Problems with mongoDB statefulsets in Kubernetes

NodeBB version

v3.7.4

NodeBB git hash

No response

NodeJS version

v20.19.4

Installed NodeBB plugins

When a member of the mongodb replicaset is not available, nodebb fails to start and enters in crashloops. Both are happening with both connection string types so mongodb:// or mongodb+srv:// :

mongodb://nodebb:XXX@mongodb-0.mongodb-replicaset-svc,nodebb-stack-mongo-1.mongodb-replicaset-svc,mongo-2.mongodb-replicaset-svc/nodebb?tls=false&replicaSet=rs0&readPreference=primaryPreferred

I removed a pod from mongodb replicaset (2/3), and restarted a nodebb pod and it fails to start. Once I add it back, nodebb starts correctly.
I also simulated a node down when testing the connection string using mongosh from an other pod and it can connect even if a member does not exist. (The service is a headless one.):

mongodb://nodebb:XXX@mongodb-0.mongodb-replicaset-svc,nodebb-stack-mongo-1.mongodb-replicaset-svc,mongo-2.mongodb-replicaset-svc,mongo-3.mongodb-replicaset-svc/nodebb?tls=false&replicaSet=rs0&readPreference=primaryPreferred

For me this looks like a bug in the application or the mongodb driver, but it is quite problematic since that is not resilient to failure. I tried changing connectTimeoutMS=100000&serverSelectionTimeoutMS=100000 to the connection string but it does not help.

It could be a misconfiguration on the nodebb side but I would suspect the driver

Database type

No response

Database version

No response

Exact steps to cause this issue

No response

What you expected

No response

What happened instead

No response

Anything else?

No response

Reproduce the failure by restarting NodeBB while one MongoDB replica-set member is unavailable, using the connection strings and timeout settings from the report. Start by inspecting NodeBB's MongoDB connection handling and driver configuration; done means NodeBB starts and remains resilient when a replica-set member is down.