Hey folks, in the past I know that the Azure CosmosDB was removed from consideration for the backing store of SpiceDB because of consistency problems. I was wondering if CITUS extension compatibility (e.g. the schema-based sharding for greater Postgres scalability) was something that you considered? It could largely help with single-region deployments that require higher scalability than a single Postgres instance can provide. I assume row-level partitioning could also be considered, though I assume it might impact relationship evaluation performance more.

i think it's something that we've talked about in passing, but my sense is that because basically all of the traffic is through the

relation_tuple

table, there isn't really a good way to break things down according to the schema. you'd need row-level partitioning, and it would probably need to be bespoke for the shape of your SpiceDB schema. I don't know of anyone who's tried this. one thing that we do know that folks are doing is using read replicas. there's docs here: https://authzed.com/docs/spicedb/concepts/datastores#read-replicas