Fix coordinator stall on transient API errors (retry backoff unit + raise UpdateFailed)

[QuintinHumphreys]

Problem

After a transient AquaTemp cloud/API error (or a failed poll around an HA restart), the climate + sensor entities stop updating and stay frozen until the config entry is reloaded. HA's DataUpdateCoordinator is meant to auto-retry on its interval, so a persistent stall points at the integration breaking that contract.

There are two root causes in managers/aqua_temp_api.py.

1. Retry backoff uses the wrong time unit (1000 s instead of 1 s)

asyncio.sleep takes seconds, but the retry delay is await sleep(1000) in both _internal_update and _perform_action:

if attempt < API_MAX_ATTEMPTS:
    await sleep(1000)   # 1000 seconds = ~16.7 minutes, not 1000 ms
    await self._internal_update(attempt + 1)

_async_update_data awaits _internal_update, so a single transient error blocks the coordinator's update method for 1000 s per retry - up to ~33 minutes across API_MAX_ATTEMPTS = 3. During that window the entities look frozen. The value was evidently intended as a short (1 s) backoff.

2. _internal_update swallows the error after exhausting retries

After the final attempt, _internal_update only logs and returns normally, so _async_update_data returns stale data as a success. The coordinator sets last_update_success = True, never raises UpdateFailed, and the entities never go unavailable - the failure is completely hidden. This is also inconsistent with _perform_action, which correctly re-raises after max attempts.

Fix

• Change both retry delays from await sleep(1000) to await sleep(1).
• Re-raise the captured error in _internal_update after the last attempt, so the coordinator's existing try/except wraps it in UpdateFailed and HA's normal retry/backoff engages.

if error is not None:
    if attempt < API_MAX_ATTEMPTS:
        await sleep(1)
        await self._internal_update(attempt + 1)
    else:
        _LOGGER.error(
            f"Failed to update (Attempt #{attempt}), Error: {error}, Line: {line_number}"
        )
        raise error

Testing

Applied to a live install (single AquaTemp heat pump). After ha core check + restart the integration loads cleanly (config entry state loaded), the API-status binary sensor is on, and temperature sensors report fresh live values on the normal poll interval. No UpdateFailed or login errors in the log.

🤖 Generated with Claude Code