What @guim31 describes is very close to what I observed at my place. Here: Gladys devient injoignable après quelques heures de fonctionnement - #9 par pierre-gilles
And I did have a Node-RED flow that used the same Telegram as Gladys. I disabled that Node-RED flow, and Gladys has been working much better since.
But not yet perfect — I had another freeze afterwards that I couldn’t analyze at the time… So I’m in ‹ observation › mode, with a Gladys scene that triggers every hour and writes a ‹ keep alive › to Telegram, to efficiently spot the next crash 
But yeah, I’m really short on time right now to sit down and seriously analyze all this…