Unexpected caching issues conflicting with CORB

[twalpole]

Unexpected caching issues conflicting with CORB

[geoffharcourt]

One further note: the specs that have these CORB problems pass with Cuprite, so I don't think this is an issue with our app or Chrome.

[twalpole]

Any chance of a reproducible test case/example?

[geoffharcourt]

Yeah, I'm sorry this isn't more specific. I'm trying to build a small enough application to produce the problem.

[twalpole]

Something along the framework of https://gist.github.com/twalpole/6711ad9aedf4449d5edb01890726e81b (Update to use apparition rather than selenium) is the simplest, although it may be too simple for your situation

[geoffharcourt]

Still working on a streamlined test case app, but adding

  browser_options: {
    "disable-web-security" => nil,
  },

to my Apparition options makes the tests work successfully (but obviously that's not a great solution).

[baffers]

+1... I've found using page.refresh after loading a page that will fire async JS will prevent the Provisional headers problem.. Obviously that solution doesn't scale. I'll give the disable-web-security a shot.

[geoffharcourt]

I haven't been able to get a small app together yet, but every time I've used a binding to halt the test and have reloaded the page once these issues resolve themselves for the rest of that example.

[twalpole]

Any update on a reproducible test case for this?

[geoffharcourt]

I tried adapting the script you provided but I can't reproduce the issue in it. I tried injecting all the headers from our report into the output to see if I could get a jQuery.get call to return the issue (so setting the same Cache-Control, etc) without success.

[geoffharcourt]

Update: I can't reproduce this by pointing the test script you suggested at our production app at a page that fails in test. I'm looking into whether this somehow has anything to do with requests that are on a non-standard HTTP port number in hopes that will be reproducible in a publicly-sharable case.

[geoffharcourt]

Something is different in newer versions of Chrome, and this problem solved itself even on older versions of our app from the time of the initial issue report. Thanks for your advice about things to try and sorry for the noise!

[geoffharcourt]

I was wrong, it's just intermittent. Holding off on reopening until I can generate this in a publicly-sharable app.

[geoffharcourt]

An update here: this appears to also be a problem in Cuprite, but Cuprite has dealt with it by setting disable-web-security out of the box: rubycdp/cuprite#71

This makes me think this issue might be related to CDP and not specific to these drivers.

[twalpole]

Hmmm -- disabling web security doesn't really seem like a good idea

[geoffharcourt]

Agreed. I just wanted to add that here since earlier in the issue I had said it wasn't a problem with Cuprite (thinking it was a unique issue in Apparition).

[MinasMazar]

I experienced the same issue. Don't know if it can help the investigation, but putting a before { Capybara.current_session.driver.headers = {} } in the spec (absolutely before any other Capybara statement) was enough to make it green.