In the meantime, I think I’ve identified the bug.
In Gladys Assistant 4.56, I introduced a new authentication logic on the WebSockets, allowing a faster connection: ideal for instant access to the dashboard on mobile.
The problem? If the instance loses the connection, it tries to reconnect with the same access_token used during the first connection. Except that this access_token has expired in the meantime and is not renewed. I’m using a new logic present in the socket.io library and I didn’t understand its behavior on disconnection.
Result: the Gladys Plus backend rejects the connection (expired JWT), and the instance enters an infinite reconnection loop.
It’s a good lesson, and a few avenues for improvement:
-
Renew the access_token in case of connection loss to resume with a valid token.
-
Add a delay before reconnecting, to avoid overloading the server in case of an infinite loop.
-
Strengthen unit tests to better cover connection loss scenarios and prevent this bug from recurring.
Sorry for the inconvenience!
I’ll keep you informed as soon as version 4.56.1 is available 