1245332 - Mozregression hangs if I restart browser (in different e10s state or with add-ons disabled/enabled)

Status: RESOLVED FIXED

(Testing :: mozregression, defect)

Description

[arni2033]

Mozregression GUI version 0.8.1   using mozregression version 2.2.1

STR:
1. Start new bisection between 2016-01-01 and 2016-01-05 (firefox)
2. When browser launches, click Menu (≡) -> [?] -> Restart with Add-ons Disabled
3. When browser restarts, click "Evaluate" button in mozregression

AR:  mozregression GUI hangs for ~1-2 minutes
ER:  No hang

This is long-standing bug. I don't know if this should be wontfix'ed.

Comment 1

[Julien Pagès (:parkouss)]

Oh, this does happen on windows, but not on linux.

Mozregression should not hang for this - it is a bug we should try to solve.

Comment 2

[Julien Pagès (:parkouss)]

So, the slowdown is in mozprocess,

https://dxr.mozilla.org/mozilla-central/source/testing/mozbase/mozprocess/mozprocess/processhandler.py#137

in "self.returncode = winprocess.GetExitCodeProcess(self._handle)" line.

Not sure what we can do about that yet, this is a call in the windows API. From the mozregression side, we could possibly implement a workaround: running the mozrunner.stop() method in a thread (in launcher.py), wait a bit for its termination (say 300 ms). If it does not complete, just return to the main thread and go on, leaving that thread the required time to complete (or not in case we close the program before).

If we implement that workaround, we should only do that for the impacted OS (e.g, windows).

Comment 3

[Julien Pagès (:parkouss)]

Note for myself:

In mozzprocess, I can fix it with the following code in the if block:

>                    winprocess.TerminateJobObject(self._job, winprocess.ERROR_CONTROL_C_EXIT)
>                    rc = winprocess.WaitForSingleObject(self._handle, 350)
>                    if rc == winprocess.WAIT_TIMEOUT:
>                        self.return_code = 127  # do not wait anymore
>                    else:
>                        self.returncode = winprocess.GetExitCodeProcess(self._handle)

Though I don't know if it is a right fix for mozprocess.

Comment 4

[Julien Pagès (:parkouss)]

So, this appears in mozregression console too (windows only again).

I had a discussion with :ahal, looks like it is not trivial to fix that in mozprocess directly. I'm inclined to write a workaround in mozregression for this special case.

Comment 5

[Julien Pagès (:parkouss)]

Attachment: do not block when stopping firefox

Hey Andrew, since :wlach is not here - and that we talked about that - I'm asking you for a review :) This is quite simple (I don't think it requires any mozregression knowledge) and working well from what I have tested.

With that, the GUI will only block for 0.7 seconds max, which is acceptable.

Comment 6

[Andrew Halberstadt [:ahal]]

Comment on attachment 8715788 [details] [review]
do not block when stopping firefox

Lgtm! Left a nit comment on the PR.

Comment 7

[Julien Pagès (:parkouss)]

Thanks!

Fixed and landed in https://github.com/mozilla/mozregression/commit/5be38e968d800d99332532f712537562712e086d.