Logging and monitoring

Three ways to see inside a running app

Log levels

• debug: very detailed, for developers. Usually off in production.
• info: normal events. "Order 9001 placed".
• warn: something odd but handled. "Payment retry 2 of 3".
• error: something failed. "Could not reach SMS provider".
• fatal: the app cannot continue and is stopping.

Structured logs and request ids

A structured log is a log line written as JSON with named fields, instead of a free sentence. Machines can search and filter it: "show all error lines for userId 42 in the last hour".

{"level":"error","time":"2026-03-14T08:21:05.120Z","msg":"payment failed","requestId":"req-7f3a","orderId":9001,"provider":"upi","durationMs":5020}

A request id is a unique id given to each incoming request, often sent back in a response header too. Every log line written while handling that request includes it. When a user reports a problem, you search one id and see the whole story.

Metrics that matter

Three numbers tell you most of what you need. Requests per second (traffic): how busy the app is. Error rate: the share of requests that fail, usually counted as 5xx responses. Latency: how long responses take.

For latency, an average hides problems. p95 latency means 95% of requests were faster than this number, and 5% were slower. If p95 is 2 seconds, one user in twenty waits 2 seconds or more, even if the average looks fine. To find it, sort all the times and pick the value at the 95% position.

Health checks, alerts and on-call

app.get('/health', async (req, res) => {
  try {
    await db.query('SELECT 1');           // can we reach the database?
    res.json({ status: 'ok' });
  } catch {
    res.status(503).json({ status: 'error', db: 'down' });
  }
});

An alert is an automatic message (phone call, SMS, chat) sent when a metric crosses a line, for example "error rate above 2% for 5 minutes" or "/health failing". On-call means one team member is responsible for answering alerts during a period, taking turns. Alert only on things that need a human, or people start ignoring alerts.