----------------------------------------
Fix Version/s: 4.0.0-alpha-1
3.1.0
3.0.1
2.6.8
2.5.17
Release Note: Bulk load no longer deletes the original input HFile after
splitting it across region boundaries. Previously the file was always deleted,
so a retried bulk load could not find it and lost its data. Now only
intermediate split files under .tmp are deleted, as HBASE-15808 intended.
Original HFiles that were split remain in the input directory, and cleaning up
that directory is up to the caller.
Resolution: Resolved
> HBASE-30501 Preserve original HFiles when splitting for bulk load
> -----------------------------------------------------------------
>
> Key: HBASE-30501
> URL: https://issues.apache.org/jira/browse/HBASE-30501
> Project: HBase
> Issue Type: Bug
> Reporter: Ujjawal Kumar
> Assignee: Ujjawal Kumar
> Priority: Major
> Labels: pull-request-available
> Fix For: 4.0.0-alpha-1, 3.1.0, 3.0.1, 2.6.8, 2.5.17
>
>
>
> LoadIncrementalHFiles automatically splits an HFile that crosses region
> boundaries into .bottom and .top files under .tmp and tracks them in an
> in-memory queue. If the invocation
> fails after splitting but before loading—for example, due to the maximum
> HFiles-per-region/family limit—the queue is lost. A new invocation may not
> rediscover .tmp files, while the original HFile may already have been
> deleted.
>
>
>
> Update the retry behavior to preserve the original HFile until the split
> outputs are successfully loaded, or persist and rediscover split outputs
> across invocations. Add tests covering failure after splitting followed by
> a fresh LoadIncrementalHFiles retry.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
