Context

When running the driver test test_stale_errors_are_ignored with serverless, it fails as an extra PoolClearedEvent is created (see here). This is because two different service_ids are created for the same connection as we have three Serverless Proxies running for our drivers MTM group (to confirm this, you can go to the Envoy Manager page in our Admin UI and search for groupId 654c0a390ae7b6753906bc56). As we see here a new service_id is generated for each Serverless Proxy process. The reason why the test passed in non gRPC situations is because service_ids used to be generated in the Atlas Proxy. And since we turned on the feature flag SINGLE_TARGET_SERVERLESS_DEPLOYMENT, all connections will go to one atlas proxy (maintaining the assumption that connections to the same instance will give us the same service id).

Definition of done

A potential solution would be to only run one Serverless Proxy for the drivers testing MTM. However this would not reflect the reality for customers as we will have multiple serverless proxies running for each MTM group in production. The definition of done would be for the Drivers team to decide what is the best path forward regarding this test.

Pitfalls

N/A