All Interactive Window tests failing

[roblourens]

Something must have gone wrong in setup because it seems that every IW test failed. I saw the same thing in another run from yesterday

https://github.com/microsoft/vscode-jupyter/runs/8232442931?check_suite_focus=true

[rebornix]

I'm seeing this again https://github.com/microsoft/vscode-jupyter/actions/runs/3046878708/jobs/4910227190

[DonJayamanne]

Looks like some issues in Core, something could not be loaded, I've seen this in the past as well, and only manaaged to get things working after re-installing VS Code.
I.e. when ever i see Could not register service workers: TypeError: Failed to register a ServiceWorker I know its a bug in VS Code core.

2022-09-13T16:59:20.8487109Z A bad HTTP response code (504) was received when fetching the script.
2022-09-13T16:59:20.8521684Z Webview fatal error: Error: Could not register service workers: TypeError: Failed to register a ServiceWorker for scope ('https://14ml6ta8ob5jtoorsm0fmomggogpr45tanlf3u6jkl91p3ra1h7t.vscode-cdn.net/insider/1a55beb2aaee64eb8d0cbe5b61b10088bbc6c895/out/vs/workbench/contrib/webview/browser/pre/') with script ('https://14ml6ta8ob5jtoorsm0fmomggogpr45tanlf3u6jkl91p3ra1h7t.vscode-cdn.net/insider/1a55beb2aaee64eb8d0cbe5b61b10088bbc6c895/out/vs/workbench/contrib/webview/browser/pre/service-worker.js?v=4&vscode-resource-base-authority=vscode-resource.vscode-cdn.net&remoteAuthority='): A bad HTTP response code (504) was received when fetching the script..
2022-09-13T16:59:20.8833668Z Error loading webview: Error: Could not register service workers: TypeError: Failed to register a ServiceWorker for scope ('https://14ml6ta8ob5jtoorsm0fmomggogpr45tanlf3u6jkl91p3ra1h7t.vscode-cdn.net/insider/1a55beb2aaee64eb8d0cbe5b61b10088bbc6c895/out/vs/workbench/contrib/webview/browser/pre/') with script ('https://14ml6ta8ob5jtoorsm0fmomggogpr45tanlf3u6jkl91p3ra1h7t.vscode-cdn.net/insider/1a55beb2aaee64eb8d0cbe5b61b10088bbc6c895/out/vs/workbench/contrib/webview/browser/pre/service-worker.js?v=4&vscode-resource-base-authority=vscode-resource.vscode-cdn.net&remoteAuthority='): A bad HTTP response code (504) was received when fetching the script..

[roblourens]

These have both been in web. This is the same sort of webview error that we have seen in vscode integration tests, microsoft/vscode#126371 (comment)

cc @mjbvz, is there anything we can do here?

If this is behind lots of flakiness in vscode-jupyter web tests I suggest we make all web tests optional or add a retry, but we clearly need to test web

[mjbvz]

@roblourens This appears to be an electron bug microsoft/vscode#125993

Unfortunately I don't know how to resolve it or work around it

[roblourens]

That issue sounds like it's specifically talking about an error like The document is in an invalid state, this is a different message, or is that issue about all "Could not register service worker" issues?

[rebornix]

Some of the interactive window execution tests are flaky in the last week.

[amunger]

The vast majority of the failures are happening in web.
The failures aren't all due to that 504 error. The one that I'm currently looking into never starts the kernel for the set of tests that fail:

Get default Remote Controller, Class name = ControllerDefaultService, completed in 0ms, has a truthy return value, Arg 1: "interactive", Arg 2: undefined, Return Value: <Return value cannot be serialized for logging>
CreateDefaultRemoteController
Intiailize notebook communications for editor Interactive-1.interactive
Test Remote Jupyter Server Output: [I 16:18:07.094 NotebookApp] Starting buffering for

[MISSING from failure case]
Test Remote Jupyter Server Output: [I 21:58:03.908 NotebookApp] Creating new notebook in 
info 21:58:03.885:  InteractiveWindow.ts.createExecutionPromise.start
verbose 21:58:03.885:  start the kernel, options.disableUI=false
Starting Jupyter Session startUsingRemoteKernelSpec

getRemoteUri: http://localhost:
list kernel specs

It's hard to see what leads to triggering the kernel to start from the logs in a successful test run though, most of the previous logs are about disposing objects from previous tests. Need to try and dig through the code to see if I can find why kernels would just stop trying to initialize for an entire suite.

[amunger]

haven't seen this for a while