Zhiwen Deng created HBASE-30084:
-----------------------------------
Summary: Use a temp directory to avoid creating dirty hfile files
when enable splitToHFile
Key: HBASE-30084
URL: https://issues.apache.org/jira/browse/HBASE-30084
Project: HBase
Issue Type: Bug
Components: regionserver
Affects Versions: 2.7.0, 3.0.0-beta-2, 2.6.6
Reporter: Zhiwen Deng
When we set "hbase.wal.split.to.hfile=true" to accelerate SCP, if the split
WAL's RS crashes unexpectedly, dirty hfiles will be generated, causing Regions
to fail to open.
The core reason is that we directly placed the hfiles in `recovered.hfiles`
instead of first generating them as temp files and then moving them to
`recovered.hfiles`.
I have fixed this issue, and the function now works correctly. Below are some
logs after the fix.
{code:java}
2026-04-15,11:50:50,695 INFO [split-log-closeStream--pool7-t14]
wal.BoundedRecoveredHFilesOutputSink - Renamed recovered hfile
hdfs://test-dev/hbase/test-dev/data/default/ycsb-test/c6c21a44cc79b5252d4765434e665d9b/C/recovered.hfiles.tmp/hbase01.xxxx%2C12700%2C1776224501472.1776225008107/950c180f92374b8dbbc7614892125242
to
hdfs://test-dev/hbase/test-dev/data/default/ycsb-test/c6c21a44cc79b5252d4765434e665d9b/C/recovered.hfiles/950c180f92374b8dbbc7614892125242
2026-04-15,11:50:50,696 INFO [split-log-closeStream--pool7-t14]
wal.BoundedRecoveredHFilesOutputSink - deleted temp dir
hdfs://test-dev/hbase/test-dev/data/default/ycsb-test/c6c21a44cc79b5252d4765434e665d9b/C/recovered.hfiles.tmp/ak-hbase01.xxxx%2C12700%2C1776224501472.1776225008107{code}
Which branch should I submit the pr to fix it?
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
