MongoDB connection pooling and performance tuning in Node.js

Why connection pooling matters

Establishing a new MongoDB TCP connection takes 20–100 ms and allocates resources on both the client and server. Under load, creating a connection per request would saturate MongoDB's connection limit within seconds. The MongoClient maintains a pool of pre-established, reusable connections — each request borrows one for its operation and immediately returns it, with minimal overhead after pool warm-up.

// WRONG — new connection per request, pools bypassed entirely
app.get('/products', async (req, res) => {
  const client = new MongoClient(uri)  // new TCP connection every time
  await client.connect()
  const data = await client.db('shop').collection('products').find().toArray()
  await client.close()
  res.json(data)
})

// CORRECT — singleton client created at startup, shared across all requests
let _db = null
export async function getDB() {
  if (_db) return _db
  const client = await MongoClient.connect(process.env.MONGO_URI, {
    maxPoolSize: 10,               // max concurrent connections
    minPoolSize: 2,                // keep 2 alive to reduce cold-start latency
    serverSelectionTimeoutMS: 5000,
    socketTimeoutMS: 45000
  })
  _db = client.db('shopDB')
  return _db
}

app.get('/products', async (req, res) => {
  const db = await getDB()  // instant — returns cached db reference
  res.json(await db.collection('products').find().toArray())
})

Start with maxPoolSize: 10. Increase it only after profiling shows requests are waiting in the pool queue under real production load.