e2e: Close pages before disabling MSW

[Turbo87]

Without this, requests still in flight (or fired by late-running hydration code) bypass the mocks once worker.disable() removes the interception and hit the real network instead, surfacing as [404] log lines from the SvelteKit vite preview web server.

[Turbo87]

@kettanaito is this something that should be mentioned in the MSW docs? (or maybe it already is, in that case apologies for the ping 😅)

[kettanaito]

Hi, @Turbo87. We've already addressed that and the proper fix is coming in 3.0 as that requires a breaking change both in the worker.stop() API and the worker script.

In fact, could you please give this branch a try and see if your change is still necessary with it? Thanks.

[Turbo87]

could you please give this branch a try and see if your change is still necessary with it?

I tried the beta from pkg.pr.new/msw@2692 against our full test suite, but it looks like the leaks are still there.

We are using @msw/playwright in this case, instead of the raw msw, so maybe that is the reason for this not fixing the issue?

Our local fix is to close all context pages before calling worker.disable() in the fixture, which prevents Playwright's Fetch.disable from racing with late requests.

[Turbo87]

damn... the change in the PR worked perfectly locally, but I've rebased #13373 on top of it and CI still shows leaked requests 😵‍💫

[Turbo87]

interestingly, removing the network.disable() (or in our case worker.disable()) seems to fix it on CI 🤔

... but locally it still shows the occasional leak 😵‍💫

[kettanaito]

We are in the process of updating @msw/playwright to use the new architecture so understandably the issue will still be reproducible there. It will take some time to that binding to adopt 3.0.

[Turbo87]

yeah, no worries. it's not breaking anything here, just unintentionally leaking requests to the dev server. feel free to ping me once @msw/playwright is ready for testing :)