Canceling a slow loading navigation without triggering a net error for HTTPS-Upgrades

[Mustafa Emre Acer]

I'm trying to cancel a navigation timeout (set by SetNavigationTimeout) without triggering a visible net error as a fix for crbug.com/41488861. I'd appreciate any suggestions that could prevent these noisy error pages for what's supposed to be "silent" upgrades.

Context:

HTTPS-Upgrades feature opportunistically upgrades HTTP navigations HTTPS. If the upgraded HTTPS load takes long, the upgrade is canceled and the original HTTP URL is loaded as a fallback. We call this fast-fallback.

To support fast-fallbacks, we introduced the concept of navigation timeouts:

• Original thread

• Code

Presently, an upgraded HTTPS load is initiated with a timeout of 3 seconds. When it times out, the HTTPS upgrade interceptor handles the failed upgrade and initiates a fallback navigation. A fallback navigation is initiated by first allowlisting the HTTP URL (so that it's not upgraded again), then loading it again in the same tab. A URL is also allowlisted if it's part of a redirect chain. 

In https://crbug.com/41488861#comment22, a combination of a slow redirect and a redirect loop is causing the net error to be visible for a short while.

This scenario seems to happen frequently enough that it's impacting our metrics for an HTTPS-Upgrade related launch.

As a potential fix, I tried introducing a CancelNavigationTimeout() and calling it on redirects, but obviously this breaks the fast-fallback logic. E.g. If an upgraded URL (https://a.com) redirects to https://slow.com, we won't fast-fallback and the upgrade will hang.

Thanks!

Mustafa

[Mustafa Emre Acer]

Another incomplete idea was to introduce a new net error code to be returned only for navigation timeouts and also exclude it from net error pages. It wasn't clear to us if this is feasible. 

[Rakina Zata Amni]

Is the problem about the navigation cancellation that's caused by the navigation timeout commits an error page, while the new navigation is still ongoing? Is it possible to use a pre-existing error code for the navigation timeout cancellation, that makes the navigation be silently cancelled instead of committing an error page (e.g. ERR_ABORTED)?

--
You received this message because you are subscribed to the Google Groups "navigation-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email to navigation-de...@chromium.org.
To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/navigation-dev/CAHafXh3jWg6ecGk0JjAMZGHt29qG1WAw8DYuK2JpiHSK5oCquA%40mail.gmail.com.

[Chris Thompson]

Interestingly, I quickly tried modifying NavigationUrlLoaderImpl::SetNavigationTimeout() to use ERR_ABORTED and it looks like it is correctly triggering the HttpsUpgradesInterceptor::MaybeCreateLoaderForResponse() flow for serving the followup redirect in the common case (e.g., navigating to http://httpbin.org/delay/10) and in the simple redirect-to-timeout case (e.g., https://httpbin.org/redirect-to?url=http%3A%2F%2Fhttpbin.org%2Fdelay%2F10). But it doesn't work for the slightly more complicated redirect-loop-with-timeout bug repro case (crbug.com/41488861#comment22) -- there instead of the flash of the net error page we get a real aborted navigation which prevents the followup redirect entirely.

On Fri, Sep 20, 2024 at 2:05 PM Mustafa Emre Acer <mea...@chromium.org> wrote:

Also, apologies, I used the wrong URL for the "original thread" link. It should be https://groups.google.com/a/chromium.org/g/navigation-dev/c/BrXS3578Vew/m/qLTYANB8CAAJ instead.

On Fri, Sep 20, 2024 at 1:58 PM Mustafa Emre Acer <mea...@chromium.org> wrote:

Hi Rakina,

> Is the problem about the navigation cancellation that's caused by the navigation timeout commits an error page, while the new navigation is still ongoing?

Yes. And because the timed-out navigation is allowlisted (because it was part of a redirect chain), its error page will show briefly, until the fallback navigation completes.

>  Is it possible to use a pre-existing error code for the navigation timeout cancellation, that makes the navigation be silently cancelled instead of committing an error page (e.g. ERR_ABORTED)?

It appears that ERR_ABORTED aborts the whole navigation including subsequent redirects. Is that expected? With ERR_TIMEOUT, SetNavigationTimeout() cancels the current redirect hop, but subsequent navigations (including the fallback navigation) still take place. 

[Mustafa Emre Acer]

Indeed, I was testing with the redirect-loop-with-timeout repro case in that bug and saw the same behavior. I also quickly tried adding a new error code (HTTPS_UPGRADE_TIMED_OUT), and the flash of the net error still appeared, this time with the updated error code. So a custom error code doesn't seem appropriate either.

[Mustafa Emre Acer]

Also, apologies, I used the wrong URL for the "original thread" link. It should be https://groups.google.com/a/chromium.org/g/navigation-dev/c/BrXS3578Vew/m/qLTYANB8CAAJ instead.

On Fri, Sep 20, 2024 at 1:58 PM Mustafa Emre Acer <mea...@chromium.org> wrote:

Hi Rakina,

> Is the problem about the navigation cancellation that's caused by the navigation timeout commits an error page, while the new navigation is still ongoing?

Yes. And because the timed-out navigation is allowlisted (because it was part of a redirect chain), its error page will show briefly, until the fallback navigation completes.

>  Is it possible to use a pre-existing error code for the navigation timeout cancellation, that makes the navigation be silently cancelled instead of committing an error page (e.g. ERR_ABORTED)?

It appears that ERR_ABORTED aborts the whole navigation including subsequent redirects. Is that expected? With ERR_TIMEOUT, SetNavigationTimeout() cancels the current redirect hop, but subsequent navigations (including the fallback navigation) still take place. 

On Fri, Sep 20, 2024 at 6:06 AM Rakina Zata Amni <rak...@chromium.org> wrote:

[Mustafa Emre Acer]

Hi Rakina,

> Is the problem about the navigation cancellation that's caused by the navigation timeout commits an error page, while the new navigation is still ongoing?

Yes. And because the timed-out navigation is allowlisted (because it was part of a redirect chain), its error page will show briefly, until the fallback navigation completes.

>  Is it possible to use a pre-existing error code for the navigation timeout cancellation, that makes the navigation be silently cancelled instead of committing an error page (e.g. ERR_ABORTED)?

It appears that ERR_ABORTED aborts the whole navigation including subsequent redirects. Is that expected? With ERR_TIMEOUT, SetNavigationTimeout() cancels the current redirect hop, but subsequent navigations (including the fallback navigation) still take place. 

On Fri, Sep 20, 2024 at 6:06 AM Rakina Zata Amni <rak...@chromium.org> wrote: