[CASSANDRA-10384] Create automated performance workloads and metrics collection for improving compaction performance

• Type: Task
• Status: Open
• Priority: Normal
• Resolution: Unresolved
• Component/s: Local/Compaction
• Assignee: Unassigned
• Reporter: Ariel Weisberg
• Created: 22/Sep/15 16:32
• Updated: 16/Apr/19 09:30

Description

This is a high level ticket (Epic or Story) and needs to be decomposed into smaller tasks.

Improving compaction performance is currently a priority. To reduce the risk of choosing what path to take and to allow us to prune paths that don't end up being productive we need to automate measuring the performance of compaction, maintaining a historical log, and collecting detailed metrics that explain why we are seeing what we are seeing.

This relates to some of the work for CASSANDRA-9259 where sub-tthe asks are closely related. Bulk reading appears to be 60% of the CPU time associated with compaction according to flight recorder. There is also almost complete overlap in the interest in automating the collection of metrics that can be use to explain and advocate for pursuing (or abandoning) specific changes.

Issue Links

• relates to: CASSANDRA-9259 Bulk Reading from Cassandra