fix(NODE-7411): iterate all servers in closeCheckedOutConnections()

[Nepomuk5665]

Summary

Fix a bug in Topology.closeCheckedOutConnections() where only the first server's checked-out connections were being closed.

The Bug

The method had a premature return statement inside the for loop:

closeCheckedOutConnections() {
  for (const server of this.s.servers.values()) {
    return server.closeCheckedOutConnections();  // Returns on first iteration!
  }
}

This caused the method to:

1. Get the first server from the servers Map
2. Call closeCheckedOutConnections() on it
3. Immediately return, skipping all remaining servers

The Fix

Remove the return statement so all servers are iterated:

closeCheckedOutConnections() {
  for (const server of this.s.servers.values()) {
    server.closeCheckedOutConnections();
  }
}

Impact

This bug affects multi-server topologies (replica sets, sharded clusters) where only the first server's connections would be properly closed when MongoClient.close() is called. This could lead to:

• Connection leaks on servers other than the first one
• Resources not being properly cleaned up during client shutdown

[tadjik1]

Hi @Nepomuk5665 thanks so much for the work on this PR. We're currently reviewing this and will get back to you soon.

[tadjik1]

@Nepomuk5665 fantastic job, thanks for pointing out this problem. Could you please modify the existing test inside test/integration/connection-monitoring-and-pooling/connection_pool.test.ts so that it validates the fix as well?

There are currently two issues with the test:

• Metadata: The metadata for the existing ConnectionCheckedInEvent test is set to single. This needs to be either removed or updated to support multi-server deployments.
• Validation: The test only executes an insert. We aren't validating that any operations or connections are being issued against secondaries. It would be better to explicitly set a readPreference to ensure multiple servers are exercised.

[Nepomuk5665]

Thanks for the feedback @tadjik1! I've updated the test with the following changes:

1. Metadata: Removed the topology: 'single' restriction so the test now runs on all topologies (single, replicaset, and sharded)

2. Validation:

   • Added readPreference: 'secondaryPreferred' to the client configuration to ensure operations can be routed to secondaries when available
   • Changed from insert to find operations since reads respect readPreference while writes always go to primary
   • Updated the failpoint to block find commands instead of insert

This ensures that in multi-server deployments, the test will exercise connections to both primary and secondary servers, validating that the fix properly checks in connections across all servers on close.