I haven't configured the connection pool limits, so they are running on the default values (--datastore-conn-pool-read-max-open=20). Since we are running 2 ECS tasks, that gives us a total of 40 read connections across the cluster. With 100 concurrent read-heavy requests like LookupResources, it makes perfect sense that we are exhausting the pool and hitting long connection wait times. This also explains why our Aurora instance metrics look completely fine! I will increase the pool limits to better handle our concurrency and re-run the benchmark. By the way, just to get your gut feeling: assuming we tune the configurations properly, do you think our current ECS specs (2 tasks with 2 vCPU / 4GB each) should generally be sufficient to handle this level of load, or would you still expect us to need more compute? Thank you so much for pointing this out!