This daemon extends the ansible-pull
method of running Ansible.
It uses S3 or HTTP file transmission instead of Git to manage distribution (easy to cache), and integrates with Prometheus monitoring.
ansible-pull
assumes that you are checking out an Ansible repository from git.
This wasn't an option for us at the scale that we needed, so we turned to HTTP file distribution.
On top of scaling, we've integrated monitoring (via Prometheus) to retain the centralized view of all of our Ansible
runs and a simple REST API to enable/disable the puller and trigger a run to give more fine-grained control of rollouts.
Ansible puller expects an HTTP endpoint, or an S3 ARN that points to a tarball with Ansible playbooks, inventories, etc.
The minimal configuration would just be a config file supplying http-url
(see below).
While the defaults have been set assuming Ansible's "Alternative Directory Layout"
it should be configurable enough to support alternative setups.
Inside of the tarball, at a minimum you'll need an inventory, playbook, and a requirements.txt
file.
The requirements file will be used to populate the Python virtual environment that Ansible will run in locally.
At a minimum it needs to contain ansible
so that Ansible will be installed in the virtualenv. A pinned version is even better.
The playbook is what will be actually run.
The inventory needs to contain the hostname of the node that Ansible puller is installed on.
To support our use of an Infrastructure monorepo, Ansible-puller will loop through an entire directory looking for inventories. It will test each of these inventories for a matching hostname and run the given playbook in the first found inventory.
Given the structure:
.
├── ansible.cfg
├── inventories/
│ ├── production/
│ └── staging/
├── roles/
└── site.yml
And the config file:
{
"http-url": "https://example.com/infra.tgz",
"ansible-inventory": ["inventories/production", "inventories/staging"],
"playbook": "site.yml"
}
Setting ansible-inventory
to ["inventories/production", "inventories/staging"]
and playbook
to site.yml
would mean that the puller would search for the correct host in production
and staging
, provided that the hosts were
a part of site.yml
's run. Use the debug
option to get more insight to the process while it is running.
Config file should be in: /etc/ansible-puller/config.json
, $HOME/.ansible-puller.json
, ./ansible-puller.json
Config Option | Default | Description |
---|---|---|
http-listen-string |
"0.0.0.0:31836" |
Address/port the service will listen on. Use 127.0.0.1:31386 to lock down the UI. |
http-proto |
https |
Modify to "http" if necessary |
http-user |
"" |
Username for HTTP Basic Auth |
http-pass |
"" |
Password for HTTP basic Auth |
http-url |
"" |
HTTP Url to find the Ansible tarball. Required if s3-arn is not set |
log-dir |
"/var/log/ansible-puller" |
Log directory (must exist) |
ansible-dir |
"" |
Path in the pulled tarball to cd into before ansible commands - usually ansible.cfg dir |
ansible-playbook |
"site.yml" |
The playbook that will be run - relative to ansible-dir |
ansible-inventory |
[] |
List of inventories to operate on - relative to ansible-dir |
venv-python |
"/usr/bin/python3" |
Path to the python version you are using for Ansible |
venv-path |
"/root/.virtualenvs/ansible_puller" |
Path to where the virtualenv will be created |
venv-requirements-file |
"requirements.txt" |
Path to the python requirements file to populate the virtual environment |
sleep |
30 |
How often to trigger run events in minutes |
start-disabled |
false |
Whether or not to start with Ansbile disabled (good for debugging) |
s3-arn |
"" |
S3 location to find the Ansible tarball. Required if http-url is not set |
s3-conn-region |
"" |
S3 connection region to use. Uses the aws-sdk-go-v2 default providers if not set |
debug |
false |
Whether or not to start in debug mode |
once |
false |
Only run the configured playbook once and then stop |
This daemon uses Ansible's json
STDOUT callback to parse the results of this run for this host.
It currently produces the number of tasks that are ok, skipped, changed, failed, or unreachable.
Metric | Description |
---|---|
ansible_puller_debug |
Whether or not debug mode is enabled |
ansible_puller_disabled |
Whether or not the puller is disabled |
ansible_puller_last_success |
Last timestamp of a successful run |
ansible_puller_last_exit_code |
Last ansible run exit code |
ansible_puller_play_summary |
Ansible metrics: changed, failures, ok, skipped, unreachable |
ansible_puller_run_time_seconds |
How long Ansible took to run to completion |
ansible_puller_running |
Whether or not the puller is currently running |
ansible_puller_runs |
How many times the puller has run |
ansible_puller_version |
Version (git sha) of the puller |
Enabling MD5 checksumming will prevent extraneous calls to download the ansible tarball from the remote.
By design, ansible_puller will look at the remote path <resource_path>.md5
to discover the live
MD5 checksum. If, for example, your resource is located at https://example.com/some/file.tgz
then
ansible_puller will look for the MD5 hash at https://example.com/some/file.tgz.md5
. The following
conditions will lead to a (re-)download of the ansible tarball:
- There is no current ansible tarball at the specified local path
- The current hash of the local ansible tarball not match the remote checksum
- The remote checksum does not exist
If a remote checksum exists then the downloaded tarball will be hashed and the resulting output will be compared to the remote checksum to validate artifact integrity.
This program expects the following to be true about its runtime environment:
- It is running as root (unless you don't need
--become
) virtualenv
is installed on the server
This project uses Go Modules. Go 1.19+ should be able to handle this transparently.
bazelisk run //:ansible_puller
or, without bazel
go run .
bazelisk test //...
bazelisk build --config=release --platforms=@io_bazel_rules_go//go/toolchain:linux_amd64 //...
bazelisk build --config=release //:ansible_puller_deb
bazelisk build --config=release //:ansible_puller_rpm
For debugging the application, use the --debug
flag, or the debug
option in the config file.
This streams the Ansible output to the console so that you can follow along in the run.
Also consider using the --once
flag to run the process just once and then exit without spinning up the webserver.