[CASSANDRA-17049] Fix rare NPE caused by batchlog replay / node decomission races

• Type: Bug
• Status: Resolved
• Priority: Low
• Resolution: Fixed
• Fix Version/s: 3.0.26, 3.11.12, 4.0.2, 4.1-alpha1, 4.1
• Component/s: Consistency/Batch Log, Consistency/Hints
• Bug Category: Degradation - Other Exception
• Severity: Low
• Complexity: Normal
• Discovered By: Code Inspection
• Platform: All
• Since Version: 3.0.0

Source Control Link

58c878e3e0356286ac1418e5e52e6f5cadd22ddb

Test and Documentation Plan

Covered by existing unit tests

• Assignee: Aleksey Yeschenko
• Reporter: Aleksey Yeschenko

Authors

Aleksey Yeschenko

Reviewers

Alex Petrov

• Created: 19/Oct/21 14:47
• Updated: 27/May/22 19:24
• Resolved: 12/Nov/21 14:46

Description

Batchlog replay process collects addresses of the hosts that have been hinted to, so it can flush hints for them to disk before confirming deletion of the replayed batches. If a node has been decommissioned during replay, however, when the time comes to flush the hints at the very end of replay, StorageService.getHostIdForEndpoint() will return null for its address, which will, down the line, cause HintsCatalog::get() to be invoked with a null host id argument, causing an NPE.

The simple fix is to check returned host ids for addresses for nulls, and collect hinted host ids instead of hinted addresses.