I did find the problem above and have fixed it in my development branch. It was a simple buffer full issue. The problem you’re reporting is different. One of my home systems has been having that problem with a MBTF of several days. I have not had enough incidents to see any kind of pattern. It seems to have started with the introduction of async HTTP. My development system uses a newer ESP/Arduino core and it is the same.
This is turning into a trade off between intermittent problems and improving general functionality. The async HTTP was huge in both keeping the sampling rate high and not timing out in poor connection environments.
The new core is generally faster and more robust in other ways, but has a temporary memory leak issue with high I/O rates.
To mitigate these problems, the development branch has a watch-dog-timer on the WiFi connection that restarts the ESP if disconnected for 2 hours. The restart takes a couple of seconds and then all of the backed up data is uploaded. Pretty harmless.
The memory leak affects the history upload in that it has to keep pausing to allow the low-level IP code to timeout and release memory. I have a commitment from the ESP folks to resolve that issue, so hopefully it’s a temporary fix.
So the next release will not require that you power cycle, and if I can figure out what I can do to fix the disconnect short of a restart, I’ll fold that in as well.
Thanks for the feedback.