Aurora supports several containerizers, notably the Mesos containerizer and the Docker containerizer. The Mesos containerizer uses native OS features directly to provide isolation between containers, while the Docker containerizer delegates container management to the Docker engine.
The support for launching container images via both containerizers has to be enabled by a cluster operator.
The Mesos containerizer is the native Mesos containerizer solution. It allows tasks to be run with an array of pluggable isolators and can launch tasks using Docker images, AppC images, or directly on the agent host filesystem.
The following example (available in our Vagrant environment)
launches a hello world example within a debian/jessie Docker image:
$ cat /vagrant/examples/jobs/hello_docker_image.aurora
hello_loop = Process(
name = 'hello',
cmdline = """
while true; do
echo hello world
sleep 10
done
""")
task = Task(
processes = [hello_loop],
resources = Resources(cpu=1, ram=1*MB, disk=8*MB)
)
jobs = [
Service(
cluster = 'devcluster',
environment = 'devel',
role = 'www-data',
name = 'hello_docker_image',
task = task,
container = Mesos(image=DockerImage(name='debian', tag='jessie'))
)
]
Docker and Appc images are designated using an appropriate image property of the Mesos
configuration object. If either container or image is left unspecified, the host filesystem
will be used. Further details of how to specify images can be found in the
Reference Documentation.
By default, Aurora launches processes as the Linux user named like the used role (e.g. www-data
in the example above). This user has to exist on the host filesystem. If it does not exist within
the container image, it will be created automatically. Otherwise, this user and its primary group
has to exist in the image with matching uid/gid.
For more information on the Mesos containerizer filesystem, namespace, and isolator features, visit Mesos Containerizer and Mesos Container Images.
The Docker containerizer launches container images using the Docker engine. It may often provide more advanced features than the native Mesos containerizer, but has to be installed separately to Mesos on each agent host.
Starting with the 0.17.0 release, image can be specified with a {{docker.image[name][tag]}} binder so that
the tag can be resolved to a concrete image digest. This ensures that the job always uses the same image
across restarts, even if the version identified by the tag has been updated, guaranteeing that only job
updates can mutate configuration.
Example (available in the Vagrant environment):
$ cat /vagrant/examples/jobs/hello_docker_engine.aurora
hello_loop = Process(
name = 'hello',
cmdline = """
while true; do
echo hello world
sleep 10
done
""")
task = Task(
processes = [hello_loop],
resources = Resources(cpu=1, ram=1*MB, disk=8*MB)
)
jobs = [
Service(
cluster = 'devcluster',
environment = 'devel',
role = 'www-data',
name = 'hello_docker',
task = task,
container = Docker(image = 'python:2.7')
), Service(
cluster = 'devcluster',
environment = 'devel',
role = 'www-data',
name = 'hello_docker_engine_binding',
task = task,
container = Docker(image = '{{docker.image[library/python][2.7]}}')
)
]
Note, this feature requires a v2 Docker registry. If using a private Docker registry its url
must be specified in the clusters.json configuration file under the key docker_registry.
If not specified docker_registry defaults to https://registry-1.docker.io (Docker Hub).
Example: # clusters.json [{ “name”: “devcluster”, … “docker_registry”: “https://registry.example.com” }]
Details of how to use Docker via the Docker engine can be found in the Reference Documentation. Please note that in order to correctly execute processes inside a job, the Docker container must have Python 2.7 and potentitally further Mesos dependencies installed. This limitation does not hold for Docker containers used via the Mesos containerizer.
For more information on launching Docker containers through the Docker containerizer, visit Docker Containerizer