"Something went wrong with the socket connection"

silvenga

It appears that Kagi can get into a weird state where SSE breaks. It appears the result of breaking causes no cookies to be sent with the request for SSE, resulting in a 302 redirect to the login page (which the UI considers an error).

I'm unsure the root cause, but the failure mode is likely undesirable.

At the very least, I would expect the UI to provide an error, rather than silently failing. However, optimally, the UI should recover more gracefully from this error.

Vlad

silvenga Is there any way you can reproduce this?

silvenga

I think this is a new problem - been seeing this quite a few times today - the symptom is similar to https://kagifeedback.org/d/2988-kagicom-fails-to-load-on-chrome - but in this state, the results never load.

silvenga

Ooh, yes, I just found a reproduction case Vlad

Ensure backwards cache is enabled (enabled in Firefox by default, some reports it's not enabled in Edge by default).
Ensure dev tools is open to record network activity.
Execute a query and click a result.
Immediately click the back button - observe that the browser used backwards cache and didn't make a new request to Kagi.
Attempt to make a new search without reloading the page, observe a request to /signin and observe that most operations in Kagi now fail.

I don't think this is the same issue as SSE breaking - since I don't observe a new SSE request. But the symptoms are the same e.g. no search results until page reload.

Vlad

silvenga Can you record a video of reproduction steps? Thanks

silvenga

Is there possibly a workaround for this? I'm getting hit by this every search session (I typically need to look at multiple results in quick succession, and this bug requires a refresh of the tab after each result clicked to mitigate).

Firefox no longer has the ability to disable bfcache - or at least Kagi can't find it. 😆

silvenga

Here's a screen capture.

I also discovered this does not happen in private windows at all - I'm guessing bfcache is disabled to reduce fingerprinting.

And I was wrong, this does cause a SSE request to trigger, which causes a failure, and the original error message in the console.

silvenga

FWIW, it wasn't called out in the release notes, but this appears fixed as of the last release (today). I am no longer seeing the SSE errors when using back-cache.

Thanks! That was incredibly frustrating, so I'm ecstatic!

Vlad

silvenga Thanks for reporting back!