Log once when unavailable, and once when back

This is a Silver rule. In quality_scale.yaml, it is called log-when-unavailable.

When the device or service cannot be reached, the integration logs that once. When it can be reached again, the integration logs that once as well. Not on every failed update in between. When you write these messages yourself, log them at info level.

When entities go unavailable, users look in the logs to find out why. One message when the connection was lost, and one when it came back, tells them what happened and when. A message on every failed poll buries that in noise, fills the log file, and hides other problems.

The coordinator does this for you. Raise UpdateFailed when the update fails. The coordinator logs the first failure, stays quiet while the failures continue, and logs at info level when fetching the data recovers:

async def _async_update_data(self) -> MyData:
    """Fetch the latest data."""
    try:
        return await self.client.get_data()
    except MyConnectionError as err:
        raise UpdateFailed(
            translation_domain=DOMAIN,
            translation_key="update_failed",
        ) from err

An entity that updates itself keeps track of whether it already logged:

class MySensor(SensorEntity):
    """A sensor that updates itself."""

    _unavailable_logged = False

    async def async_update(self) -> None:
        """Fetch the latest value."""
        try:
            data = await self.client.get_data()
        except MyConnectionError as err:
            self._attr_available = False
            if not self._unavailable_logged:
                _LOGGER.info("The device is unavailable: %s", err)
                self._unavailable_logged = True
            return

        self._attr_available = True
        self._attr_native_value = data.temperature
        if self._unavailable_logged:
            _LOGGER.info("The device is available again")
            self._unavailable_logged = False

• Pylint warns when an error is logged at warning level or higher earlier in the same block that raises it to Home Assistant, like logging before raising UpdateFailed (home-assistant-log-and-raise). Home Assistant already reports those exceptions.
• The rest is checked in review.

If the integration has no entities, this rule does not apply.

• entity-unavailable: mark entities unavailable when appropriate.