You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
k8ssandra forces you to use RF of 3 (per DC) for system_auth, reaper_db, etc. This is problematic for two reasons:
With larger clusters, it's a best practice to increase system_auth RF. I discovered this week that the operator will undo my RF changes.
The reaper_db doesn't always need to be replicated elsewhere. It's hard coded to use SERIAL CL, which causes instability when LWT queries time out across remote DCs.
I'm not sure what the appropriate configuration should be here. I can think of a few ways to approach it.
Use the same config as user keyspaces. If the option isn't supplied, use the default.
A simplified version where you can simply override the RF per DC. This would be as simple as supplying the keyspace name and the RF per-dc.
A hybrid, where you can tell the default RF in each DC then override per-dc.
I think I lean towards option 2 or 3. Curious what you think. Happy to contribute a patch once we decide on what the config should look like.
┆Issue is synchronized with this Jira Story by Unito
┆Issue Number: K8OP-309
The text was updated successfully, but these errors were encountered:
k8ssandra forces you to use RF of 3 (per DC) for system_auth, reaper_db, etc. This is problematic for two reasons:
I'm not sure what the appropriate configuration should be here. I can think of a few ways to approach it.
I think I lean towards option 2 or 3. Curious what you think. Happy to contribute a patch once we decide on what the config should look like.
┆Issue is synchronized with this Jira Story by Unito
┆Issue Number: K8OP-309
The text was updated successfully, but these errors were encountered: