[CASSANDRA-10343] Erroneous partition deletion events are delivered to indexers

• Type: Bug
• Status: Resolved
• Priority: Normal
• Resolution: Fixed
• Fix Version/s: 3.0.0 rc1
• Component/s: Legacy/CQL, Legacy/Local Write-Read Paths
• Severity: Normal
• Assignee: Sam Tunnicliffe
• Reporter: Sam Tunnicliffe

Authors

Sam Tunnicliffe

Reviewers

Sylvain Lebresne

• Created: 15/Sep/15 15:01
• Updated: 16/Apr/19 09:30
• Resolved: 16/Sep/15 08:43

Description

In AtomicBTreePartition#addAllWithSizeDelta we check update.deletionInfo().isLive() to determine whether the PartitionDeletion should be passed to the supplied IndexTransaction. This is incorrect and it results in the PartitionDeletion being passed on if either a partition delete or any range tombstones are present in the update.

We need to change:

  if (!update.deletionInfo().isLive())

to

  if (!update.deletionInfo().getPartitionDeletion().isLive())

Note that the built-in indexes don't currently do anything with the deletionTime and so aren't affected, but any custom implementation which didn't examine it properly might remove data from its index where it shouldn't