Hello, Guix!

This patch provides the basis for 'guix deploy', implementing what I've
referred to as the "simple case" in my progress reports: in-place
updates to machines (physical or virtual) whose name and IP address we
know well. Do note that these commits depend on Ludovic's implementation
of 'remote-eval'.[1]

There's certainly more to be done with this -- the GSoC period is far
from over, and I'm hoping to use that time to implement more complex
use-cases such as automatically provisioning virtual machines in the
cloud. I'm submitting a patch series now per the recommendation of my
mentors to break the project into a few chunks to submit over the
duration of the summer.

Quite a bit has changed since my last email about this.[2] For one,
GOOPS is no longer used. Machine declarations now look just like any
other sort of declaration in Guix.

#+BEGIN_SRC scheme
(use-modules (gnu) (guix))
(use-machine-modules ssh)
(use-service-modules networking ssh)
(use-package-modules bootloaders)

(define %system
  (operating-system
   (host-name "gnu-deployed")
   (timezone "Etc/UTC")
   (bootloader (bootloader-configuration
                (bootloader grub-bootloader)
                (target "/dev/vda")
                (terminal-outputs '(console))))
   (file-systems (cons (file-system
                        (mount-point "/")
                        (device "/dev/vda1")
                        (type "ext4"))
                       %base-file-systems))
   (services
    (append (list (service dhcp-client-service-type)
                  (service openssh-service-type
                           (openssh-configuration
                            (permit-root-login #t)
                            (allow-empty-passwords? #t))))
            %base-services))))

(list (machine
       (system %system)
       (environment 'managed-host)
       (configuration (machine-ssh-configuration
                       (host-name "localhost")
                       (identity "./id_rsa")
                       (port 2222)))))
#+END_SRC scheme

There are a number of other differences here as well. For one, the SSH
configuration now has an 'identity' field for specifying a private key
to use when authenticating with the host. Any key management scheme you
might have set up in '~/.ssh/config' will also work if the 'identity'
field is omitted.

The 'environment' field is where we declare how machines should be
provisioned. In this case, the only type of provisioning that's been
implemented is 'managed-host' -- the "simple case" of in-place updates
to a machine that's already running GuixSD. The parameters for
provisioning are given in the form of an environment-specific
configuration type. In the example, this is 'machine-ssh-configuration',
which describes how 'guix deploy' should make an SSH connection to the
machine. I'm sure you can imagine something along the lines of a
'machine-digitalocean-configuration', describing some parameters for a
droplet.

There are two things in this patch series that I'd like comments on in
particular.

First, I still haven't figured out the whole testing situation. The
tests, as of now, spin up a virtual machine, create a machine instance,
deploy that to the virtual machine, and then make assertions about
changes made to the system. These tests were originally in the system
test suite as they deal with virtual machines, but I've since moved it
into the normal Guix test suite because of how much needs to be done on
the host side -- I spent an absurd amount of time trying to fit a call
to 'deploy-machine' into a derivation that could be run by the system
test suite, but I just wasn't able to make it work. I'm hoping someone
will have thoughts about how we can test 'guix deploy'. Should we have
them disabled by default? Is there some way to implement them in the a
system test suite that I've overlooked? Should the tests be included at
all?

Second, I'd like some suggestions on how to go about the documentation.
I have a cursory description of how to invoke the command-line tool, and
an example of a deployment specification, but I'm wondering if the
documentation should be split up into multiple sections across the
manual -- especially if we're going to have multiple 'environment' types
with their own configuration records down the line.

I look forward to your comments.

Regards,
Jakob

[1]: https://lists.gnu.org/archive/html/guix-patches/2019-06/msg00201.html
[2]: https://lists.gnu.org/archive/html/guix-devel/2019-06/msg00078.html

David Thompson (1):
  Take another stab at this whole guix deploy thing.

Jakob L. Kreuze (5):
  ssh: Add 'identity' keyword to 'open-ssh-session'.
  gnu: Add machine type for deployment specifications.
  Export the (gnu machine) interface.
  Add 'guix deploy'.
  doc: Add section for 'guix deploy'.

 Makefile.am             |   4 +-
 doc/guix.texi           | 103 +++++++++
 gnu.scm                 |   8 +-
 gnu/local.mk            |   5 +-
 gnu/machine.scm         |  89 ++++++++
 gnu/machine/ssh.scm     | 355 +++++++++++++++++++++++++++++++
 guix/scripts/deploy.scm |  90 ++++++++
 guix/ssh.scm            |   3 +-
 tests/machine.scm       | 450 ++++++++++++++++++++++++++++++++++++++++
 9 files changed, 1103 insertions(+), 4 deletions(-)
 create mode 100644 gnu/machine.scm
 create mode 100644 gnu/machine/ssh.scm
 create mode 100644 guix/scripts/deploy.scm
 create mode 100644 tests/machine.scm

-- 
2.22.0