[CASSANDRA-15179] multiple validation compactions on single table

• Type: Bug
• Status: Triage Needed
• Priority: Normal
• Resolution: Unresolved
• Component/s: Consistency/Repair
• Platform: All
• Assignee: Unassigned
• Reporter: Nathan Jackels
• Created: 22/Jun/19 00:36
• Updated: 22/Jun/19 00:36

Description

In a 2 node cluster I'm observing that after a repair is started, any repairs started leaves validation compactions running on the coordinator node.

The non-coordinator correctly rejects the repair with

ActiveRepairService.java:525 - Cannot start multiple repair sessions over the same sstables

but the coordinator did not. When testing to confirm the behavior I was able to increase the executing validations from 11 to ~18 running on the same table and an additional 17 [compactions] pending with high confidence they were pending validations. (see attached scrubbed files).

Repairs were started with: nodetool repair – keyspace_01 table_02

Initially I thought this might be fixed with CASSANDRA-14332, but that is a revert of changes from CASSANDRA-13797 which was released in 3.0.15 and I'm running 3.0.10 so shouldn't be affected. So far in the ~1.5 hours since noticing this I haven't been able to recreate the issue.

I'm will soon be validating 3.0.18 and will retest with that release if I can figure out how to recreate.

Environment Info:

• Cassandra 3.0.10
• 10.0.0.1: Windows 10 / 4C8T / 16 GB DDR4 / JRE 1.8.191
• 10.0.0.2: Windows 7 / 4C8T / 16 GB DDR3 / JRE 1.8.191