🧪 Beta 8
Entities went unavailable for a few minutes at a time after every session — if you use an external reference sensor for cooldown. Every update of that sensor re-armed the 5-minute fallback poll, so a sensor reporting every minute (typical for a BLE H&T) silenced polling entirely. The heater was then only heard from through its own pushes — about every 15 minutes when idle — and 10 minutes after each one it counted as stale, taking every entity with it until the next push. Measured on 26 Sep: three outages of ~4.5 minutes, 15 minutes apart; the same pattern shows up after earlier sessions. Fixed: sensor updates refresh the entities without touching the poll schedule.
Cooldown flicker guard counts sensor readings, not evaluations. It is meant to wait for three readings below the end temperature, but it counted every evaluation — and one reading is evaluated by the sensor update, by pushes and by polls. A single low outlier could therefore end the session early. Each report now counts once.
A reload could hang the integration for good. Tearing down the cloud connections had no timeout, so a half-dead connection left the config entry in unload in progress until Home Assistant was restarted. Every teardown step is now bounded (5 s), and polling stops as soon as the shutdown begins.
Verified against the maintainer's own Xenio data; the Fenix side shares only the connection teardown change.
Note: this is a pre-release. Enable "Show beta versions" for the integration in HACS to test it.