Ansible has the capability to do operations on multi-tier applications in a coordinated way, making it easy to orchestrate a sophisticated zero-downtime rolling upgrade of our web application. This is implemented in a separate playbook, called ``rolling_update.yml``.
What's going on here, and why are there no tasks? You might know that Ansible gathers "facts" from the servers before operating upon them. These facts are useful for all sorts of things: networking information, OS/distribution versions, etc. In our case, we need to know something about all of the monitoring servers in our environment before we perform the update, so this simple play forces a fact-gathering step on our monitoring servers. You will see this pattern sometimes, and it's a useful trick to know.
This is just a normal play definition, operating on the ``webservers`` group. The ``serial`` keyword tells Ansible how many servers to operate on at once. If it's not specified, Ansible will parallelize these operations up to the default "forks" limit specified in the configuration file. But for a zero-downtime rolling upgrade, you may not want to operate on that many hosts at once. If you had just a handful of webservers, you may want to set ``serial`` to 1, for one host at a time. If you have 100, maybe you could set ``serial`` to 10, for ten at a time.
- The ``serial`` keyword forces the play to be executed in 'batches'. Each batch counts as a full play with a subselection of hosts.
This has some consequences on play behavior. For example, if all hosts in a batch fails, the play fails, which in turn fails the entire run. You should consider this when combining with ``max_fail_percentage``.
The ``pre_tasks`` keyword just lets you list tasks to run before the roles are called. This will make more sense in a minute. If you look at the names of these tasks, you can see that we are disabling Nagios alerts and then removing the webserver that we are currently updating from the HAProxy load balancing pool.
The ``delegate_to`` and ``loop`` arguments, used together, cause Ansible to loop over each monitoring server and load balancer, and perform that operation (delegate that operation) on the monitoring or load balancing server, "on behalf" of the webserver. In programming terms, the outer loop is the list of web servers, and the inner loop is the list of monitoring servers.
Note that the HAProxy step looks a little complicated. We're using HAProxy in this example because it's freely available, though if you have (for instance) an F5 or Netscaler in your infrastructure (or maybe you have an AWS Elastic IP setup?), you can use modules included in core Ansible to communicate with them instead. You might also wish to use other monitoring modules instead of nagios, but this just shows the main goal of the 'pre tasks' section -- take the server out of monitoring, and take it out of rotation.
The next step simply re-applies the proper roles to the web servers. This will cause any configuration management declarations in ``web`` and ``base-apache`` roles to be applied to the web servers, including an update of the web application code itself. We don't have to do it this way--we could instead just purely update the web application, but this is a good example of how roles can be used to reuse tasks:
In this example, we use the simple HAProxy load balancer to front-end the web servers. It's easy to configure and easy to manage. As we have mentioned, Ansible has built-in support for a variety of other load balancers like Citrix NetScaler, F5 BigIP, Amazon Elastic Load Balancers, and more. See the :ref:`working_with_modules` documentation for more information.
For other load balancers, you may need to send shell commands to them (like we do for HAProxy above), or call an API, if your load balancer exposes one. For the load balancers for which Ansible has modules, you may want to run them as a ``local_action`` if they contact an API. You can read more about local actions in the :ref:`playbooks_delegation` section. Should you develop anything interesting for some hardware where there is not a core module, it might make for a good module for core inclusion!
Now that you have an automated way to deploy updates to your application, how do you tie it all together? A lot of organizations use a continuous integration tool like `Jenkins <https://jenkins.io/>`_ or `Atlassian Bamboo <https://www.atlassian.com/software/bamboo>`_ to tie the development, test, release, and deploy steps together. You may also want to use a tool like `Gerrit <https://www.gerritcodereview.com/>`_ to add a code review step to commits to either the application code itself, or to your Ansible playbooks, or both.
Depending on your environment, you might be deploying continuously to a test environment, running an integration test battery against that environment, and then deploying automatically into production. Or you could keep it simple and just use the rolling-update for on-demand deployment into test or production specifically. This is all up to you.
For integration with Continuous Integration systems, you can easily trigger playbook runs using the ``ansible-playbook`` command line tool, or, if you're using :ref:`ansible_tower`, the ``tower-cli`` or the built-in REST API. (The tower-cli command 'joblaunch' will spawn a remote job over the REST API and is pretty slick).
This should give you a good idea of how to structure a multi-tier application with Ansible, and orchestrate operations upon that app, with the eventual goal of continuous delivery to your customers. You could extend the idea of the rolling upgrade to lots of different parts of the app; maybe add front-end web servers along with application servers, for instance, or replace the SQL database with something like MongoDB or Riak. Ansible gives you the capability to easily manage complicated environments and automate common operations.
