Details
-
Improvement
-
Status: Resolved
-
Normal
-
Resolution: Fixed
-
None
-
Operability
-
Challenging
-
All
-
None
-
Description
While ultimately we could serve to have a real scheduler and single entry point for compactions, for now some moderate progress seems in order. There are cases where, during expansions, you can end up in a state where you can't finish upgrading sstables as they keep getting aborted by Incremental Repair and other operations that ultimately should be lower priority and not have the ability to cancel these operations.
The idea here is that tasks like upgradestables and cleanup (and other tasks that have the same priority) will get serialised since markAllCompacting is synchronising on data, and higher priority tasks will be able to still cancel others via runWithCompactionsDisabled.
Attachments
Issue Links
- causes
-
CASSANDRA-18635 Test failure: org.apache.cassandra.distributed.test.UpgradeSSTablesTest
- Resolved
- fixes
-
CASSANDRA-17890 Support high PRIORITY thread pool for use trriged compaction
- Resolved