[CASSANDRA-16944] Single partition reads can read more SSTables than required

• Type: Bug
• Status: Resolved
• Priority: Normal
• Resolution: Fixed
• Fix Version/s: 3.0.26, 3.11.12, 4.0.2
• Component/s: Legacy/Local Write-Read Paths
• Bug Category: Code
• Severity: Low
• Complexity: Normal
• Discovered By: Code Inspection
• Platform: All
• Since Version: 3.0.0

Source Control Link

https://github.com/apache/cassandra/commit/9f492844d5401d4d856c3bf5b908ba84d1b92a0c

Test and Documentation Plan

The patches add new unit tests and distributed tests. It also update existing unit tests.

• Assignee: Benjamin Lerer
• Reporter: Benjamin Lerer

Authors

Benjamin Lerer

Reviewers

Caleb Rackliffe

• Created: 10/Sep/21 13:08
• Updated: 16/Mar/22 17:02
• Resolved: 28/Sep/21 16:14

Description

For some scenarios involving row deletions, range deletions or static columns, the logic of SinglePartitionReadCommand.queryMemtableAndSSTablesInTimestampOrder might trigger more SSTables reads that expected.

For row deletions and range deletions the reasons is that the logic do not take them into account. Once we hit a deleted row (caused by a row deletion or a range deletion) with a timestamp higher than the one of the next SStable we know that we can stop reading more SSTables.

For static columns the problems seems to have been introduced by the changes in CASSANDRA-16671.