Sugar Valley NeoPool - Home Assistant Integration

I did some digging with Claude:

What changed with the Mosquitto add-on update

The add-on 7.x update was much bigger than it looks: it moved Mosquitto from 2.0.22 to 2.1.2 and updated the authentication plugin (mosquitto-go-auth) to a new major version 3.0.0. And there are currently three open issues reporting that this new auth plugin crashes the whole broker with a nil-pointer panic, several times a day for some users:

Every time it crashes, the watchdog restarts the broker and all clients get dropped — which fits your symptoms perfectly: it hits your WallPanel tablet too (client-agnostic), entities come back for ~10 minutes and then die again, and the LWT stays “Offline” because the broker keeps dying mid-session.

Please check this: open the Mosquitto add-on log and look for lines containing panic: runtime error or SIGSEGV. If you find them, you’re hitting this bug — please add your logs and a :+1: to #4655, it needs visibility because it’s getting no maintainer attention so far.

How to downgrade the add-on (your blocker)

You’re right, there’s no command to install an older add-on version — that’s by design, unfortunately. The only supported way is restoring a backup:

  1. Go to Settings → System → Backups. When you updated the add-on via the UI, HA should have automatically created a partial backup of the old version, named something like core_mosquitto 6.5.2.
  2. Restore that backup → you’re back on add-on 6.5.2 = Mosquitto 2.0.22, the version everything worked on.
  3. Important: disable auto-update on the Mosquitto add-on afterwards, otherwise it will just re-upgrade itself.

CLI equivalent if you prefer: ha backups list then ha backups restore <slug> --addons core_mosquitto.

If that automatic backup doesn’t exist, there’s no official rollback path. If I was you, I’d use EMQX instead of Mosquitto. There should be an EMQX addon in HA addons. I use HA Core in Docker, I don’t use HAOS, so can’t help you there but it should be easy.

For completeness

The “exceeded timeout” keepalive regression in Mosquitto itself (eclipse-mosquitto#3226) was fixed in 2.0.22 and shouldn’t be present in 2.1.2. And your observation about the timezone jump triggering a disconnect matches an old wall-clock bug (eclipse-mosquitto#3405) that was also fixed in 2.0.22 — real phenomenon, but a red herring for your current problem.

So: check the add-on log for the panic, restore the 6.5.2 backup, and let me know how it goes. If you can, share the add-on log here (or on the GitHub issue) — it would help confirm the diagnosis.

I use HA OS and can confirm that EMQX runs perfectly as an add-on (even as a cluster node): I would also recommend using EMQX instead of Mosquitto.

Here is one of several guides on how to install it.

Hi Alex,
I use Claude too, but probably I don’t asking the questions correctly.
Confirmed: restored the addon_core_mosquitto_6.4.0 backup and the disconnections stopped. No panic/SIGSEGV in my logs though — just constant exceeded timeout on both SmartPool and my WallPanel tablet, so it looks broker-side but maybe a different failure mode than the auth plugin crash. Thanks a lot for the pointer!

Thank you Curzon.
I will evaluate it in case I still have mosquito problems

No problem. But if I was you, I would switch to EMQX. Mosquitto is very basic.

Alex, I’ve follow your suggestion because after 4 hours again problem of connection. Now with EQMX is stable, I see in these days if no other problem.
Sorry guys, the major problem was mosquito broker!