There's currently three performance tests

• shard_connection_storm_locust
• shard_connection_storm_rate_limited_locust
• shard_overloading_locust

The connection storm workloads were implemented to test the first aspect of the project which is to allow the connection pool to retry a connection after the SSL handshake was aborted.

The shard_overloading_locust might require a bit more effort. We think the amount of load the sharded cluster will be able to process is significantly higher as retargeting happens, so many tests should be run with different amount of loads and limits. To make the load shedding and retargeting happen, the ingress request rate limiter should be enabled at a rate slightly lower than what would overload the server.