GitLab

I discovered that CI runners could be manually activated for multiple
projects. It means I don't need to add a config section for each runner,
and can instead have a global one.

Playbook to use with ./chi-appliance configure to setup a CI
infrastructure on top of Chameleon nodes.

This playbook:
 - installs gitlab-runner
 - adds role to install or run ci on aml

I'm reluctant to add CI tokens in plaintext inside the infra repo.
Instead, ansible-vault is used to keep them encrypted.

The openstack client libraries bury the actual error message reported by
the openstack servers inside a Requests.Response object. This patch
makes sure we retrieve this message, as it is more useful than the top
level one.

We don't have a use for a appliance-local ssh key for now, so let's not
add that trouble to our setup.

Just enough tasks and templates to end up with an idle gitlab CI runner
installed and running on the node.

It should make it easier to organize tasks, and give us easy access to
"become" for an entire group of tasks.

[fix] enforce region for blazar client

See merge request !3

Since Chameleon changed to a multi-region setup the blazar client is
acting up. It looks like the current upstream doesn't support that
feature by default.

In theory, the session already contains the region info, but since the
client creation code doesn't use it by default, we force it. This is
adapted from the Chameleon-recommended client library.

Ansible improvements

See merge request !2

Until we figure out a way to add host keys to the config without
trouble, disable host key checking when calling ansible.

Not doing so could result in blocked calls, as ansible was asking
questions during the run.

Usefull for users that want to try a different set of roles.

Ansible allows us to wait for the node to become available, which should
allow us to launch the `configure` command as soon as `create --wait`
completes.

Uses default wait_for_connection module, which should poll for 10
minutes.

Requires that we disable gather_fact, and reactivate it after the
connection succeeds.

Stack create and delete take an option to wait for the operation to
complete. It doesn't actually wait for the operation to complete, just
for the Openstack side of it to be done. For creation for exemple, the
nodes might still be booting.

Ansible recommend a specific directory layout to deal with complex
playbooks. As we expect this configuration to become trickier, let's
adopt good practices right away.

Previous code was blocking on stdout after subprocess completed with
success, as `poll()` would return 0, and `not 0` is true.

Region support

See merge request !1

Reflecting changes in OpenRC file after Chameleon update on 2018/05/29.

Just realized that the blockquotes were not the right format for command
lists. Should be using code blocks instead.

Add a script to help with general configuration issues.

More doc to explain what is going on in these scripts.

ansible/inventory.yaml is generated by `chi-appliance configure`

Supporting the two regions of chameleon (TACC & UC) requires changes
in the clouds.yaml/secure.yaml format, to avoid having a secure file
with twice the info. We use the region support in os-client-config for
that.

I had to change the scripts a little bit, as the cloud name is now fixed
but the region_name has to be specified.

Forgot to add the pipenv install command to make sure that the
dependencies are installed on a new machine.

Thanks to @tpatki

Will allow appliance management more easily, in a script separated from
the lease stuff.

Uses ansible under the hood to configure the appliance.

This is not ideal. Ansible has tons of issues when we launch it from
there, and the configuration become tedious.

It might be easier to deploy ansible on the frontend and rerun
everything inside, or to create a python script that deals with ansible
directly.

Add an option to create haswell allocations, but default to skylake
nodes.

Skylakes are required for power management.

Using pipenv as the environment manager, implement as clouds.yaml
workflow and a script to manage leases on chameleon.

Note that the clouds.yaml only support chi-tacc for now. We will need to
add another region for UC later.

Derived from the complex appliances documentation.
I'm still in the process of understanding the whole softwareconfig part,
but it's a start.