alice

Alice is an autoscaler used at Notonthehighstreet. It is designed to scale up and down server and application instances based on daily demand, and is designed to be flexible and easy to extend.

Alice is written in Go and runs as a standalone background process in our production environment. Using metrics from a monitoring provider, it can make decisions on how to scale an inventory of resources. An inventory can be any set of resources that provide the capacity to deal with current demand - for example, the servers in a server farm, or the number of docker containers running on an instance of your application.

To support different monitoring providers and resources types, we have to write a monitor or inventory plugin to talk to the different backend APIs.

The currently supported backends are:

• Monitors: Datadog, Stats directly from Mesos
• Inventories: Marathon applications, AWS EC2 instances (via autoscaling groups)

It's relatively easy to write plugins for additional backends as required.

In addition to this, Alice is also easy to integrate with Slack and Fluentd.

Build

We use glide for vendoring dependencies, so once you have cloned this repository you will want to install it. This is usually fairly straightfoward:

OSX: brew install glide

Ubuntu: sudo add-apt-repository ppa:masterminds/glide && sudo apt-get update &&sudo apt-get install glide

Once this is installed you can install all the dependencies (into ./vendor by default):

glide install

To build Alice, just go build.

Configuration

When Alice starts up it reads ./config/config.yaml which should be relative to the executable's working directory. Copying and editing the config/config.yaml.dist file is a good place to start.

The main body of the configuration is under the managers section of the config file. Alice can manage multiple resource inventories at a time. A manager is a grouping of an inventory to be managed, a monitor from which to collect metrics, and a strategy which determines what scaling should be done on the inventory taking the current metrics into account.

A simple manager configuration might look like this:

managers:
  # Unique name of this manager
  my_web_application:
    monitor:
      # A datadog plugin example
      name: datadog
      api_key: xxxxxx
      app_key: xxxxxx
      time_period: 5m  # Most recent data point must be within this time
      metrics:
        active_users:
          query: avg:application.users.active{*}

    inventory:
      # A marathon application plugin example
      name: marathon
      settle_down_period: 3m
      url: http://marathon.example.com:8080
      app: www_example_com  # Application ID in marathon
      minimum_instances: 1
      maximum_instances: 10

    strategy:
       # A ratio strategy plugin example
      name: ratio
      ratios:
        active_users:
          metric: 100  # Keep one application instance for every 100 active users
          inventory: 1

How to test the software

The tests for Alice can be run using go test like this: go test -race -cover $(go list ./... | grep -v /vendor/)

Alternatively you can run make test

Getting help

If you have questions, concerns, bug reports, etc, please file an issue in this repository's Issue Tracker.

Getting involved

If you'd like to get involved and make Alice better, we'd be happy to accept your help!

Please read CONTRIBUTING for more information on how to contribute to Alice.