Kamil Braun comments

Results 302 comments of


                                            Kamil Braun

perf_simple_query - tps and tasks_per_op regression

Thanks @patjed41. Given that the regression is much less significant than we initially thought, and we don't know how to avoid it, I'm closing the issue.

cql/describe: hide cdc log tables

If the table can be ALTERed then we should emit it. Since emitting a CREATE is wrong -- because creating the base table with `cdc: enabled` will automatically create the...

cql/describe: hide cdc log tables

https://github.com/scylladb/scylladb/pull/18467#issuecomment-2102281994 @temichus @fruch The PR from Patryk is queued on next but not yet in master. Maybe that's the reason? Maybe there's a difference between what branches the different dtest...

cql/describe: hide cdc log tables

> https://github.com/scylladb/scylladb/pull/18467#issuecomment-2102281994 > @temichus @fruch The PR from Patryk is queued on next but not yet in master. Maybe that's the reason? Maybe there's a difference between what branches the...

`nodetool removenode` may fail if we concurrently try to do `DROP INDEX...` resulting in leaving a dead DB node as part of the DB cluster

The fact that a failed removenode spoils the rest of the SCT run because it tried to do something else to the dead node (which was supposed to be removed)...

`nodetool removenode` may fail if we concurrently try to do `DROP INDEX...` resulting in leaving a dead DB node as part of the DB cluster

Did the DROP INDEX succeed?

`nodetool removenode` may fail if we concurrently try to do `DROP INDEX...` resulting in leaving a dead DB node as part of the DB cluster

Failure happened here: ``` Apr 13 05:20:48.453347 longevity-parallel-topology-schema--db-node-4bdf0ea1-3 scylla[7325]: [shard 0:strm] node_ops - removenode[e603c4dc-685c-435e-a343-f125e4581815]: Node 10.4.8.169 failed removenode_sync_data verb: seastar::rpc::remote_verb_error ({shard 0: std::runtime_error (repair[7c1d569e-55dc-44df-a701-44028be68f69]: 2 out of 576 ranges failed,...

`nodetool removenode` may fail if we concurrently try to do `DROP INDEX...` resulting in leaving a dead DB node as part of the DB cluster

Nodes indeed are trying to sync schema for some time: ``` Apr 13 05:20:44.263074 longevity-parallel-topology-schema--db-node-4bdf0ea1-3 scylla[7325]: [shard 1:strm] migration_manager - Pulling schema from 10.4.8.169:0 Apr 13 05:20:44.270270 longevity-parallel-topology-schema--db-node-4bdf0ea1-3 scylla[7325]: [shard...

`nodetool removenode` may fail if we concurrently try to do `DROP INDEX...` resulting in leaving a dead DB node as part of the DB cluster

Judging by `Pulling schema from` we're calling `migration_manager::sync_schema` But this should not do a pull in Raft mode, it should do a read barrier ```cpp future migration_manager::sync_schema(const replica::database& db, const...

`nodetool removenode` may fail if we concurrently try to do `DROP INDEX...` resulting in leaving a dead DB node as part of the DB cluster

Ok, this: ```cpp if (schema_map.empty()) { co_return; } if (!_enable_schema_pulls) { co_await _group0_barrier.trigger(_as); co_return; } ``` was added in c6c83474939. The run used 0be61e51d329 which was before