Previously: (II) Architecture

In Maestro we typically use a Maestro master server and multiple Maestro agents. Each Maestro Agent is just a small service where the actual work happens, it processes the work sent by the master, via ActiveMQ, and executes the plugins with the data received.

The two main goals of the agent are load distribution and heterogeneous composition support. The more agents running, the more compositions that can be executed in parallel, and compositions can target specific agents based on its features, such as architecture, operating system,… which is a must for development environments. For simplicity each agent can only run one composition at a time, but you could have multiple agent processes running in a single server.

It uses Puppet Facter to gather the machine facts (operating system, memory size, cloud provider data,…) and sends all that information to the master, that can use it to filter what compositions run in the agent. For instance I may want to run a composition in a Windows agent, or in an agent that has some specific piece of software installed. Facter supports external facts so it is really easy to add new filtering capabilities, and not be just limited to what Facter provides out of the box. A small text file can be added to /etc/facter/facts.d/ and Facter would report it to the master server.

Agents are installed alongside with all the tools that may be needed, from Git, to clone repos, to Jenkins swarm to reuse the agents as Jenkins slaves, or mcollective agents to allow updating the agent itself automatically with Puppet when new manifests are deployed to the Puppet master. In our internal environment any commit to Puppet manifests or modules automatically trigger our rspec-puppet tests, the deployment of those manifests to the Puppet master, and a cascading Puppet update of all the machines in our staging environment using MCollective. All our Puppet modules are likewise built and tested on each commit and a new version published to the Puppet Forge automatically using rspec-puppet and Puppet Blacksmith.

Maestro also supports manually assigning agents to pools, and matching compositions with agent pools, so compositions can be limited to run in a predefined set of agents.

The agent process is written in Ruby and runs under JRuby in the JVM, thus supporting multiple operating systems and architectures, and the ability to write extensions in Java or Ruby easily. It connects to the master’s Composition Execution Engine through ActiveMQ using STOMP for messaging.

Plugins

Plugins are small pieces of code written in Java or Ruby that run in the agent to execute the actual work. We have made all plugins available in GitHub so they can be used as examples to create new plugins for custom tasks.

Plugins can be added to Maestro at runtime and automatically show up in the composition editor. The plugin manifest defines the plugin images, what tasks are defined, and what fields in each task. Based on the workload received, the agent downloads and executes the plugin, which just accesses the fields in the workload and do the actual work, whatever it might be, sending output back to LuCEE and populating the composition context.

For instance the Fog plugin can manage multiple clouds, such as EC2, where it can start and stop instances. The plugin receives the fields defined in the composition (credentials, image id,…), calls the EC2 API, streams the status to the Maestro output (successfully created, instance id,…) and puts some data (ids of the instances created, public ips,…) in the composition context for other tasks to use. All of that in less than 100 lines of code.

The context is important to avoid redefining field values and provide some meaningful defaults, so if you have a provision task and a deprovision task, the values in the the latter are inherited from the former.

Agent cloud manager

The agent cloud manager is a service that runs on Google Compute Engine and watches a number of Maestro installations to provide automatic agent scaling. Based on preconfigured parameters such as min/max number of agents for each agent pool, max waiting time,… and the current status of each agent pool queue, the service can start new machines from specific images, suspend them (destroy the instance but keep the disk), or completely destroy them.

We are also giving a try to Docker instead of using full vms and have created a couple interesting Docker images on CentOS for developers, a Jenkins swarm slave image and a build agent image that includes everything we use at development: Java, Ant, Maven, RVM (with 1.9, 2.0, 2.1, JRuby), Git, Svn, all configurable with credentials at runtime.

Previously: (I) Workflow

Maestro architecture is basically defined by a master server and multiple agents, written in Java and Ruby (JRuby) for the backend and JavaScript for the frontend using AngularJS, and integrating several open source services. It is quite heterogeneous, with multiple languages, build tools, packages,… using the best tool for the job in each part of the stack.

Master

The master services include

• Maestro REST API
• End user web interface
• Composition Execution Engine (LuCEE)
• ActiveMQ for STOMP messaging
• PostgreSQL (or MySQL)
• MongoDB

Maestro REST API

The REST API is a webapp written in Java, using Spring, packaged with a Jetty server. It is documented with Swagger annotations that generate a really nice web interface automatically that allows trying all the operations from the browser.

It handles caching, security, based on LDAP or database records, and delegates to the Composition Execution Engine (LuCEE) typically through LuCEE REST API but also via STOMP messaging to avoid continuous polling.

It also implements handlers to execute compositions from Github, Git, SVN,… on commit callbacks.

