1192485 - mozregression can't handle independently terminated process

Status: RESOLVED FIXED

(Testing :: mozregression, defect)

Description

[Ray Satiro]

User Agent: Mozilla/5.0 (Windows NT 6.1; Win64; x64; rv:42.0) Gecko/20100101 Firefox/42.0
Build ID: 20150807030210

Steps to reproduce:

I'm using mozregression in both the command line 0.40 with Python 2.7.10 in Windows 7 x64 and also in the GUI 0.3 using mozregression 0.37. I'm establishing regression windows for Firefox but sometimes it terminates or I accidentally close it and the regression stops working.



Actual results:

In the case of the GUI it just hangs and nothing happens. In the case of the command line tool it says "OSError: Could not terminate process" and exits. This made establishing a regression window very difficult.


Expected results:

I think you should add something in both so that if a process terminates (either due to user or firefox terminates or crashes etc) the regression can still continue. I must've restarted my regression six times although with a smaller window each time.

Comment 1

[Julien Pagès (:parkouss)]

Ray, Thanks a lot for reporting this issue!

Hm, I often close manually the browser (on Linux, less often on windows) and it works well in "normal" cases.

I suspect it is more in case the browser terminates in a hard way or crashes on Windows - and this is not expected by mozprocess:

https://dxr.mozilla.org/mozilla-central/source/testing/mozbase/mozprocess/mozprocess/processhandler.py#142

I will work on this (mozregression should definitely not crash for this), but I would like to be sure that the normal case (ie, the user closing the browser by clicking on the cross) works for you. Else, there is something else that I don't get yet. In that case, please copy paste a stack trace of the error you got with the command line, and provide me the exact command line you use and how you are closing the browser so I can try to reproduce and investigate.

Comment 2

[Julien Pagès (:parkouss)]

Attachment: handle errors when running the tested application

This is the command line only part of the equation. Some additional code is required for the GUI.

mozregression should not crash when trying to run a build now. In worst case, it will skip the erroneous build.


Copy paste of my commit message:

This handle cases when we got some exception while:

    1. trying to install the application
    2. trying to start the application
    3. trying to stop the application.

For cases 1 and 2, this is an "uncoverable" error. So the raw
exception is printed, and we force skipping the build.

Case 3 can be recovered, so the only thing we do is print the raw
error, but we keep the verdict that the user entered and the
app_info dict extracted from the installed application.

Comment 3

[William Lachance (:wlach)]

Comment on attachment 8645629 [details] [review]
handle errors when running the tested application

This looks great, thanks!

Comment 4

[Julien Pagès (:parkouss)]

Thanks Will, merged in https://github.com/mozilla/mozregression/commit/96e4739f75d5481914b4489f29ab7208ea3cd229.

Comment 5

[Ray Satiro]

I upgraded to mozregression-0.41. Although I still see "OSError: Could not terminate process" messages mozregression no longer terminates and I am able to continue the bisect. Thanks for your work fixing this issue. 

I notice I see this messages quite often:

Exception trying to use job objects;
falling back to not using job objects for managing child processes
Traceback (most recent call last):
  File "C:\Python27\lib\site-packages\mozprocess\processhandler.py", line 295, in _execute_child
    winprocess.AssignProcessToJobObject(self._job, int(hp))
  File "C:\Python27\lib\site-packages\mozprocess\winprocess.py", line 51, in ErrCheckBool
    raise WinError()
WindowsError: [Error 5] Access is denied.

I think if that fails the first time it should know it probably won't succeed any other time and fall back by default.

Also I realize this might not be the appropriate place but since I have your ear I was wondering where the best place is to ask questions about mozregression? Is there a forum or a board or a list?

Comment 6

[Julien Pagès (:parkouss)]

I'm glad the patch helps!

So looking at the stack trace, this seems to be a mozprocess error to me. We should report that in the right component.

I think the best place to ask questions about mozregression is on irc (irc.mozilla.org, channel #ateam). You can also find some information here:

http://mozilla.github.io/mozregression/


:wlach, do you agree with my answer here ? And should we open a mozprocess bug ?

Comment 7

[William Lachance (:wlach)]

(In reply to Julien Pagès from comment #6)

> :wlach, do you agree with my answer here ? And should we open a mozprocess
> bug ?

Yeah, I think resolving this bug and opening a mozprocess one makes sense. :)

Comment 8

[Julien Pagès (:parkouss)]

Well, I suppose it is related to bug 1188863, so I won't open a new bug. Feel free to open a new one if I'm missing something.