Re: how to create and test a new service definition

[ng0]

On Sun, 14 May 2017, ng0 wrote:

> On Sun, 14 May 2017, Catonano wrote:
>
> >  2017-05-13 22:06 GMT+02:00 Ludovic Courtès <address@hidden>:
> >
> > >  Vincent Legoll <address@hidden> skribis:
> > >
> > > > >  The best way to test your code is to write an ‘operating-system’
> > > > >  declaration that uses the new service and to instantiate it in a VM
> > > > >  with
> > > > >  ‘guix system vm’.
> > > >
> > > >  Should that be working properly (out-of-the-box) when you're already in
> > > >  a qemu VM (recursive virtualization) ?
> > > >
> > > >  I ask because I'm getting:
> > > >
> > > >  [...]
> > > >  ERROR: qemu failed "qemu-system-x86_64"
> > >
> > >  What were the lines above this one?  This tool tries to use KVM if it
> > >  seems available.  Maybe in your case it “seems” to be available (as in
> > >  /dev/kvm exists) but is actually unusable?
> > >
> > > > >  Once you’ve done that, you can also write an automated test for the new
> > > > >  service; see <https://gnu.org/s/guix/news/guixsd-system-tests.html>.
> > > >
> > > >  I'm far from there, I have a *really* hard time getting back to guixsd.
> > >  For
> > > >  instance it took me very long time to find back the GUIX_PACKAGE_PATH
> > > >  env var. This looks under-documented, or I don't understand how one is
> > > >  to
> > > >  work on custom or new packages, etc...
> > >
> > >  ‘GUIX_PACKAGE_PATH’ is documented at
> > >  <https://www.gnu.org/software/guix/manual/html_node/Package-
> > >  Modules.html#index-GUIX_005fPACKAGE_005fPATH>.
> > >
> > >  The workflow for defining packages is described at
> > >  <https://www.gnu.org/software/guix/manual/html_node/Defining-Packages.html
> > > >  ,
> > >  and that for contributing them is at
> > >  <https://www.gnu.org/software/guix/manual/html_node/
> > >  Submitting-Patches.html>.
> > >
> > >  There’s probably room for improvement though.  What changes/additions
> > >  would you suggest?
> >
> >  Let me chime in here: I'd suggest to state the workflow or defining
> >  services, as the one or deining packages is stated
> >
> >  But I'm not thinking about how to define and use The scheme structures.
> >  The
> >  examples are enough for that.
> >
> >  I'm thinking about things like how to deal with qemu, if you want to test
> >  your services in a vm
> >
> >  Is there any other way to get your feet wet with services ?
> >
> >  George suggested me the trick to link a checked out master branch to
> >  .local/guix/profile (or maybe it's the other way around: it's liinking
> >  .local/guix/profile to a checked out master branch)
> >
> >  Is anyone using this to work on services ? How, exactly ?
> >
> >  Thanks
>
> My current way of testing services is usually on bare-metal.
> I had bad experiences with one network based service (gnunet)
> and it being terrible to debug, so I took the QEMU out of there.
> I test things this way which will not break my current generation like 
> darkhttpd or the gopherd I submitted as a question.
> Other, potential boot breaking, stuff I test on other systems.
>
> This is also because I never really figured out the best way
> to run network based services in qemu in Guix. I have no issues
> with qemu, but on Guix many months back it behaved unlike the
> qemu I knew from Gentoo.

I have to append to this: I tried debugging darkhttpd with a shared VM
now and it is much more useful to get a bourne-like shell than
a frozen system if you make a syntax error in the container description.

So that's definitely a plus for the VM.

> There is lots of room for services in the documentation.
> There should be a "contributing services" section in chapter 8,
> currently I learn this the way I learned to do packages, by
> repetition and failure until I see how to debug stuff.. which is already 
> helping since I am starting to see patterns for debugging with darkhttpd (it 
> helped that it simply worked and then randomly broke in further attempts).