README.md

Ansible tutorial

Variables again

So we've setup our loadbalancer, and it works quite well. We grabbed variables from facts and used them to build the configuration. But Ansible also supports other kinds of variables. We already saw ansible_host in inventory, but now we'll use variables defined in host_vars and group_vars files.

Fine tuning our HAProxy configuration

HAProxy usually checks if the backends are alive. When a backend seems dead, it is removed from the backend pool and HAproxy doesn't send requests to it anymore.

Backends can also have different weights (between 0 and 256). The higher the weight, the higher number of connections the backend will receive compared to other backends. It's useful to spread traffic more appropriately if nodes are not equally powerful.

We'll use variables to configure all these parameters.

Group vars

The check interval will be set in a group_vars file for haproxy. This will ensure all haproxies will inherit from it.

We just need to create the file group_vars/haproxy below the inventory directory. The file has to be named after the group you want to define the variables for. If we wanted to define variables for the web group, the file would be named group_vars/web.

haproxy_check_interval: 3000
haproxy_stats_socket: /tmp/sock

The name is arbitrary. Meaningful names are recommended of course, but there is no required syntax. You could even use complex variables (a.k.a. Python dict) like this:

haproxy:
    check_interval: 3000
    stats_socket: /tmp/sock

This is just a matter of taste. Complex vars can help group stuff logically. They can also, under some circumstances, merge subsequently defined keys (note however that this is not the default ansible behaviour). For now we'll just use simple variables.

Hosts vars

Hosts vars follow exactly the same rules, but live in files under host_vars directory.

Let's define weights for our backends in host_vars/host1.example.com:

haproxy_backend_weight: 100

and host_vars/host2.example.com:

haproxy_backend_weight: 150

If we'd define haproxy_backend_weight in group_vars/web, it would be used as a 'default': variables defined in host_vars files overrides varibles defined in group_vars.

Updating the template

The template must be updated to use these variables.

global
    daemon
    maxconn 256
{% if haproxy_stats_socket %}
    stats socket {{ haproxy_stats_socket }}
{% endif %}

defaults
    mode http
    timeout connect 5000ms
    timeout client 50000ms
    timeout server 50000ms

listen cluster
    bind {{ ansible_eth1['ipv4']['address'] }}:80
    mode http
    stats enable
    balance roundrobin
{% for backend in groups['web'] %}
    server {{ hostvars[backend]['ansible_hostname'] }} {{ hostvars[backend]['ansible_eth1']['ipv4']['address'] }} check inter {{ haproxy_check_interval }} weight {{ hostvars[backend]['haproxy_backend_weight'] }} port 80
{% endfor %}
    option httpchk HEAD /index.php HTTP/1.0

Note that we also introduced an {% if ... block. This block enclosed will only be rendered if the test is true. So if we define haproxy_stats_socket somewhere for our loadbalancer (we might even use the --extra-vars="haproxy_stats_sockets=/tmp/sock" at the command line), the enclosed line will appear in the generated configuration file (note that the suggested setup is highly insecure!).

Let's go:

ansible-playbook -i step-11/hosts step-11/haproxy.yml

Note that, while we could, it's not necessary to run the apache playbook since nothing changed, but we had to cheat a bit for that. Here is the updated haproxy playbook :

- hosts: web
- hosts: haproxy
  tasks:
    - name: Installs haproxy load balancer
      apt: pkg=haproxy state=installed update_cache=yes

    - name: Pushes configuration
      template: src=templates/haproxy.cfg.j2 dest=/etc/haproxy/haproxy.cfg mode=0640 owner=root group=root
      notify:
        - restart haproxy

        - name: Sets default starting flag to 1
          lineinfile: dest=/etc/default/haproxy regexp="^ENABLED" line="ENABLED=1"
          notify:
            - restart haproxy 

      handlers:
        - name: restart haproxy
          service: name=haproxy state=restarted

See? We added an empty play for web hosts at the top. It does nothing. But it's here because it will trigger facts gathering on hosts in group web. This is required because the haproxy playbook needs to pick facts from hosts in this group. If we don't do this, ansible will complain saying that ansible_eth1 key doesn't exist.

Now on to the next chapter about "Migrating to Roles!", in step-12.