Marcin Maliszkiewicz

Actually when I was testing schema was the only stable thing (dependant on seed)

This should have been fixed since scylla-2022.1.0 by https://github.com/scylladb/scylladb/commit/a51529dd1547fe029e8c259f8b03891106f6edc0

> > > Ah, it is because the patch set uses write_mutations command to write to the table and this command is unguarded (does not work with read-modify-write). > >...

> I think this commit: > > > cql3: auth: add a way to create mutations without executing > > is the next best thing after sliced bread. If we...

> So thinking about it a bit, why did we have a system_auth keyspace in the first place? I think it's because we wanted to manage its replication strategy from...

v2: - comment fixes/additions - lifetime management around data_value_or_unset (now moving vector) - minor bugs (removed extra static, etc) - lost quorum test added Now this branch should pass all...

> Please adjust the TODO list at least. > I will > What does the lack of abort_source support mean? Problems in shutdown? That can't be deferred. Duno, it travels...

> > > Please adjust the TODO list at least. > > > > > > I will > > > What does the lack of abort_source support mean? Problems...

> I just noticed that the necessary modifications to snapshot transfer code are missing - see Kamil's comment on the service levels PR ([#16585 (review)](https://github.com/scylladb/scylladb/pull/16585#pullrequestreview-1800490842)) which had the same issue....

v3: - updated TODOs - fixed problem with passing data_value_or_unset across shards (has shared_ptr inside) - make one default_authorizer::modify a coroutine to avoid uaf on query string - rebase