2 / 4 WebSocket, SSE & Polling

Everybody Back at Once

About 200,000 WebSocket clients are spread over several nodes. One node is restarted, its 40,000 clients all detect the close and immediately redial, and the surviving nodes fall over under the handshake load. What reconnect policy prevents the cascade?

A

A longer idle timeout on the load balancer, so the surviving connections are not reaped while the restarted node comes back into rotation

B

Exponential backoff with random jitter, so retry attempts are spread across a widening window instead of arriving together

C

A fixed five second retry interval for every client, which caps the handshake rate at a predictable value

D

Immediate reconnect capped at three attempts, after which the client falls back to short polling on a fixed two second timer