Day 27 - From Automated to Automatic - Event-Driven Infrastructure Management with Ansible
Daniel Bodky
Overview
A universal truth and recurring theme in the DevOps world is automation. From providing infrastructure to testing code to deploying to production, many parts of the DevOps lifecycle get automated already. One popular technology for managing infrastructure and configuration in an automated way is Ansible, but are we fully utilizing its capabilities yet?
This presentation will give a broad overview of Ansible and its architecture and use-cases, before exploring a relatively new feature, Event-driven Ansible (EDA). Analzying applications of event-driven Ansible, participants will see that automated management is nice, but automatic management is awesome, not just regarding DevOps principles, but also in terms of reaction times, the human tendency for minor mistakes, and toil for operators.
Participants will get first-hand insights into Ansible, its strengths, weaknesses, and the potential of event-driven automation within the DevOps world.
NOTEThe below content is a copy of the lab repository’s README for convenience.
Event-Driven Ansible Lab
This is a lab designed to demonstrate Ansible and how Event-Driven Ansible (EDA) builds on top of its capabilities.
The setup is done with Ansible, too. It will install Ansible, EDA, Prometheus, and Alertmanager on a VM to demonstrate some of the capabilities of EDA.
Prerequisites
To follow along with this lab in its entirety, you will need three VMs:
NOTEIf you want to skip Ansible basics and go straight to EDA, you’ll need just the
eda-controller.example.com
VM and can skip the others.
VM name | OS |
---|---|
eda-controller.example.com | CentOS/Rocky 8.9 |
company.example.com | CentOS/Rocky 8.9 |
webshop.example.com | Ubuntu 22.04 |
You’ll need to be able to SSH to each of these VMs as root using SSH keys.
Lab Setup
Clone the repository and create a Python virtual environment
git clone https://github.com/mocdaniel/lab-event-driven-ansible.gitcd lab-event-driven-ansiblepython3 -m venv .venvsource .venv/bin/activate
Install Ansible and other dependencies
pip install -r requirements.txt
Create the inventory file
---webservers: hosts: webshop.example.com: ansible_host: <ip-address> webserver: apache2 company.example.com: ansible_host: <ip-address> webserver: httpdeda_controller: hosts: eda-controller.example.com: ansible_host: <ip-address>
Install Needed Roles and Collections
ansible-galaxy install -r requirements.yml
Run the Setup Playbook
After you created the inventory file and filled in the IP addresses, you can run the setup playbook:
ansible-playbook playbooks/setup.yml
Due to a known bug with Python on MacOS, you need to run export NO_PROXY="*"
on MacOS before running the playbook
Demos
Lab 1: Ansible Basics
Ansible from the CLI via ansible
Ansible from the CLI via ansible
The first example installs a webserver on all hosts in the webservers
group. The installed webserver is defined as a host variable in the inventory file hosts.yml
(see above).
ansible \ webservers \ -m package \ -a 'name="{{ webserver }}"' \ --one-line
Afterwards, we can start the webserver on all hosts in the webservers
group.
ansible \ webservers \ -m service \ -a 'name="{{ webserver }}" state=started' \ --one-line
Go on and check if the web servers are running on the respective hosts.
[!HINT] Ansible is idempotent - try running the commands again and see how the output differs.
Ansible from the CLI via ansible-playbook
Ansible from the CLI via ansible-playbook
The second example utilizes the following playbook to gather and display information for all hosts in the webservers
group, utilizing the example role from the lab repository.
---- name: Example role hosts: webservers gather_facts: false vars: greeting: "Hello World!" pre_tasks: - name: Say Hello ansible.builtin.debug: msg: "{{ greeting }}" roles: - role: example post_tasks: - name: Say goodbye ansible.builtin.debug: msg: Goodbye!
ansible-playbook \ playbooks/example.yml
Lab 2: Event-Driven Ansible
Receive Generic Events via Webhook
Receive Generic Events via Webhook
If you followed the setup instructions for the EDA lab, you should already have a running EDA instance on the eda-controller.example.com
VM.
