jupyter kernel start on a compute server -- sometimes it hangs on starting

[williamstein]

I just saw the following on a computer server, and it's obviously a very serious bug, so will get fixed soon:

1. Created a compute server with the Tensorflow image
2. Immediately set a brand new Jupyter notebook to run on that compute server and selected a Python kernel.
3. When I tried to run import tensorflow as tf the startup just sat there with nothing running. Restarting the kernel didn't help.

WORKAROUND: I set the compute server for the notebook back to "Shared Resources" for a few seconds, then back to the compute server. Then everything immediately worked properly.

[williamstein]

For what it is worth, yesterday I made a bunch of videos involving compute servers and Jupyter and didn't hit this issue once. So it's a bit hard to reproduce.

[williamstein]

I disabled the pool and in my testing haven't seen this since. The pool is much less necessary for compute servers. I'll re-open this if the problem appears again.