The "retrieve" permission is the most used one, and the one that we have focused most of our load testing on. We are seeing response times of multiple seconds, sometimes tens of seconds per request (when we are doing a load test with 7 requests per second). We expect these requests to take a few hundred milliseconds. We have 2 spiced pods running in Kubernetes connected to an RDS Aurora serverless db with 0.5-5 acus. We have enabled tracing, and we can see that most of the time spicedb is waiting for the database. However each request is doing ~65 db queries, and contains about 500 spans, which we see as the main issue