Tomasz Grabiec

If this Fixes #16411, we need to open another bug for truncate being unsafe with repair.

> > New in v3: > > > > * Changed `storage_service::get_erm_when_table_has_no_migrations` to allow vnode tables. This way we avoid situation with truncate table concurrently with bootstrap. Truncate table will...

> > > > New in v3: > > > > > > > > * Changed `storage_service::get_erm_when_table_has_no_migrations` to allow vnode tables. This way we avoid situation with truncate table...

> * Tomasz Grabiec ***@***.***> [24/09/18 23:31]: > The split seems a bit arbitrary to me. For example, why is read_schema_for_keyspaces() in schema_applier.cc, and other schema reading methods still in...

> @tgrabiec can tablets logic cause a timeout of local queries? I don't see how. Maybe tablet migration overloads the disk for some reason.

> > Also, regarding `service::abstract_read_executor::make_digest_requests` above, does that mean this coming from read repair? > > No. Why would we repair local table. Local read wouldn't send a digest request,...

Maybe the read fails on replica side and we just don't see the cause of failure because it's not logged or propagated?

> > Maybe the read fails on replica side and we just don't see the cause of failure because it's not logged or propagated? > > This is a read...

> There is a lot of traces there. Not all of them are the same. I also found this one in handle_tablet_split_finalization: What's the timestamp of that one? It's strange,...

> Just a guess > > Maybe a call to "storage_service/quiesce_topology" can be doing those reads ? This call doesn't do any reads.