The information on this page is written for Vamp v0.9.3

Blueprints are execution plans - they describe how your services should be hooked up and what their topology should look like at runtime. This means you reference your breeds (or define them inline) and add runtime configuration to them.

Blueprints allow you to add the following extra properties:

  • Gateways: a stable port where the service can be reached.
  • Clusters and services: a cluster is a grouping of services with one purpose, i.e. two versions (a/b) of one service.
  • Environment variables: a list of variables (interpolated or not) to be made available at runtime.
  • Dialects: a dialect is a set of native commands for the underlying container platform, i.e. Docker or Mesosphere Marathon.
  • Scale: the CPU and memory and the amount of instance allocate to a service.
  • Conditions: how traffic should be directed based on HTTP and/or TCP properties.
  • SLA and escalations: SLA definition that controls autoscaling.

Example - key concepts of blueprints

name: my_blueprint                        # Custom blueprint name
  8080/http: my_frontend/port
  my_frontend:                            # Custom cluster name.
    gateways:                             # Gateway for this cluster services.
      routes:                             # Makes sense only with
        some_cool_breed:                  # multiple services per cluster.
          weight: 95%
          condition: User-Agent = Chrome
        some_other_breed:                 # Second service.
          weight: 5%
    services:                             # List of services
          ref: some_cool_breed
        scale:                            # Scale for this service.
          cpu: 2                          # Number of CPUs per instance.
          memory: 2048MB                  # Memory per instance (MB/GB units).
          instances: 2                    # Number of instances
          ref: some_other_breed           # Another service in the same cluster.  
        scale: large                      # Notice we used a reference to a "scale". 
                                          # More on this later.


A gateway is a “stable” endpoint (or port in simplified sense) that almost never changes. When creating the mapping, it uses the definition (my_frontend/port in this case) from the “first” service in the cluster definition you reference. This service can of course be changed, but the gateway port normally doesn’t.

Please take care of setting the /tcp or /http (default) type for the port. Using /http allows Vamp to record more relevant metrics like response times and metrics.

Read more about gateways.


gateways are optional. You can just deploy services and have a home grown method to connect them to some stable, exposable endpoint.

Clusters and services

In essence, blueprints define a collection of clusters. A cluster is a group of different services, which will appear as a single service and serve a single purpose.

Common use cases would be service A and B in an A/B testing scenario - usually just different versions of the same service (e.g. canary release or blue/green deployment).

Clusters are configured by defining an array of services. A cluster can be given an arbitrary name. Services are just lists or arrays of breeds.

   - breed: 
      ref: my_cool_service_A      # reference to an existing breed
     breed:                       # shortened inline breed
       name: my_cool_service_B
       deployable: some_container

Clusters and services are just organisational items. Vamp uses them to order, reference and control the actual containers and gateways and traffic.

This all seems redundant, right? We have a reference chain of blueprints -> gateways -> clusters -> services -> breeds -> deployable. However, you need this level of control and granularity in any serious environment where DRY principles are taken seriously and where “one size fits all” doesn’t fly.


Vamp allows you to use container driver specific tags inside blueprints. We call this a “dialect”. Dialects effectively enable you to make full use of, for instance, the underlying features like mounting disks, settings commands and providing access to private Docker registries.

We currently support the following dialects:

  • docker:
  • marathon:

Docker dialect

The following example show how you can mount a volume to a Docker container using the Docker dialect.

Example blueprint - using the Docker dialect

name: busybox
        name: busybox-breed
        deployable: busybox:latest
          "/tmp": ~

Vamp will translate this into the proper API call. Inspecting the container after it’s deployed should show something similar to this:

"Volumes": {
      "/tmp": "/mnt/sda1/var/lib/docker/volumes/1a3923fa6108cc3e19a7fe0eeaa2a6c0454688ca6165d1919bf647f5f370d4d5/_data"

Marathon dialect

This is an example with Marathon that pulls an image from private repo, mounts some volumes, sets some labels and gets run with an ad hoc command: all taken care of by Marathon.

We can provide the marathon: tag either on the service level, or the cluster level. Any marathon: tag set on the service level will override the cluster level as it is more specific. However, in 9 out of 10 cases the cluster level makes the most sense. Later, you can also mix dialects so you can prep your blueprint for multiple environments and run times within one description.

example blueprint - using the Marathon dialect

Notice the following:

  • Under the marathon: tag, we provide the command to run in the container by setting the cmd: tag.
  • We provide a url to some credentials file in the uri array. As described in the Marathon docs ( - using a private Docker repository) this enables Mesos to pull from a private registry, in this case where these credentials are set up.
  • We set some labels with some arbitrary metadata.
  • We mount the /tmp to in Read/Write mode.
name: busy-top:1.0
        name: busybox
       cmd: "top"
         environment: "staging"
         owner: "buffy the vamp slayer"
             containerPath: "/tmp/"
             hostPath: "/tmp/"
             mode: "RW"


Scale is the “size” of a deployed service. Usually that means the number of instances (servers) and allocated CPU and memory.

Scales can be defined inline in a blueprint or they can defined separately and given a unique name. The following example is a scale named “small”. POST-ing this scale to the /scales REST API endpoint will store it under that name so it can be referenced from other blueprints.

Example scale

name: small   # Custom name.

cpu: 2        # Number of CPUs per instance.
memory: 2gb   # Memory per instance, MB/GB units.
instances: 2  # Number of instances.

What next?

Last updated on March 14, 2017