[Feature Request] Add Vagrant Executer Support

I think it would be a great benefit to add Vagrant executor support as it supports multiple vm types and it is how some developers support their development environment and would want to emulate it on the CI.

It would operate by knowing when the vagrant vm is up, running and stopped. And would run the commands via vagrant ssh -c ''.

While I have manage to this with set of specific commands, when a build is canceled or randomly stalled, the vagrant vm is never stopped or destroyed causing server resource issues.

Added feature proposal label

@bass_rock I'd love to chat with you about how you're doing this currently -- it looks like we have the same desired outcome. Where's best to reach you to collaborate?

@AlSegno hangouts works. bassrock408@gmail.com

@bass_rock invite sent.

+1 on this!

One of the benefits of having vagrant, is that one can also use the vagrant catalogue of images for virtualbox, vmware, esxi, parallels, etc. - so it would be docker-like but with VMs instead of containers. So for GitLab, having a vagrant runner makes sense to support all the underlying virtualisation technologies, and, for the user it's a benefit because one could reuse much of the VM stock available for Vagrant (especially for OSX, iOS and Windows builds).

Would like to collaborate on this, if possible.

I've been running Vagrant VMs under Jenkins for a while with the same basic resourcing issues, looking to switch some over to Gitlab-CI and decided to consolidate some scripts.

https://github.com/neerolyte/vagrant-sad is a first pass, I'll probably add more functionality in the future, thought some people following this issue might be interested as an intermediate solution though.

Added executors label

Milestone changed to %Backlog

mentioned in issue #1349

mentioned in issue #1658 (closed)

Was there any work already done with this? vagrant-sad sounds interesting, but how do I integrate it into GitLab CI?

I have the following in .gitlab-ci.yml for most jobs running vagrant stuff:

before_script:
  - git clone https://github.com/neerolyte/vagrant-sad.git
  - max_run_age=0 ./vagrant-sad/clean-up-vbox-vms
  - ...any other before stuff...

Also a specific job set up with:

max_run_age=0 max_off_age=0 ./vagrant-sad/clean-up-vbox-vms

as the script and then use triggers from cron to run that - so that it never the more intensive job runs when nothing else is running (we currently have one CI runner capable of running Vagrant VMs and concurrency limited to 1 on that runner).

This is not a fully scalable solution, but it's adequate "for now".

Thank you @neerolyte ; how do you provision the boxes ? I was expecting a vagrant up somewhere ?

So provisioning is done exactly however it's done when running Vagrant manually, i.e one of our stacks is called "php56" and has:

php56:
  tags:
    - shell-vagrant
  script:
    - vagrant up php56
    - vagrant ssh php56 -c "... steps to run tests ..."
    - vagrant destroy -f

To be clear, none of this is particularly tightly integrated with GitLab CI - just enough to get it working.

just an idea : if the existing VirtualBox executor was enhanced so that the VM to be run could be specified in the gitlab-ci.yml, wouldn't it be close to a vagrant support ?

@bbdb I don't see how. Vagrant is a way of provisioning multiple VMs together and having them get predictable network, file shares, CPU etc. The Virtualbox runner would require a different base box (i.e we may not have git in the Vagrant base box) and I assume can only run a single VM per job?

It'd be like asking if providing a LXC runner was close to docker support.

I was not clear, sorry. The nice thing I see with vagrant support is that a runner can be configured to run a vm selected in the gitlab-ci file, at runtime. On the other hand, the virtualbox executor tights the runner with a specific VM, at runner configuration time. I was just wondering if the virtualbox executor could be enhanced to allow selection of the VM at run time.

+1

Just like docker executor, it would be possible to define a vagrant box to use, or use the one defined in config.toml.

job2:
  stage: build
  box: ubuntu/trusty64
  before_script:
    - sudo apt install -y ruby
    - bundle install
  script:
    - bundle exec rake spec
  tags:
    - vagrant