Reconnect to kernel on restart action

[kevin-bates]

This pull request addresses the issue raised in #8103 and closed by #8432. That change was not sufficient as the reconnect() was embedded within the _handleRestart() method - which is conditionally called as a result of execution states of autorestarting or restarting, only the latter of which is returned by the server and only when the kernel is automatically restarted. No execution status is returned from the server upon completion of a manual restart request. As a result, the fixed code will never be reached upon a manual restart.

Please note that relative to the originating issue, neither JupyterHub nor (arguably) Enterprise Gateway are related. The issue is really about whether a kernel's ZMQ ports are refreshed on manual restarts. For local kernels, that is atypical. For remote kernels deployed across managed clusters, it is the norm since the kernel's physical location is likely to change. EG happens to be a way kernels are deployed across a cluster.

This issue does not occur when using the Notebook UI - whose kernel restart behavior includes the reconnect sequence. As a result, this change brings parity with Notebook.

References

JupyterHub+JupyterLab+JupyterEnterpriseGateway Restart Kernel hangs #8013

Reconnect a websocket when a kernel is restarted. #8432

Code changes

We now explicitly call reconnect() immediately following the conclusion of the kernel restart request. This change was recommended by @blink1073 in #8013 (comment).

User-facing changes

Restarts of kernels, irrespective of location, now yield a working (connected) kernel.

Backwards-incompatible changes

None - this change introduces parity with the Notebook UI wrt kernel restarts.

[jupyterlab-dev-mode]

Thanks for making a pull request to JupyterLab!

To try out this branch on binder, follow this link:

[blink1073]

Thanks!

[blink1073]

Congratulations @kevin-bates on your first PR to JupyterLab! 🎉