End user web interface

The end user UI is written in AngularJS using the AngularJS Bootstrap components and Less stylesheets. It connects to the REST API, so everything that can be done through the webapp can also be automated using the REST API (automation, automation, automation!). I have found Angular really nice to work with besides the service, factory, provider,… complicated abstractions, with good modularity and the ability to reuse third party plugins.

Built with Maven and Grunt (better for the Javascript parts), using Bower to manage all the Javascript dependencies (angular core, bootstrap, ladda button spinner,…), and Karma + PhantomJS, for headless UI tests without needing a real browser.

Composition Execution Engine (LuCEE)

LuCEE is a webapp that manages the execution of compositions, sending/receiving work to/from the agents through ActiveMQ STOMP queues, and storing state in the PostgreSQL database. LuCEE uses the Ruote workflow engine for work scheduling, and manages the compositions queue and agent routing, so basically checks what compositions need to be executed and decides in what agent to execute them, based on composition requirements, free agents, and other factors ie. prioritizing previously used agents that would likely have a cached copy of sources and dependencies to speed things up.

It is written in Ruby, it was quick to implement a first version, with a simple REST API using Sinatra and a STOMP connector to send messages to the Maestro REST webapp through ActiveMQ.

It is packaged as a JRuby war with Warbler, and both LuCEE and the REST API wars are run in the same Jetty server, all packaged as an RPM for easier deployment.

ActiveMQ

ActiveMQ handles all the comunication between LuCEE, the REST API webapp, and the agents using multiple STOMP queues. All the comunication between LuCEE and agents such as workloads, agent output, agent status,… is sent over a queue so it can be easily scaled across a high number of agents.

LuCEE also pushes changes in the database to the REST API webapp so it can update the caches without needing continuous polling.

PostgreSQL

LuCEE uses PostgreSQL (or MySQL or any other SQL database using Ruby Datamapper) as main storage to save compositions, projects, tasks,… The SQL database is also used by the REST API webapp to store permissions and user data when not using LDAP.

MongoDB

We found that in order to do more complex dashboards and reports we needed to store all sort of unstructured data from the plugins, from run time or status to anything that a plugin developer may want such as GitHub payload data received or test stacktrace. That data is sent by the agents to LuCEE and then stored in MongoDB, and can be queried directly (all your data belong to you) or through a reporting pane in the webapp.

Next: (III) Agents

At MaestroDev we have been building what may be called, for lack of a better name, a DevOps Orchestration Engine, and is long overdue to talk about what we have been doing there and most importantly, how.

The basics of the application is to tie together the different systems involved in a Continuous Delivery cycle: Continuous Integration server, SCM, build tools, packaging tools, cloud resources, notification systems,… and streamline the process through these different tools. So it hooks into a bunch of popular tools to orchestrate interactions between them, an example:

This workflow, or as we call it, composition, will

1. download a war file from a Maven repository (previously built by Jenkins)
2. start an Amazon EC2 instance with Tomcat preinstalled
3. deploy the war
4. checkout the acceptance tests from Git
5. run some tests with Maven (Selenium tests using SauceLabs) against that instance
6. wait for an user to confirm before moving to the next step (to record the human approval or to do some extra manual tests if needed)
7. destroy the Amazon EC2 instance

Maestro provides a nice web UI that gives visibility over the composition execution and an aggregated log from all the tools that run during the composition in a single place.

But the power comes with the combination of compositions together, as there are tasks for typical flows, such as running forking and joining compositions, call another composition in case of a failure, or waiting for a composition to finish.

Here we have a more complex setup with five compositions tied together.

• * – A composition that calls compositions 1 and 2.
• 1 – A Jenkins build
• 2 – The acceptance tests composition mentioned before
• 2a – Notification composition in case the acceptance tests fail
• 3 – Deployment to production

So you can see that compositions are not just limited to build, test, deploy. The tasks can be combined as needed to build your specific process.

Tasks are contributed by plugins, easily written in Ruby or Java, and define what fields are needed in the UI and what to do with those fields and the composition context. Maestro includes a lot of prebuilt tasks, publicly available on GitHub, from executing shell scripts to Jenkins job creation or Amazon Route 53 record management, but anything.

All the tasks share a common context and use sensible defaults, so if the scm checkout path is not defined it creates a specific working directory for the composition, and that is reused by the Maven, Ant,… plugins to avoid copying and pasting the fields. That’s also how a EC2 deprovision task doesn’t need any configuration if there was a provision task before in the composition, it will just deprovision those instances started previously in the composition by default.

You can take a look at our Maestro public instance, showing some examples and builds of public projects, mostly Puppet modules that are automatically built and deployed to the Puppet Forge, and Maestro plugins build and release compositions. In next posts I’ll be talking about the technologies used and distributed architecture of Maestro.

Next: (II) Architecture