which is to say that nothing you're doing from a SpiceDB load standpoint seems particularly out of the ordinary, the results you're getting are unexpected, and the thing that's sticking out to me right now as being different is the use of aurora serverless

there are fixes in Bulk checking

you'll need to use HEAD

I should have been more specific when I said "latest version"

(we also publish to quay and gcr depending on where you're pulling from)

(and you can override the

imageName

in the operator if that's how you're using it)

I switched to AWS RDS and one advantage that I found is consistent response due to predictable and fixed instance size. WIth aurora many queries were timed out due to exhausting connections pool, but after several running Aurora scales the instance and gains more connections. I conduct my load testing using different spicedb tags, my results and opinion I posted in this comment https://discord.com/channels/844600078504951838/1350114521594986567/1350809224564834377. So, following my conclusion I think the problem in SpiceDB schema and how many quries we must send to resolve all permissions. Or maybe you conducted load testing against different database, such as Spanner or CockroachDB? I'm wondering maybe queries to Spanner or CockroachDB are way faster and optimized to handle more concurrent requests? Do you have any opinion on this?

it doesn't matter which datastore is used

if SpiceDB has to wait for a connection, it will take longer to process

generally we recommend scaling the number of connections in the pool until the wait time metric goes to 0