retry_and_liveness.md

Retry and Liveness

The producer manages websocket lifecycle in two dimensions:

• retry and reconnect
• liveness tracking

Reconnect and retry

Connection attempts are controlled by:

• :ws_retry_opts
• :ws_retry_fun

After a failed connect or bootstrap failure:

• :ws_retry_fun is called with the current retry state
• the returned :delay is used before the next attempt
• the updated retry state is stored for the next failure

After successful reconnect:

• retry state resets to the initial :ws_retry_opts
• frame_handler_state also resets to its initial baseline

Liveness

Liveness is refreshed by:

• inbound data payloads
• inbound pong frames
• {:skip, new_state} frame-handler results

The producer checks inactivity with :ws_timeout.

If the websocket is idle beyond that timeout, the producer treats it as a timeout failure and reconnects.

Important consequence

If your protocol emits heartbeats or acknowledgements that should not be sent downstream, handle them with {:skip, new_state} rather than dropping them outside the frame handler. That keeps the connection healthy without polluting Broadway messages.