[PHOENIX-5473] Index write failures during index rebuilds should not change index table state

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 4.15.0, 5.1.0, 4.14.3
• Assignee: Kadir OZDEMIR
• Reporter: Kadir OZDEMIR
• Created: 12/Sep/19 01:02
• Updated: 21/Dec/19 00:55
• Resolved: 17/Sep/19 17:29

Description

With PHOENIX-5018, index full rebuilds are done within UngroupedAggregateRegionObserver coprocessors like index partial rebuilds. Index partial rebuilds have been only used for mutable tables. This means that UngroupedAggregateRegionObserver had been only used for rebuilding mutable tables. With PHOENIX-5018, UngroupedAggregateRegionObserver is now used for both partial and full rebuilds and both immutable and mutable index tables. When an index replay write initiated by UngroupedAggregateRegionObserver fails, the failure is handled by PhoenixIndexFailurePolicy which disables the index table if the index write still fails after it is tried the configured number of times. PhoenixIndexFailurePolicy is also used by the Phoenix client code. However, PhoenixIndexFailurePolicy had been only applied to mutable tables and index write failures never disabled an immutable index table. This behavior has changed with PHOENIX-5018 since UngroupedAggregateRegionObserver is also used to rebuild immutable index tables. Index rebuild failures should not disable immutable tables (and mutable tables).

Issue Links

• links to: GitHub Pull Request #586