[ujjawal4046]

Summary

Backports HBASE-30501 to branch-2.

Preserve original top-level HFiles after splitting so a failed bulk load can be retried by a fresh process. Only intermediate split inputs already under .tmp are deleted after being split again.

This is a manual adaptation because branch-2 still implements the logic in LoadIncrementalHFiles, while the master change was made in BulkLoadHFilesTool.

Upstream master PR: #8819

Test

mvn -pl hbase-server -PrunLargeTests -Dhadoop.profile=3.0 \
  -Dwarbucks.skip=true \
  -Dtest=TestLoadIncrementalHFilesSplitRecovery#testFreshRetryAfterSplitFailure \
  -Dcheckstyle.skip=true -Dspotless.check.skip=true test

Result: 1 test run, 0 failures, 0 errors, 0 skipped.

[Umeshkumar9414]

I have cherry-picked it to branch-2.6 and branch-2.5 as well.
On a side note, If possible please refrain from using force push when addressing comments, it is not recommanded.

[Umeshkumar9414]

Thanks @ujjawal4046 for your contribution.