[CASSANDRA-15657] Improve zero-copy-streaming containment check by using file sections

• Type: Improvement
• Status: Resolved
• Priority: Normal
• Resolution: Fixed
• Fix Version/s: 4.0-beta1, 4.0
• Component/s: Legacy/Streaming and Messaging
• Change Category: Performance
• Complexity: Normal
• Platform: All

Source Control Link

https://github.com/apache/cassandra/commit/b05fe7ab010218f1fb23b3192e2aea719a9611de

Test and Documentation Plan

CI: https://circleci.com/workflow-run/a36ff018-cbb5-455f-88bb-40bfd1ca4cea
 

• Assignee: Zhao Yang
• Reporter: Zhao Yang

Authors

Zhao Yang

Reviewers

Dinesh Joshi, T Jake Luciani

• Created: 23/Mar/20 15:12
• Updated: 15/May/20 08:48
• Resolved: 18/Apr/20 18:59

Description

Currently zero copy streaming is only enabled for leveled-compaction strategy and it checks if all keys in the sstables are included in the transferred ranges.

This is very inefficient. The containment check can be improved by checking if transferred sections (the transferred file positions) cover entire sstable.

I also enabled ZCS for all compaction strategies since the new containment check is very fast..

Issue Links

• blocks: CASSANDRA-15694 Statistics upon streaming of entire SSTables in Netstats is wrong
• fixes: CASSANDRA-14586 Performant range containment check for SSTables
• is related to: CASSANDRA-15783 test_optimized_primary_range_repair - transient_replication_test.TestTransientReplication