[TEPHRA-43] TransactionVisibilityFilter can prematurely drop delete markers

• Type: Bug
• Status: Closed
• Priority: Blocker
• Resolution: Fixed
• Affects Version/s: 0.3.0
• Fix Version/s: 0.3.1
• Component/s: hbase-compat-0.96, hbase-compat-0.98
• Assignee: Gary Helmling
• Reporter: Gary Helmling
• Created: 05/Nov/14 20:45
• Updated: 14/Jul/16 05:52
• Resolved: 20/Nov/14 22:30

Description

In TransactionVisibilityFilter, we have logic to interpret empty values as delete markers, and drop any subsequent versions for the deleted column. When this is applied to a normal client region scan, this works correctly. However, there is a problem when this is applied in some cases on flush and compaction.

During compaction, we can only drop the delete markers themselves when performing a major compaction. On flush and minor compaction, the delete markers must be retained, otherwise older cells from the same column, which are present in other store files not included in the compaction, may later become visible again due to the removal of the delete markers.

Issue Links

• is related to: TEPHRA-42 Coprocessor to clear deleted data in tephra