Atlas Cluster connections and memory allocation that can not be explained


We have started experiencing fairly high response times on our M10 cluster (around 1000-1300ms, previously 200-300ms) the past month. This cluster has not been publicly deployed, and the only connections that should have been made to the cluster is from our automated tests and our development team (whose homes are the only whitelisted IPs).

While troubleshooting this using the shard analytics dashboards we found something weird. Even though we haven’t touched the cluster or its connected realm apps for the past month, it has constantly had around 60 connections to it. We tried to restart the cluster by pausing it for 1h, and then resuming it, but the connections were still there.

Furthermore, when looking at the Memory graph, our virtual memory has been around 2-3GB, while our resident memory has not ever exceeded 550MB, which according to the information tab could be because of the amount of open connections to the database.

We’re considering upgrading our cluster to M30, but before we are still interested in knowing what might be causing this. Are these kind of mysterious connections common? Can they affect our response times this much?

Thanks in advance!


These sound like internal connections required for the distributed system to maintain internal availability heartbeats, as well as for MongoDB Atlas’ built in monitoring data