[CASSANDRA-15627] sstable not in the corresponding level in the leveled manifest

• Type: Bug
• Status: Open
• Priority: Normal
• Resolution: Unresolved
• Component/s: Local/Compaction, Local/Compaction/LCS
• Bug Category: Correctness - API / Semantic Implementation
• Severity: Normal
• Complexity: Normal
• Discovered By: Workload Replay
• Platform: All
• Assignee: Unassigned
• Reporter: David Capwell
• Created: 10/Mar/20 01:40
• Updated: 10/Mar/20 01:41

Description

I get the following warning logs when running smoke tests

Live sstable /cassandra/d1/data/ks/table-cce7c54b5abf3f369bb7659a74e9e963/mf-71-big-Data.db from level 0 is not on corresponding level in the leveled manifest. This is not a problem per se, but may indicate an orphaned sstable due to a failed compaction not cleaned up properly.

There are no other warning logs and no error logs; so compaction doesn’t have anything saying there was a failure.

Schema

CREATE TABLE ks.table (
  pk1 ascii,
  pk2 bigint,
  ck1 ascii,
  ck2 ascii,
  ck3 ascii,
  v1 int,
  v2 ascii, 
  PRIMARY KEY ((pk1,pk2), ck1, ck2, ck3)
) WITH comment = 'test table'
  AND gc_grace_seconds = 1
  AND memtable_flush_period_in_ms = 100
  AND compression = {'class': 'LZ4Compressor'}
  AND compaction = {'class': 'LeveledCompactionStrategy', 'only_purge_repaired_tombstones': true}
  AND CLUSTERING ORDER BY (ck1 DESC,ck2 ASC,ck3 DESC);

test

• run simulated queries for 30 minutes
• run incremental repair in a loop (once one completes run the next)