It looks like the HA monitoring in mongodb-core is only performed on the servers that are in the replica set when the driver first connects. Servers that get added to the replica set later are not monitored. I think the monitoring should cover all available servers.

This caught us by surprise recently when we replaced some members of a replica set, and connections to the new servers kept timing out because there was no monitoring traffic to keep them active.

The issue can be reproduced with a simple client like the following:

1. Run the client with the URL of a replica set.
2. After the client has connected, add another secondary to the replica set.
3. Check the serverHeartbeatStarted messages to see if the new secondary is being monitored.

#!/usr/bin/env node
'use strict';

const MongoClient = require('mongodb').MongoClient;

const SDAM_EVENTS = [
  'serverOpening',
  'serverClosed',
  'serverDescriptionChanged',
  'topologyOpening',
  'topologyClosed',
  'topologyDescriptionChanged',
  'serverHeartbeatStarted',
  'serverHeartbeatSucceeded',
  'serverHeartbeatFailed'
];


const url = process.argv[2];
if (!url) {
  console.error('Usage: tiny_mongo_client.js mongodb://url');
  process.exit(1);
}

(new MongoClient()).connect(url).then((db) => {
  console.info(`Connected to ${url}`);
  SDAM_EVENTS.forEach((name) => {
    db.topology.on(name, (event) => { console.warn(`${new Date()} ${name}`, event); });
  });
});

With driver versions 2.2.25 and 2.2.26, no serverHeartbeatStarted messages are logged for the new secondary. Instead, the connection times out and is reopened every 30 seconds. I have not tested older versions to compare the behavior.

I've also seen odd behavior where if you add and remove a secondary, and then add a different secondary, the driver may wait 10 seconds after the timeout before reopening the connection. But I think that should be less of a concern if the connection doesn't time out.