Originally from the User Slack
@Nick_Vladiceanu: Hey there!
We are seeing warnings regarding the Tombstones for system_schema.tables:
querier - Read 157 live rows and 24367 tombstones for system_schema.tables partition key "placeholder" {{1369899501664607956, pk{000f7265636f6d6d656e646174696f6e73}}} (see tombstone_warn_threshold)
with that, we see some degradation in performance, mostly reads become slower. Could that be the cause?
We do create and delete tables quite often. Thanks
@avi: What version are you running? Are you constantly creating and dropping tables?
It shouldn’t affect data-path performance.
@Nick_Vladiceanu: we are currently using version 5.2.18 in Kubernetes. Every 2-3 hours a bunch of new tables get created and old ones deleted (~15 tables).
Any suggestion where to look into? We start observing performance degradation, like slower responses and clients report errors connecting to Scylla
ConnectionException('Host has been marked down or removed')
@avi: 5.2 is long out of support.
Newer versions don’t have that tombstone in schema problem
esp. once you enable raft for schema management
@Nick_Vladiceanu: could that tombstone problem be the reason of read performance degradation?
@avi: It’s unlikely. It slows down driver connection, not not regular reads
Maybe frequent driver connections are slowing down the cluster
@Nick_Vladiceanu: ok, so the tombstone problem might be the reason of
ConnectionException('Host has been marked down or removed')
and as a consequence, this might trigger a reconnect on the driver(s) which might affect the performance of the cluster 
Thanks a lot, I’m looking into what is the next version I could upgrade to. Is there any page with the list of supported versions of OS Scylla?
@Guy: AFAIR, the last two major versions are supported, so that would now be 6.2. You can read more in this FAQ.