[CASSANDRA-18734] SAI result retriever is filtering too many rows

• Type: Bug
• Status: Resolved
• Priority: Normal
• Resolution: Fixed
• Fix Version/s: 5.0-alpha2, 5.0, 5.1
• Component/s: Feature/SAI
• Epic Link: Storage Attached Indexes (Phase 2)
• Bug Category: Degradation - Performance Bug/Regression
• Severity: Critical
• Complexity: Challenging
• Discovered By: Performance Regression Test
• Platform: All
• Since Version: 5.0-alpha

Source Control Link

https://github.com/apache/cassandra/commit/9ce86e0ff8b6344b528a0640f9dafa23f97dd85a

Test and Documentation Plan

Current circle CI runs are in the first comment for 5.0 and trunk

• Assignee: Mike Adamson
• Reporter: Mike Adamson

Authors

Mike Adamson

Reviewers

Andres de la Peña, Caleb Rackliffe

• Tester: Mike Adamson
• Created: 08/Aug/23 12:51
• Updated: 12/Sep/23 12:57
• Resolved: 07/Sep/23 10:59

Description

Performance tests on SAI have shown that the number of rows being filtered for wide row partitions is more than expected. For a 10k row/partition test - limit 10, the following has been observed.

This is not the expected outcome of row-awareness and needs investigating. The number of rows read should, roughly, match the number of partitions read.