878479 - Win7 tests on try fail with baffling timeouts when hgweb doesn't serve machine-configuration.json

Status: RESOLVED WORKSFORME

(Release Engineering :: General, defect)

Description

[Randell Jesup [:jesup] (needinfo me)]

See https://tbpl.mozilla.org/?tree=Try&rev=a652c78969d2 for an example.  This is a big problem for me; I need to retrigger some tests on Win7 to try to verify a fix for a beta uplift.

Comment 1

[Randell Jesup [:jesup] (needinfo me)]

Downloading a build and trying to install it gives "This file does not have a valid digital signature..." and publisher says "Unknown Publisher".  Per Philor, this will cause tests to fail to run, but rail pointed out it may be failing setting the resolution.

No Win7 tests are starting (mochitests or crashtests).  You can see my Try pushes or Monica Chew's from earlier in the morning.  

I see these errors before the timeouts...

10:43:41     INFO - ##### Running run-tests step.
10:43:41     INFO - #####
10:43:41     INFO - Running pre test command run mouse & screen adjustment script with 'c:\mozilla-build\python27\python.exe ../scripts/external_tools/mouse_and_screen_resolution.py --configuration-url http://hg.mozilla.org/%(repo_path)s/raw-file/%(revision)s/testing/machine-configuration.json'
10:43:41     INFO - Running command: ['c:\\mozilla-build\\python27\\python.exe', '../scripts/external_tools/mouse_and_screen_resolution.py', '--configuration-url', u'http://hg.mozilla.org/try/raw-file/a652c78969d2/testing/machine-configuration.json'] in C:\slave\test\build
10:43:41     INFO - Copy/paste: c:\mozilla-build\python27\python.exe ../scripts/external_tools/mouse_and_screen_resolution.py --configuration-url http://hg.mozilla.org/try/raw-file/a652c78969d2/testing/machine-configuration.json

Comment 2

[Phil Ringnalda (:philor)]

Yeah, I was totally on the wrong track with signing, and the actual problem is that in http://mxr.mozilla.org/build/source/mozharness/external_tools/mouse_and_screen_resolution.py#30 we don't give urllib2 a timeout value, which means it takes the default value from httplib, which has no default value so it takes the default value from socket, which has a default timeout of infinity, so when hgweb isn't serving raw-files and instead is timing out (hellishly frequent until now, who knows whether the hg upgrade will solve it), mouse_and_screen_resolution.py will wait forever, and we'll just get a buildbot timeout, instead of a nice "Could not fetch url '%s'".