MRU list for server connections is now super slow with kernel work

[IanMatthewHuff]

One basic scenario to demo this:

1. Start up JupyterServer A from CPU A
2. From VSCode on CPU B connect to JupyterServer A
3. Close down the connection in VSCode on CPU B
4. Close down JupyterServer A on CPU A
5. Start up a new JupyterServer B on CPU A
6. Open up the "Select Remote or Local" dialog on VSCode in CPU B

In this scenario the only difference between servers A and B is the token. As such, when trying to populate the MRU list it's going to try sending lots of kernel API requests with the old A token to the new B server. In the command windows for the jupyter server on CPU B you can see all the 403 forbidden requests for this. In this scenario for me opening up the "Remote or Local" command and actually seeing the selection box takes over 30 seconds with no progress shown.

[DonJayamanne]

Do we even need to check if the url is alive (reachable)?
Why not do it in the background and return what even information has completed?

Unsure how kernel work world have impacted this, as it's unrelated.

[IanMatthewHuff]

Honestly, right now I'm having trouble replicating the really bad perf. Now it's only a 4-5 second wait, which might be acceptable. But getRunningServerDetails is getting called for every server in the list many of which might be remote. And it's going to call Kernel.listRunning on everything in there. Seems like this would always have the possibility of hitting bad cases. Can we change the pick items after we've submitted them? If not I don't think any degree of hitting servers and querying kernels would be acceptable personally.

[rchiodo]

We should have a way to tell if it's running or not though? Wouldn't that be useful to the user?

[DonJayamanne]

Should we add telemetry to see how long it takes to popup this list in the wild?

[rchiodo]

Yeah I think so.

[DonJayamanne]

We already do. SelectJupyterURI

[IanMatthewHuff]

From discussion. We don't want to hit a server at all to pop up a MRU list. Proposed plan.

Save a list of X (5? 10?) servers in global VS Code cache. With each save the last time that we were connected to it. Display the name and the last time we connected to it in the MRU list.

[DonJayamanne]

Validated