Hernan Gelaf-Romer created HBASE-29239:
------------------------------------------
Summary: Subsequent runs of re-splitting HFiles can fail because
we don't cleanup the MR directory
Key: HBASE-29239
URL: https://issues.apache.org/jira/browse/HBASE-29239
Project: HBase
Issue Type: Bug
Components: backup&restore
Reporter: Hernan Gelaf-Romer
Recently, I've discovered that it's possible to break backups if the
MapReduceHFileSplitterJob fail to run successfully.
The MR job first populates the target directory, but fails at some other point.
When re-tried, the MR fails because the target directory already exists. We
should clean up the target bulkload directory if the HFileSplitter job fails
for any reason to avoid this failure mode
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
