Jitter Worth The Name

A node holding 40,000 WebSocket clients is restarted. Every client runs this reconnect loop, and the surviving nodes still fall over.

let attempt = 0;

function connect() {
  const ws = new WebSocket(URL);
  ws.onopen = () => { attempt = 0; };
  ws.onclose = () => {
    const delay = Math.min(30000, 250 * 2 ** attempt);
    attempt += 1;
    setTimeout(connect, delay + Math.random() * 100);
  };
}

Which change actually spreads the redial load?

Multiply the existing random term by 10, giving 0 to 1,000 ms of jitter

Draw the whole delay at random: const delay = Math.random() * Math.min(30000, 250 * 2 ** attempt); and pass that to setTimeout with no additive term

Raise the cap from 30,000 ms to 120,000 ms

Reset attempt only after the socket has stayed open for 30 seconds, not in onopen