Here's a different approach to printing via a ESP(-like) device:
This project implements part of the OctoPrint API on the ESP. So it actually replaces OctoPrint in that regard. But if one OctoPrint instance could send a whole printjob to another instance, that would take away most of the wifi stability issues.
Given the size of the wifi megathread on here, I can't even begin to imagine trying to support people trying to flash an ESP8266. For those who already have them, it's a cool solution in theory. For anyone who doesn't have one, I'd suggest a Pi.
Finally watching this. @foosel, with regard to ~40% of all prints being canceled, can you fetch a histogram on the duration to cancel? I suspect the vast majority are at "first layer" or before.
and it's possible I could be convinced to work on the Grafana problem. PhantomJS is gone, chromium headless is supposed to be simple, and I've wanted to bolt Grafana onto my local stuff anyhow, so I would hopefully have a development env.
on the undervoltage graph, showing 3k instances, is that unique instances? So out of ~15k octoprint instances phoning home regularly, 20% are undervoltage?
If I don't work on Grafana I am thinking it's time for a packaged plugin that pops up a modal on every 'start print' for things that are extremely bad- undervoltage, Pi Zero or ancient Pi, etc.
With respect to the new firmware error—reporting feature for the next major release, I think it would be useful in that graph to also know what firmware that is. So if "Marlin 1.1.5", e.g., is throwing lots of a certain kind of error then we could know this.
With respect to ESP3D, I was able to create a working rig with the rfc2217 redirected serial connection to Zero W and 3A+ Raspberry Pi computers and provided some stats here. That said, I don't think it's the community's or foosel's responsibility to build business-class solutions for printer farms. I would suggest that they need to pay for development or to simply eat the cost of one-printer-one-Pi as part of their cost of doing business. (I don't honestly think that these ESP-related chips have enough memory for the task.)