Actually, it looks like the test is already marked as flaky with [Failure Pass] in regular TestExpectations, but I guess something is now also causing it to time out with --site-per-process.

The other test exception is linked to  issue 599670 , which speculatively blames the problem on some bots (and so - is probably a separate issue from what Site Isolation Linux experiences).

I am able to repro the timeout locally, so let me try to see if I can bisect this problem down to a single CL.

Unfortunately, the local timeout happens not only at the revision for build 20043, but also for earlier builds... (I've tried a few different gn configs, including the one used by Site Isolation Linux bot).

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/20196d32b0b98118006a080f9da90f0a09b1c15a

commit 20196d32b0b98118006a080f9da90f0a09b1c15a
Author: Lukasz Anforowicz <lukasza@chromium.org>
Date: Wed Aug 23 18:51:44 2017

Disable resource-parameters-ipv6.html layout test in --site-per-process mode.

NOTRY=true

Bug:  758075 
Change-Id: I0f7e430ff0bb4a9a71f2558aad283869fc81ab25
Tbr: alexmos@chromium.org
Reviewed-on: https://chromium-review.googlesource.com/628401
Commit-Queue: Łukasz Anforowicz <lukasza@chromium.org>
Reviewed-by: Łukasz Anforowicz <lukasza@chromium.org>
Cr-Commit-Position: refs/heads/master@{#496748}
[modify] https://crrev.com/20196d32b0b98118006a080f9da90f0a09b1c15a/third_party/WebKit/LayoutTests/FlagExpectations/site-per-process

horo@, could you PTAL?

On my local machine, when running the test with --site-per-proces I get a "text diff" failure (covered by LayoutTests/TestExpectations) before your r484009 and I get a timeout after this change.  This change seems like a valid suspect, although please note the strange discrepancy between local VS bot repro noted in #c2 and #c3.

Also - I wonder if you could also take a look at  issue 599670  (the "text diff" expectation is present uninvestigated for more than a year;  

Even before my patch r484009, there was a bug with Site Isolation and DevTools.
When Site Isolation is enabled, the document resource of iframe is not correctly
shown in the DevTools network tab.

Reproduce step:
(1) Go http://horo-t.github.io/tmp/20170828/index.html
(2) Open DevTools network tab.
(3) Click "Submit" button which opens
    https://horo-t.github.io/tmp/20170828/index.html in a iframe.
    (Note: http"s")

Expected: "index.html" is shown as loaded state in the DevTools network tab.

Actual:
 (1) Before my patch r484009
   (1-1) with SiteIsolation (--site-per-process)
     (1-1-1) without PlzNavigate (--disable-features=browser-side-navigation) 
       - "index.html" is shown as "canceled" state.
     (1-1-2) with PlzNavigate (--enable-features=browser-side-navigation)
       - "index.html" is not shown.
   (1-2) without SiteIsolation: Working as expected
   
 (2) After my patch r484009
   (2-1) with SiteIsolation (--site-per-process)
     (2-1-1) without PlzNavigate (--disable-features=browser-side-navigation) 
       - "index.html" is shown as "pending" state.
     (2-1-2) with PlzNavigate (--enable-features=browser-side-navigation)
       - "index.html" is not shown.
   (2-2) without SiteIsolation: Working as expected

In (1-1-1), the request is canceled when the renderer process received 
FrameMsg_SwapOut IPC message. FrameFetchContext::DispatchDidFail() calls
"probe::didFailLoading(GetFrame(), ...);" and
InspectorNetworkAgent::DidFinishLoading() calls GetFrontend()->loadingFinished().
So the request is show as "canceled" in DevTools.

My patch r484009 changed "probe::didFailLoading(GetFrame(), ...);" to
"probe::didFailLoading(GetFrame()->GetDocument(), ...)" which can't call 
InspectorNetworkAgent::DidFinishLoading(), because GetFrame()->GetDocument()->GetFrame()
is null now.
So GetFrontend()->loadingFinished() is not called and the request is show as
"pending" in DevTools.

I haven't investigated what happens with PlzNavigate and Site Isolation.
But anyway, is this bug a known issue?

The request not showing up with PlzNavigate and --site-per-process is a known problem, see  issue 750901 .  I think we've known about issues with the behavior without PlzNavigate as well.  There's  issue 750898  for cookies not showing up; not sure if there's a separate bug filed for the wrong request state.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/62db39d8dec89187c2518d0046b7ca28d9fc747d

commit 62db39d8dec89187c2518d0046b7ca28d9fc747d
Author: Lukasz Anforowicz <lukasza@chromium.org>
Date: Wed Oct 04 23:23:03 2017

Disable exceptions for tests that have "healed" themselves.

Bug:  758075 ,  700535 ,  582522 ,  602497 ,  616905 
Bug:  623268 ,  623265 ,  678492 
Change-Id: I1bdc648c9aeb97cc1d16b0d8693d04e0c76030de
Reviewed-on: https://chromium-review.googlesource.com/700983
Commit-Queue: Łukasz Anforowicz <lukasza@chromium.org>
Reviewed-by: Alex Moshchuk <alexmos@chromium.org>
Reviewed-by: Nasko Oskov <nasko@chromium.org>
Cr-Commit-Position: refs/heads/master@{#506557}
[modify] https://crrev.com/62db39d8dec89187c2518d0046b7ca28d9fc747d/third_party/WebKit/LayoutTests/FlagExpectations/site-per-process

Both Site Isolation FYI bots are green - closing this as fixed.