[natekspencer]

Summary

Fixes the library side of bretterer/home-assistant-rivian#305. After a restart, setup kept failing because the web socket was closed with 1013 "Rate limited". Each failed retry left a client reconnecting, and each new client restarted its backoff, so the limit never cleared until a fresh login.

Changes

• Stop reconnecting with nothing to deliver. A monitor with no subscriptions stops after a disconnect instead of reconnecting forever. A failed subscribe closes the connection when nothing else is subscribed, and so does removing the last subscription.
• Share the backoff per user session. A new client waits out another client's backoff instead of starting over. The backoff only holds off opening new connections, and the monitor re-checks it after waiting.
• Never lose a backoff. It's recorded as soon as the cause is known (a 1013 close or a missing connection_ack), so close() can't drop it.
• Fail subscribes fast, and say why. A subscribe stops waiting when the connection closes and logs the cause: rate limited, not acknowledged, or closed (with the server's reason, e.g. Unauthenticated). Failures caused by a backoff are warnings; others are errors.
• New public properties on Rivian:
  • ws_backoff_remaining: seconds until the web socket may reconnect;
  • ws_rate_limited: whether the web socket is backing off after a rate limit, for any client on the session;
  • ws_unauthenticated: whether the server closed the socket as unauthenticated.
• Keep connections up through callback errors. A subscription callback that raises is logged, and the connection keeps going.

Behavior changes for callers

• Unsubscribing every subscription closes the web socket. The next subscribe reconnects, and that counts toward the server's rate limit.
• Failed subscribes return sooner. While backing off, subscribe returns None without connecting.
• To trigger reauth, use ws_unauthenticated and ws_rate_limited instead of reading _ws_monitor.
• Integrations still need to close the client when setup fails.

Tests

tests/test_ws_monitor.py covers:

• stopping when nothing is subscribed;
• the shared backoff, including a backoff pushed back by another client and an open connection that isn't blocked;
• fast, explained subscribe failures (rate limited, not acknowledged, unauthenticated) and their log levels;
• a backoff surviving close();
• closing on the last unsubscribe;
• callback errors;
• the per-client rate-limit flag.

The backoff fixture uses a fake clock, so the tests don't depend on the random part of the delay.

🤖 Generated with Claude Code