running nbclient against enterprise gateway

[jgeraerts]

I'm using nbclient to run notebooks against an enterprise gateway on a k8s cluster. The issue i'm running into is when a kernel gets killed (e.g by OOMKiller) the enterprise gateway will restart the kernel immediately. The nbclient is using is_alive checks which succeed, because the kernel got restarted, but there is no code executing anymore.

On the iopub channel there is a restarting status being posted, however this is not being processed by nbclient and it keeps on waiting for a result which will never happen because the kernel got restarted.

The code on https://jupyter-enterprise-gateway.readthedocs.io/en/latest/developers/kernel-manager.html together with the Autokill.ipynb notebook from the testcases demonstrates the issue.