Issue 2406 - Update pinned Nginx version to 1.8.0-1~precise

See https://issues.adblockplus.org/ticket/2406 for more information.

[Felix Dahlke, 2015-04-23 07:44:09]

LGTM, assuming we roll this out carefully.

[mathias, 2015-04-23 08:23:21]

On 2015/04/23 07:44:09, Felix H. Dahlke wrote:
> LGTM, assuming we roll this out carefully.

I discovered this because my Travis-CI builds have failed
(https://travis-ci.org/mjhennig/adblockplus-infrastructure/builds/59648495).
After verifying that this actually affects OUR servers, their OS and -version...

mathias@server4:~$ sudo apt-get update # ... rödel
mathias@server4:~$ apt-cache show nginx | head -2
Package: nginx
Version: 1.8.0-1~precise

... and preparing this patch-set, I've first checked that Travis-CI now works as
expected
(https://travis-ci.org/mjhennig/adblockplus-infrastructure/builds/59651470), and
just subsequently put this in review.

To make a long story short: Of course we will be careful, and actually that
being careful does not start at roll-out time - but way earlier... :)

[Wladimir Palant, 2015-04-23 11:19:51]

On 2015/04/23 08:23:21, matze wrote:
> ... and preparing this patch-set, I've first checked that Travis-CI now works
as
> expected

This means exactly nothing. Provisioning didn't fail, that's great. It doesn't
mean however that our nginx configuration continues to work as it should.
Personally, my idea was waiting for nginx 1.8.1 before rolling this out - the
current version has lots of changes that haven't been tested by a wide audience
yet. We'll need time to test the VMs anyway.

[mathias, 2015-04-23 11:28:57]

On 2015/04/23 11:19:51, Wladimir Palant wrote:
> On 2015/04/23 08:23:21, matze wrote:
> > ... and preparing this patch-set, I've first checked that Travis-CI now
works
> as
> > expected
> 
> This means exactly nothing. Provisioning didn't fail, that's great. It doesn't
> mean however that our nginx configuration continues to work as it should.
> Personally, my idea was waiting for nginx 1.8.1 before rolling this out - the
> current version has lots of changes that haven't been tested by a wide
audience
> yet. We'll need time to test the VMs anyway.

I did not claim that this is enough testing. But I do claim that this is part of
a careful process in the first place. So, "means exactly nothing" is a bit
insulting. And of course there is more testing required, and I am actually
working on that right now. It just would make no sense if the provisioning of
any Nginx-using host failed in the first place.

By the way, because of our current Puppet code and since we use the fabulous
Ubuntu, there is no short-term choice for us other than upgrading. The so-far
used version has vanished from the in-use APT resource, which caused the issue
reported in #2406 in the first place. I believe I mentioned before that many
Ubuntu package maintainers "latest s**t from head" practice is one of the major
reasons why I would never recommend using Ubuntu on a server in the first place.