Extend, enhance and regularly cover tombstones handling in SCT workloads
Opened this issue · 2 comments
mykaul commented
- Add mixed insert/update/delete/read workload, not mixed update/read. I don't think cassandra-stress is prepared for that, and I don't think our other tools are either (do they even perform validation on read?)
- Add correctness check (see above) to ensure there's no data ressurection
- Performance tests - at least stress testing with many tombstones
- ICS + TWCS
The focus should be on tombstone_gc=repair, but not exclusively.
pehala commented
We need to extend existing SCT tests to have batchlog writes in the workload
- From scylladb/scylladb#20260
bhalevy commented
- Add mixed insert/update/delete/read workload, not mixed update/read. I don't think cassandra-stress is prepared for that, and I don't think our other tools are either (do they even perform validation on read?)
The deletes mix should contain: partition deletes, single row deletes, and (clustering) range deletions.
Also, add TTL-based expiration to the mix.