Better handling of failed ipykernel installations & ipykernel detection

[DonJayamanne]

We have plenty of scenarios where users are prompted to isntall ipykernel again & again.
This could happen when:

• pyzmq is not installed or is busted (Kernel died with exit code 1073741845.  #5963)
• ipython_genutils is not installed
• Conda env activation fails
• Or other dependencies are not installed or busted

ipyhton_genutils is a great example of an upstrema bug, (ipykenrle removed that dependency), but was required by other packages in jupyter (which didn't specify that as a requirement).

Whats going on

• Jupyter will get if ipykernel is installed (by importing ipykernel)
• Python extension will prompt user to isntall ipykernel
• Installation is a success
• Jupyter will get a message saying it was not installed (as importing ipykernel fails)
• No permissions to install in powershell

Thus, though ipykernel installatino ran successfully in temrinal, user gets a message that its not installed

Suggestion

• Failure to activate conda env results in failure to detect if ipykernel is installed #8119
  • Add telemetry to see how often this fails for conda users (if failed, then nothing will work)
  • Increase the timeout for conda environments (or retry with a higher timeout?)
  • Possible solutions here Perf improvements for starting kernels (perf resulting from checking packages & dependencies) #6694
• Temporarily enable permissions to run powershell scripts in that powershell session
  • IPyKernel doesn't install correctly when powershell is default #421
  • Install IPyKernel into all Python environments without the terminal #9376
• Installer for modules does not surface any errors when checking for if a module is installed #1257
  • Fixes Kernel died with exit code 1073741845.  #5963
  • Jupyter Server: No Kernel Python: Not Started #1128
• Installing ipykernel requires admin on Windows #1253
• IPyKernel doesn't install correctly when powershell is default #421
• Add telemetry in python extension to see if installation is successful and ipykernel cannot be imported (this is an indication of something wrong with the package)
• If this happens we can notify the user and ask them to do somethign (kernel doctor?)
  • Display message with possible solutions & links to wiki Kernel startup error messages should contain more information about what went wrong and how to resolve it #8354
• If jupyter extension keeps asking users to install ipykernel > n times without running a cell succesfully,
  • Add telemetry to detect this scenario
  • then display another button in the prompt withy more informaiton to install it them selves
  • Not required, as we haven't had reports of users complaining about having to install ipykernel multiple times.
• More tests with more environnments?
  • E.g. we found python extension boke recently, how do we prevent this in the future?
• Do we pull installation into jupyter extension?
  • Pip install the DS packages in Jupyter extension if Python extension fails to install them #8325
• Display error message with instructions to install ipykernel in cell output #8327
• Display kernel startup failure messages in cell output #8351
• Kernel startup error messages should contain more information about what went wrong and how to resolve it #8354
• Move installation of all DS packages into Jupyter Extension #8457

[DonJayamanne]

Closing this issue, as we've addressed all of them