[HBASE-28081] Snapshot working dir does not retain ACLs after snapshot commit phase

• Type: Bug
• Status: Resolved
• Priority: Blocker
• Resolution: Fixed
• Fix Version/s: 2.6.0, 2.4.18, 2.5.6, 3.0.0-beta-1
• Component/s: acl, test
• Hadoop Flags: Reviewed
• Assignee: Viraj Jasani
• Reporter: Duo Zhang
• Created: 13/Sep/23 10:26
• Updated: 17/Oct/23 11:24
• Resolved: 30/Sep/23 18:49

Description

TestSnapshotScannerHDFSAclController is failing 100% on flaky dashboard.

 

After snapshot is committed from working dir to final destination (under /.hbase-snapshot dir), if the operation was atomic rename, the working dir (e.g. /hbase/.hbase-snapshot/.tmp) no longer preserves ACLs that were derived from snapshot parent dir (e.g. /hbase/.hbase-snapshot) while creating first working snapshot dir. Hence, for new working dir, we should ensure that we preserve ACLs from snapshot parent dir.

This would ensure that final snapshot commit dir has the expected ACLs regardless of whether we perform atomic rename of non-atomic copy operation in the snapshot commit phase.

 

Issue Links

• is broken by: HBASE-28042 Snapshot corruptions due to non-atomic rename within same filesystem
• links to: GitHub Pull Request #5437