[Qa-jenkins-dev] setup up the fdroid build env (Re: pull request

[Hans-Christoph Steiner]

Holger Levsen:
> Hi,
> 
> On Donnerstag, 11. Februar 2016, Hans-Christoph Steiner wrote:
>>> I've now commited + deployed another attempt of fixing this. Triggering a
>>> job run now…
>>
>> Ok, we're back to where we were before with the fail on timeout.  My
>> guess is that its trying to use an old broken instance.  Can you delete
>> the old vagrant/VirtualBox setup?  Those both should now be totally
>> self-contained in the jenkins job, and wiped on each build.  That is
>> done like this:
>>
>> rm -rf ~/.vagrant.d
>> rm -rf ~/VirtualBox\ VMs
>> rm -rf ~/.config/VirtualBox/
> 
> Doing so now. I'm wondering whether we want a job (which is never run 
> automatically) for doing this more easily and reliable…

I do that now in ./jenkins-build-makebuildserver.  it forces vagrant and
VirtualBox to put their home dirs into the WORKSPACE, so they'll be
wiped when WORKSPACE is wiped.  Our jenkins box runs a `git clean` every
time it is run, so starting from scratch.


>> Just be sure nothing else is using vagrant or VirtualBox, since this
>> would delete all the settings and VMs!
> 
> there's nothing.
> 
>> If that fails, then we'll have to launch the VirtualBox GUI to see where
>> the failure is. It'll work over X11 forwarding.
> 
> one potential difference between your and this setup which came to my mind: 
> we're using jessie *without* systemd atm, because there was an issue with 
> schroots. Are you using systemd or sysv?

Plain jessie, so systemd.  I think at this point, if the VirtualBox
kernel modules are loading on the box where the job is run, then the
next step is to fire up the VirtualBox GUI.


>> > As for http_proxy, env vars are not cleared, but then also not sent to
>> > the VM, where apt is running. The http_proxy is for .deb caching,
>> > right? Since that's already happening locally, so this shouldn't be
>> > using much bandwidth.
>
> Does that mean there's no proxy in use by the VM? That's not nice.

I don't know if VirtualBox automatically uses proxies from an env var.
The most manageable thing probably would be to force all of the
VirtualBox's net traffic through a SOCKS proxy.  But I think this is
also likely a case where it'll be hard to get this setup, but with only
minimal benefits.  The cached jessie32.box provides a reasonably
up-to-date jessie install, and the apt cache is stored between job runs.
 So really, the full git clones of fdroidserver.git and fdroiddata.git
are probably a lot more bandwidth than the apt downloads.

.hc