Repository navigation
Pass subscription errors to an on_error callback - #249
Conversation
`start_gear_guard_live_stream` starts a live camera stream the way the Rivian app does and returns its Kinesis Video WebRTC config (signaling endpoint, ICE servers and role), for a viewer to join: - It subscribes to `gearGuardLiveConfig` with a fresh command id, then sends the signed `security.video_monitoring.start_live_streaming` command with that same id; the cloud routes the config by it. - `parallax.encode_live_streaming` builds the command's payload from a new `LiveStreamingRequest` schema; `GEAR_GUARD_CAMERAS` names the cameras. - A refused start (e.g. a camera the vehicle doesn't have) raises RivianApiException straight away, through the subscription's `on_error`. It needs Gear Guard video monitoring on and a paired phone. The vehicle serves one session at a time, about 3 minutes each and an hour a day. Builds on bretterer#249. Verified against a 2026 R1S (Gen 2) running 2026.36.01: the config arrived within seconds of the start when the vehicle was awake, and the front and right cameras streamed in a WebRTC viewer. A sleeping vehicle can take longer to wake than the default 40 second timeout. Not tested: the rear, left and bed cameras, the R2's cabin camera. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FxWHBTrHU6eTPotwMWgtJF
4c27ea0 to
3adad09
Compare
natekspencer
left a comment
There was a problem hiding this comment.
Thanks, this is a useful hook and existing callers are unaffected. One change before merging:
on_error isn't guarded like the next callbacks. In _receiver, the next callback runs inside try/except Exception so a failing callback doesn't take the connection down, but subscription[2](data) is called bare. If it raises, the exception escapes the receive loop and the receiver task dies. _monitor replaces it about a second later without reading the exception, so it surfaces as "Task exception was never retrieved". #250's handler can hit this: it indexes errors[0], which raises if the server ever sends the error payload as an object instead of a list. Wrapping the call the same way (and supporting a coroutine on_error like next does, if you want them symmetric) would fix it:
try:
if inspect.iscoroutinefunction(on_error):
await on_error(data)
else:
on_error(data)
except Exception: # pylint: disable=broad-except
_LOGGER.exception("Error in a subscription error callback")A test with an on_error that raises, checking the connection stays up, would cover it.
|
Good catch, fixed. |
|
Thanks, the They will land in this order, so rebasing in the same order keeps the later ones from conflicting again:
I've moved this to draft for now; please mark it ready for review once it's rebased. |
The server ends a subscription with an `error` message when it fails, for example when the vehicle refuses a command the subscription waits on. The web socket only logged those at debug, so a caller waiting on the subscription could only time out. Subscriptions can now take an `on_error` callback that gets the error message; existing callers are unchanged.
An `on_error` that raised escaped the receive loop and ended the receiver task. Both callbacks now run through `_run_callback`, which logs a failure instead of raising and awaits coroutine callbacks.
4b2e11d to
e8fe7d9
Compare
`start_gear_guard_live_stream` starts a live camera stream the way the Rivian app does and returns its Kinesis Video WebRTC config (signaling endpoint, ICE servers and role), for a viewer to join: - It subscribes to `gearGuardLiveConfig` with a fresh command id, then sends the signed `security.video_monitoring.start_live_streaming` command with that same id; the cloud routes the config by it. - `parallax.encode_live_streaming` builds the command's payload from a new `LiveStreamingRequest` schema; `GEAR_GUARD_CAMERAS` names the cameras. - A refused start (e.g. a camera the vehicle doesn't have) raises RivianApiException straight away, through the subscription's `on_error`. It needs Gear Guard video monitoring on and a paired phone. The vehicle serves one session at a time, about 3 minutes each and an hour a day. Builds on bretterer#249. Verified against a 2026 R1S (Gen 2) running 2026.36.01: the config arrived within seconds of the start when the vehicle was awake, and the front and right cameras streamed in a WebRTC viewer. A sleeping vehicle can take longer to wake than the default 40 second timeout. Not tested: the rear, left and bed cameras, the R2's cabin camera.
|
Rebased onto main, ready for review. |
natekspencer
left a comment
There was a problem hiding this comment.
Thanks, the shared _run_callback guard and the new test look good.
When a subscription fails (e.g. the vehicle refuses a command it's waiting on), the server sends an
errormessage that was only logged at debug, so callers could only time out. Subscriptions now take an optionalon_errorcallback. Existing callers are unchanged.