Originally from the User Slack
@Nick_Vladiceanu: Hi folks,
we are seeing quite a lot of warnings regarding Tombstones in system_schema keyspace, Scylla 5.4.9:
WARN 2025-05-05 10:12:02,070 [shard 0:stat] querier - Read 210 live rows and 4109 tombstones for system_schema.tables <partition-range-scan> (-inf, +inf) (see tombstone_warn_threshold)
WARN 2025-05-05 10:11:10,929 [shard 0:stat] querier - Read 288 live rows and 10003 tombstones for system_schema.columns <partition-range-scan> (-inf, +inf) (see tombstone_warn_threshold)
As far as I understand, this can hurt the cluster performance. Can we do something to clean up those tombstones faster? Since the gc_grace_seconds = 604800 and it’s immutable for this keyspace, what are the options? Thanks
@Botond_Dénes: Major compaction on the keyspace can help.
These tables are not read all the time, only on startup and on schema changes.
@Nick_Vladiceanu: I have tried to execute a few times
nodetool compact system_schema
Although it runs without issues, the warnings about tombstones are still present 
We create and remove tables every 2-3 hours
@Botond_Dénes: In a recent version (don’t remember which exactly) we added a different method to purge tombstones, this would allow getting rid of most of them.
> We create and remove tables every 2-3 hours
With this use-case you will have many tombstones, no way to avoid it.
The above mentioned improvement will help though.
You can look around in the release notes of later releases (6.0+) to see which one has this improvement. In any case I suggest upgrading, 5.4 is EOL for a while now.
@Nick_Vladiceanu: Thanks a lot Botond. We have recently upgraded to 5.4 as preparation for the 6.x upgrade, which will follow soon. I will post here updates after the upgrade to 6.x.