Optica

Optica is a service for registering and locating nodes. It provides a simple REST API.

Nodes can POST to / to register themselves with some parameters. Humans can GET / to get a list of all registered nodes. GET also accepts some parameters to limit which of the registered nodes you see.

Why Optica?

We love the node registration features of chef server. However, we run chef-solo here at Airbnb. We use optica as alternate node registration system.

Installation

Use bundler! To install all the dependencies:

$ bundle install

Dependencies

Zookeeper

Optica is a front-end to a data store. At Airbnb, this data store is Apache Zookeeper.

Why Zookeeper?

we consider optica information critical data, with high uptime requirements
we already rely critically on Zookeeper to connect our infrastructure; we strive to ensure maximum uptime for this system
the load patterns of optica (many reads, infrequenty writes) match what zookeeper provides

Rabbitmq

Some parts of our infrastructure are asynchronous; we rely on notification of converges to know, for example, when some kinds of deploys have completed (or failed). For this reason, Optica generates events in rabbitmq for every converge.

Usage with Chef

We've included a sample notifier which reports back to optica on every chef converge. It's in this repo in reporter.rb, just make sure to substitute the correct value for the optica_server option. To use it, we added the chef-handler cookbook. Then, we do the following (in our common cookbook, which is applied to every role):

directory node.common.notifier_dir

cookbook_file `reporter.rb` do
  path File.join(node.common.notifier_dir, 'reporter.rb')
end

chef_handler 'notifier' do
  action    :enable
  source    File.join(node.common.notifier_dir, 'reporter.rb')
end

If you wish to register additional key-value pairs with your node, simply add them to node.optica.report:

default.optica.report['jvm_version'] = node.java.version

Usage on the command line

Optica has a very minimal query syntax, and errs on the side of returning more information than you need. Really, the only reason for the query parameters is to limit the amount of data transferred over the network. We can get away with it because all of the complex functionality you might wish for on the command line is provided by JQ.

JQ examples

Let's define a basic optica script:

#!/bin/bash

my_optica_host='https://optica.example.com'
curl --silent ${my_optica_host}/?"$1" | jq --compact-output ".nodes[] | $2"

With this in your $PATH and the right subsitution for your optica endpoint, here are some examples:

Getting all hostnames by role:

I run this, then pick a random one to ssh into when, e.g., investigating issues.

$ optica role=myrole .hostname

How many of each role in us-east-1a or 1b?

See what the impact will be of an outage in those two zones:

$ optica az=us-east 'select(.az == "us-east-1a" or .az == "us-east-1b") | .role' | sort | uniq -c | sort -n

Monitor the progress of a chef run on a role

Useful if you've just initiated a chef run across a large number of machines, or are waiting for scheduled runs to complete to deploy your change:

$ optica role=myrole '[.last_start, .failed, .hostname]' | sort

Usage with Fabric

We've included a sample fabfile.py to get you started. Simply replace optica.example with the address to your optica install.

Cleanup

Optica relies on you manually cleaning up expired nodes. At Airbnb, all of our nodes run in Amazon's EC2. We have a regularly scheduled task which grabs all recently terminated instances and performs cleanup, including optica cleanup, on those instances.

Cleanup is accomplished by calling DELETE on optica. For instance:

$ curl -X DELETE http://optica.example.com/i-36428351

Development

You'll need a copy of zookeeper running locally, and it should have the correct path for optica:

$ zkServer start
$ zkCli
[zk: localhost:2181(CONNECTED) 0] create /optica ''
Created /optica
[zk: localhost:2181(CONNECTED) 1] quit
Quitting...

The example config is set up to talk to your local zookeeper:

$ cd optica
$ cp config.json.example config.json

Edit the default config and add your EC2 credentials.

We run optica via unicorn. To spin up a test process on port 4567:

$ unicorn -p 4567

0xflotus/optica