Under heavy load, using a context with timeout in a find operation, if timeout is reached, derives in a weird behavior of driver connection pool.
Connections starts to be discarded by timeout and pool fall in a snow ball of discard/reconnect attemps until driver panics. 
In tests running about 2 minutes, we observe, using a pool monitor, about 27k opened  and 26k closed connections, with a minPoolSize of 30 and a maxPoolSize of 100.

To be clear, this is a sniped of code tested:

filter := bson.M{"store_id": bson.M{"$in": idsList}}
ctx, cancel := context.WithTimeout(context.Background(), 500*time.Millisecond)
defer cancel()

fOptions := options.Find().
   SetMaxTime(500*time.Millisecond).
   SetBatchSize(int32(len(idsList)))

records, err := m.storesCollection.Find(ctx, filter, fOptions)

where:
idList is a variable length slice of ids to search (indexed field, unique)
ctx: Problematic context with timeout
fOptions–>MaxTime: works ok

Two files attached:
1- Test result with metrics and error messages
2- console output with traces