Docker Service

Docker Service Class

Classes:

Docker(*[, agent, docker_version, registry, ...])

Deploy docker agents on the nodes and registry cache(optional)

class enoslib.service.docker.docker.Docker(*, agent: List[Host] | None = None, docker_version: str | None = None, registry: List[Host] | None = None, registry_opts: Dict | None = None, bind_var_docker: str | None = None, swarm: bool = False, credentials: Dict | None = None, nvidia_toolkit: bool | None = None)

Deploy docker agents on the nodes and registry cache(optional)

This assumes a debian/ubuntu base environment and aims at producing a quick way to deploy docker and optionally a registry on your nodes.

If an NVidia GPU is detected on a node, the nvidia-container-toolkit will be also installed automatically. See https://docs.nvidia.com/datacenter/cloud-native/ Installation of the nvidia-container-toolkit can be forced with nvidia_toolkit=True, or prevented with nvidia_toolkit=False.

Examples

# Simply install latest docker agent without any registry
docker = Docker(agent=roles["agent"])

# Install a specific version of docker agent (recommended)
docker = Docker(agent=roles["agent"], docker_version="25.0")

# Don't install nvidia-container-toolkit even if a nvidia GPU
# is detected on the target hosts.
docker = Docker(agent=roles["agent"], nvidia_toolkit=False)

# Install and use an internal registry on the specified host
docker = Docker(agent=roles["agent"],
                registry=roles["registry"])

# Install and use an internal registry on the first agent
docker = Docker(agent=roles["agent"],
                registry_opts=dict(type="internal"))

# Use an external registry
docker = Docker(agent=roles["compute"] + roles["control"],
                registry_opts = {"type": "external",
                                 "ip": "192.168.42.1",
                                 "port": 4000})

import logging

import enoslib as en

en.init_logging(level=logging.INFO)
en.check()

conf = (
    en.VagrantConf()
    .add_machine(roles=["control"], flavour="tiny", number=1)
    .add_machine(roles=["compute"], flavour="tiny", number=1)
    .add_network(roles=["mynetwork"], cidr="192.168.42.0/24")
    .finalize()
)

# claim the resources
provider = en.Vagrant(conf)
roles, networks = provider.init()

# generate an inventory compatible with ansible
roles = en.sync_info(roles, networks)

docker = en.Docker(registry=roles["control"], agent=roles["compute"])
docker.deploy()
docker.backup()
docker.destroy()

# destroy the boxes
provider.destroy()

import logging
from pathlib import Path

import enoslib as en

en.init_logging(level=logging.INFO)
en.check()

job_name = Path(__file__).name

CLUSTER = "paravance"

conf = en.G5kConf.from_settings(
    job_type=[], job_name=job_name, walltime="0:30:00"
).add_machine(roles=["control"], cluster=CLUSTER, nodes=2)

provider = en.G5k(conf)
roles, networks = provider.init()


registry_opts = dict(type="external", ip="docker-cache.grid5000.fr", port=80)

d = en.Docker(
    agent=roles["control"],
    docker_version="25.0",
    bind_var_docker="/tmp/docker",
    registry_opts=registry_opts,
    # Optional credentials for docker hub
    # credentials=dict(login="mylogin", password="mytoken"),
)
d.deploy()


# Release all Grid'5000 resources
provider.destroy()

Parameters:

• agent (list) – list of enoslib.Host where the docker agent will be installed

• docker_version (str) – major version of Docker to install. Defaults to latest.

• registry (list) – list of enoslib.Host where the docker registry will be installed.

• registry_opts (dict) – registry options. The dictionary must comply with the schema.

• bind_var_docker (str) – If set the default docker state directory (/var/lib/docker/) will be bind mounted in this directory. The rationale is that on Grid’5000, there isn’t much disk space on /var/lib by default. Set it to False to disable the fallback to the default location.

• swarm (bool) – Whether a docker swarm needs to be created over the agents. The first agent will be taken as the swarm master.

• credentials (dict) – Optional ‘login’ and ‘password’ for Docker hub. Useful to access private images, or to bypass Docker hub rate-limiting: in that case, it is recommended to use a token with the “Public Repo Read-Only” permission as password, because it is stored in cleartext on the nodes.

• nvidia_toolkit (bool) – Whether to install nvidia-container-toolkit. If set to None (the default), Enoslib will try to auto-detect the presence of a nvidia GPU and only install nvidia-container-toolkit if it finds such a GPU. Set to True to force nvidia-container-toolkit installation in all cases, or set to False to prevent nvidia-container-toolkit installation in all cases.

backup()

(Not implemented) Backup docker.

Feel free to share your ideas.

deploy()

Deploy docker and optionally a docker registry cache.

destroy()

(Not implemented) Destroy docker

Feel free to share your ideas.