Connection Management and Pooling

Module 1 already said one `MongoClient`. The pool is inside that client: a handful of TCP sockets to the replica set, borrowed per operation. You do not open a socket in the route. You do not connect() per request. This lesson is the knobs, shutdown, and what exhausts the pool.

const client = new MongoClient(process.env.MONGO_URI, {
  maxPoolSize: 20,
  minPoolSize: 2,
  maxIdleTimeMS: 60_000,
  serverSelectionTimeoutMS: 5_000,
});
await client.connect();

process.on("SIGTERM", async () => {
  await client.close();
  process.exit(0);
});

Default maxPoolSize is 100 — often more than a single Node process needs. Too high: you multiply processes × 100 and hit Atlas connection limits. Too low: requests wait in the wait queue (waitQueueTimeoutMS) and fail while MongoDB is fine. serverSelectionTimeoutMS is 'no primary in this many ms', not query time (maxTimeMS is later).

• Long-running server — connect once at boot, close on SIGTERM.
• Serverless / Next.js — cache the client on globalThis so warm invocations reuse the pool (web-apps lesson).
• Tests — one client per worker, or you leak pools.