WebSocket keep alive

WebSocket keep-alive behavior is not only ping traffic. A correct client also needs reconnect, resubscribe, and state rebuild paths after the socket changes.

Basic Ping Loop

const keepAlive = setInterval(() => {
  if (ws.readyState === WebSocket.OPEN) {
    ws.send(JSON.stringify({ op: "ping" }));
  }
}, 25_000);

Clear the interval when the socket closes permanently or the application stops the stream. Do not let old intervals continue after reconnecting.

Keep-Alive Model

Concern	Client behavior
Liveness	Send periodic ping messages only when the socket is open
Backoff	Reconnect with capped exponential backoff and jitter
Subscription restore	Resubscribe after the new socket opens, using tracked active subscriptions
Local state	Treat reconnect as a state transition that may require a fresh snapshot
Gap events	Mark local state unsafe until the gap is resolved or output is rebuilt
Observability	Log close code, reason, retry count, channel set, and restored symbols

Timer And Retry Guardrails

Keep-alive bugs usually come from stale timers. Give each socket instance its own ping interval, reconnect timer, and close handler. When a socket closes permanently or is replaced, clear those handles before creating the next socket. Otherwise the client can send duplicate pings, open multiple reconnect loops, or restore subscriptions twice.

Guardrail	Behavior
Ping interval	One active interval per socket instance
Reconnect timer	Cancel when a new socket opens or the job stops
Retry budget	Stop after the configured budget and surface the failure
Jitter	Add random delay so many clients do not reconnect together
Subscription restore	Restore from tracked state after open, not before
State rebuild	Mark local books or alerts unsafe until a new baseline is loaded

Reconnect Sequence

Stop old timers

Clear ping intervals and pending reconnect timers attached to the closed socket.

Wait with jitter

Use capped backoff so network churn does not become a reconnect storm.

Open a new socket

Authenticate the new connection through the same safe key path.

Restore subscriptions from tracked state

Recreate only the active channel-symbol pairs recorded by the client.

Rebuild unsafe state

For local books, replay, or alert state, use a fresh snapshot or replay checkpoint before trusting output again.

Production Review

A client that only handles onopen and onmessage is a smoke test. A production stream consumer also handles onerror, onclose, keep-alive, replay stop conditions, gap events, unsubscribe behavior, and backpressure. Keep-alive behavior should be visible in logs so support can distinguish normal network churn from auth failure, rate pressure, or a broken consumer.

Backpressure Link

Keep-alive and backpressure interact. If the message queue is growing, a healthy ping loop does not mean the stream is safe to use. Record queue depth or processing lag beside reconnect logs, and slow replay or reduce subscriptions when the consumer cannot keep up.

Next Step

Use WebSocket connection for the startup path and WebSocket limits for subscription and replay sizing.

Connect

Stream

Replay

WebSocket keep alive

Basic Ping Loop

Keep-Alive Model

Timer And Retry Guardrails

Reconnect Sequence

Production Review

Backpressure Link

Next Step

​Basic Ping Loop

​Keep-Alive Model

​Timer And Retry Guardrails

​Reconnect Sequence

​Production Review

​Backpressure Link

​Next Step

Basic Ping Loop

Keep-Alive Model

Timer And Retry Guardrails

Reconnect Sequence

Production Review

Backpressure Link

Next Step