If you navigate to /etc/edacontroller/rulebook.yml
on the VM, you’ll see the following rulebook:
---- name: Listen to webhook events hosts: all sources: - ansible.eda.webhook: host: 0.0.0.0 port: 5000 rules: - name: Debug event output condition: 1 == 1 action: debug: msg: "{{ event }}"
- name: Listen to Alertmanager alerts hosts: all sources: - ansible.eda.alertmanager: host: 0.0.0.0 port: 9000 data_alerts_path: alerts data_host_path: labels.instance data_path_separator: . rules: - name: Restart MySQL server condition: event.alert.labels.alertname == 'MySQL not running' and event.alert.status == 'firing' action: run_module: name: ansible.builtin.service module_args: name: mysql state: restarted - name: Debug event output condition: 1 == 1 action: debug: msg: "{{ event }}"
For this part of the lab, the first rule is the one we’re interested in: It listens to a generic webhook on port 5000
and prints the event’s metadata to its logs.
To test this, we can use the curl
command to send a POST
request to the webhook /endpoint
from the VM itself:
curl \ -X POST \ -H "Content-Type: application/json" \ -d '{"foo": "bar"}' \ http://localhost:5000/endpoint
If you now check the logs of the EDA controller, you should see the following output:
journalctl -fu eda-controller
Jan 11 16:35:29 eda-controller ansible-rulebook[56882]: {'payload': {'foo': 'bar'}, 'meta': {'endpoint': 'endpoint','headers': {'Host': 'localhost:5000', 'User-Agent': 'curl/7.76.1', 'Accept': '*/*', 'Content-Length': '21','Content-Type': 'application/x-www-form-urlencoded'}, 'source': {'name': 'ansible.eda.webhook', 'type': 'ansible.eda.webhook'},'received_at': '2024-01-11T15:35:29.798401Z', 'uuid': '6ebf8dd2-60a2-455a-9383-97b81f535366'}}
A rule that always evaluates to true
is not very useful, so let’s change the rule to only print the the value of foo
if the foo
key is present in the event’s payload, and no foo :(
otherwise:
---- name: Listen to webhook events hosts: all sources: - ansible.eda.webhook: host: 0.0.0.0 port: 5000 rules: - name: Foo condition: event.payload.foo is defined action: debug: msg: "{{ event.payload.foo }}" - name: No foo condition: 1 == 1 action: debug: msg: "no foo :("
Send the same curl
request again and check the logs, you should see a line saying bar
now.
Let’s also try a curl
request with a different payload:
curl \ -X POST \ -H "Content-Type: application/json" \ -d '{"bar": "baz"}' \ http://localhost:5000/endpoint
This time, the output should be no foo :(
.
Restarting Services Automatically with EDA
Restarting Services Automatically with EDA
The last lab is more of a demo - it shows how you can use EDA to automatically react on events observed by Prometheus and Alertmanager.
For this demo, the second ruleset in our rulebook is the one we’re interested in:
- name: Listen to Alertmanager alerts hosts: all sources: - ansible.eda.alertmanager: host: 0.0.0.0 port: 9000 data_alerts_path: alerts data_host_path: labels.instance data_path_separator: . rules: - name: Restart MySQL server condition: event.alert.labels.alertname == 'MySQL not running' and event.alert.status == 'firing' action: run_playbook: name: ./playbook.yml - name: Debug event output condition: 1 == 1 action: debug: msg: "{{ event }}"
With this rule, we can restart our MySQL server if it’s not running! But how do we get the event to trigger? With Prometheus and Alertmanager!
When you ran the setup playbook, it installed Prometheus and Alertmanager on the eda-controller.example.com
VM. You can access the Prometheus UI at http://<eda-controller-ip>:9090
and the Alertmanager UI at http://<eda-controller-ip>:9093
.
It also installed a Prometheus exporter for the MySQL database that runs on the server.
With this setup, we can now shut down our MySQL server and see what happens - make sure to watch the output of the EDA controller’s logs:
systemctl stop mysqljournalctl -fu edacontroller
Within 30-90 seconds, you should see EDA running our playbook and restarting the MySQL server. You can track that process by watching the Prometheus/Alertmanager UIs for firing alerts.
Once you see the playbook being executed in the logs, you can check the MySQL state once more:
systemctl status mysql
MySQL should be up and running again!