Damn.
There’s been some discussion in this other thread too. I just posted there a bit.
I’ve managed to keep my harvester online now for the past 12 hours, which is the first time it didn’t stop during the night. What seemed to be the fix for me was to decrease plotting.
I have a Ryzen 7 5800x CPU (8 core, 16 thread), and 16 GB RAM. I’ve been plotting 3 in parallel on my 1 TB SSD, so that I’ve used 12 threads and 12GB RAM. However, I’ve been checking from Task Manager / Performance Monitor my memory usage, and even with only 2 plots (that seem to consume only 2,5 GB each based on the Task Manager), my memory usage keeps jumping between 40 and 70 % depending on the plotting phase. So, with a few other programs and GPU miners, it might not be impossible that during some heavy phase, I just run out of memory, and the harvester is the least prioritized, leading it to shut down or who knows what. No other program has ever been affected, nor full node or plotting itself. But regardless of what’s going on, at least now it seems to be working for me.
Btw, I think I also kept getting these websocket errors, so I went through everything possible from allowing Windows Defender to pass everything through Chia, rejecting any sleeping from any program, HDD etc, updating all the software, stopped plotting into the farming folder etc, and nothing helped. This is also quite mysterious since I believe I have been plotting 3 in parallel for the past 1 month. So either this bug/issue was introduced only a few versions ago, or I’ve been extremely lucky being able to win some Chia even while harvesting being stopped >50 % of time.