On Tue, 18 Feb 2020, Dmitry V. Levin wrote:

> On Tue, Feb 18, 2020 at 06:39:01AM +0300, Ivan Zakharyaschev wrote:
> > This is needed not to send requests to nginx before it is ready to
> > listen: when the main process forks and exits, nginx is considered to
> > be ready.
> > 
> > Without this change, I observed a failure in the http test, but not
> > the https one, probably, because the delay in the https test was
> > larger due to an extra package being built:
> > 
> > Run Testcase (22/29) test-apt-method-http
> > Building package: simple-package
> > Test for successful execution of apt-get update …
> > Err http://localhost x86_64 release
> >   Could not connect to localhost:8080 (127.0.0.1). - connect (111 Connection refused)
> > 
> > However:
> > 
> > Run Testcase (23/29) test-apt-method-https
> > Building package: simple-package
> > Building package: conflicting-package-one
> > Test for successful execution of apt-get update … PASS
> > Test that package(s) are not installed with rpm -q simple-package … PASS
> > Test for successful execution of apt-get install simple-package … PASS
> > Test that package(s) are installed with rpm -q simple-package … PASS
> > Pinning invalid key in apt
> > Test for failure in execution of apt-get update … PASS
> > Test that package(s) are not installed with rpm -q conflicting-package-one … PASS
> > Test for failure in execution of apt-get install conflicting-package-one … PASS
> > Test that package(s) are not installed with rpm -q conflicting-package-one … PASS
> 
> Apparently, this fix is insufficient, I experience sporadic failures
> in these nginx-based tests.  Something racy is still there.

It's a bit difficult for me to help to find out the reason, because I 
don't know how to reproduce any other failures than those fixed by my 
patch...

-- 
Best regards,
Ivan