Prior to this commit, we placed modules local to a users installation
in the `site` directory. This was just a convention and the name
`site` doesn't clearly convey what it is for.
After this commit, we place modules local to a users installation in
the `site-modules` directory. This makes it more clear to users
that this is a directory that modules go i. When users start
with bolt they won't even know what a control-repo is and
renaming site to site-modules gives them a better idea of why
they should put their modules with tasks in them. Also see:
https://tickets.puppetlabs.com/browse/BOLT-1108
r10k generates a .r10k-deploy.json file since version 2.1.0 which was
released on October 28, 2015. New users of the control-repo are not
likely to have a so old version of r10k, so remove this dead code.
The appropriate ruby interpreter is determined by the config_version.sh
shell script which explicitely use it to start these ruby scripts.
Removing the execute bit ensure users will not run these script with the
wrong Ruby version.
- Fix shebang: `bash` is not always in `/bin/`, and since the script
does not have bashism, rely on `sh` which is always in `/bin/`;
- Use `/opt/puppetlabs/puppet/bin/ruby` if this file exist and is
executable, otherwise use `ruby` from $PATH;
- Use `code_manager_config_version.rb` if `.r10k-deploy.json` is found,
and `config_version.rb` in all other cases.
This commit moves the "where did all the previous code go" section to
the bottom as it's been a while since that change was made. Nowadays,
people new to Control Repositories will find this and won't understand
the reference as they never knew about previous versions.
Now the README starts right away with information on what this project
is and how to use it.
Also cleaned up some of the Markdown syntax to make it easier to read.
This is mainly a style and readability change.
Prior to this, on masters whose hostname is actually their FQDN, the
config_version script would show the entire FQDN. On nodes with really
longs FQDN's, it was not very nice to look at.
This takes the hostname of the master, splits it on dots (.) and takes
the first segment.
Now this: compile-master-02.int.lab.dmz.company-name.net-production-48fd18ab
Is this: compile-master-02-production-48fd18ab
Prior to this, the config_version.rb script (used for r10k) attempted to
use the system ruby to parse the script. This caused problems on Puppet
masters that don't have `ruby` in PATH.
This fixes that by hardcoding the puppet-agent's ruby in the shebang.
Prior to this, modules that were deployed with r10k into the ./modules
directory weren't being ignored by git.
When doing local development or testing, it's nice to be able to run
'r10k puppetfile install' to pull down modules from the Puppetfile.
After this commit, those modules won't be tracked by git.
This commit enables the control repo to use Hiera 5 environment-level
hiera hierarchy. This means adding a hiera.yaml to the repo, and moving
hieradata/ => data/.
We should do this to the control-repo template new customers base off of
because in a Hiera 5 world, the global hiera.yaml should be very minimal
(possibly even ONLY having the console level), and everything else
(nodes, common) belongs in the environment hiera.yaml.
This control-repo template is how people start using Puppet. It should
reflect using our most modern technologies.
Prior to this, the config_version script just showed the commit ID of
the version of code being compiled. This commit includes the compiling
Puppet master's hostname and environment name in the config_version.
This is very useful for debugging when a Puppet master is failing and
you have multiple masters behind a load balancer.
The output of config_version now looks like this:
pupmaster01-production-ac9785273a10
Without this patch the control repository does not contain any scripts
to execute CI jobs for Puppet code deployment. This patch addresses the
problem by adding a Gitlab CI job configuration with three jobs defined
across two stages. The test stage executes first with a Syntax and a
Lint job executing in parallel. If an environment branch has been
pushed, one matching `developemnt`, `testing`, `production` or starting
with the prefix `playground`, the deploy stage is executed.
The deploy stage requires a Gitlab secret environment variable,
PUPPET_TOKEN. This environment variable is the puppet access token
configured with Code Manager.
The test stage runs the following syntax checks:
* yaml files (*.yml, *.yaml)
* json files (*.json)
* bash scripts (*.sh)
* ruby scripts (*.rb)
* puppet manifests (*.pp)
* erb files (*.erb)
The behavior of the checks are to check only files modified relative to
a base branch, defaulting to `production`. This avoids running syntax
checks against files which have not been modified by the merge request.
The check uses `git diff --name-status` to identify changed files across
multiple commits in a topic branch.
The lint checks rely on bundler and the Gemfile to install numerous lint
checks.
Gem libraries are installed in a per-job location in $HOME. This is an
intentional compromise between installing into a system GEM_PATH
element, which would create library conflicts with other jobs, and the
job workspace, which would cause gem libraries to be installed from the
network on each job invocation.
Prior to this commit, if you used windows bash git when you clone
down the repo these files would get incorrect permissions which
make them unexecutable.
After this commit, due to some windows bash git magic I don't
understand it appears that adding the shebang to the beginning of
the file causes windows bash git to change the permissions to
so the file is executable.
This resolves https://github.com/puppetlabs/control-repo/issues/40
Prior to this commit, we mitigated issues with hiera-eyaml causing
a memory leak by setting max_requets_per_instance to 0
After this commit, we go back to the default for
max_requests_per_instance because the hiera-eyaml memory leak
has been resolved for months if you use the newest version
Prior to this commit, the control-repo was an example of the
structure of a control repo but it also included puppet code to
help setup code manager and instructions to get that all setup
in a very specific way.
This was great for users that wanted to follow those instructions
exactly but wasn't great for people just looking for an example to
start from.
After this commit, the control-repo will be just an example once
again and a new repo somehwere else will pop up to provide the
explicit instructions on how to use the example with code manager.
There are links added to puppetlabs/control-repo to a new repo
that will have a version of the code that once lived in
puppetlabs/control-repo
Prior to this commit, there was a stash profile in the site dir
of this control-repo.
After this commit, the profile has been moved to it's own repo
where it can be used more generally and not tightly coupled to this
repository.
