Skip to content

Latest commit

 

History

History
178 lines (132 loc) · 4.38 KB

README.md

File metadata and controls

178 lines (132 loc) · 4.38 KB

Run Nomad from a Docker container

Docker Pulls

This repository builds a Docker image to run the Nomad scheduler.

The image is mostly useful for testing purpose, when you want to ship a small stack running Nomad along other containers.

Get the image using:

docker pull ghcr.io/multani/nomad

or:

docker pull multani/nomad

See below for examples on how to start a Nomad client or server.

This image is meant to be run with host network privileges. Nomad itself can be configured:

  • either by bind-mounting HCL/JSON configuration files into /etc/nomad

  • and/or by setting the configuration content directly into the NOMAD_LOCAL_CONFIG environment variable (see examples below).

You also need to bind-mount the following directories (unless you really now what you are doing):

  • /var/run/docker.sock: to access the Docker socket, used by Nomad Docker's driver
  • /tmp: default temporary directory used by Nomad's -dev mode

You can run the container as a non-root user, in which case you should set the NOMAD_DISABLE_PERM_MGMT environment variable to any value. This is especially useful when running standalone Nomad servers.

The repository produces a dockerized version of Nomad following Hashicorp's model for their Dockerized Consul image. It is based on the work from djenriquez/nomad.

To run:

You can use the Docker Compose file to get started:

docker compose up

The relevant Docker Compose bits are:

version: '2.1'

services:
  nomad:
    image: multani/nomad
    build: .
    command: agent -dev
    privileged: true
    network_mode: host
    environment:
      NOMAD_LOCAL_CONFIG: |
        data_dir = "/nomad/data/"

    volumes:
      - /var/run/docker.sock:/var/run/docker.sock:rw
      - /tmp:/tmp

Or you can configured Nomad on dedicated host with the following command lines.

Server

Notes for the server:

  • It doesn't need to run as root.
  • If you decide to run it as non-root, the Nomad data directory must have the proper permissions.
docker run -d \
  --name nomad \
  --net host \
  --user nomad \
  -e NOMAD_DISABLE_PERM_MGMT=true \
  -e NOMAD_LOCAL_CONFIG='
server {
  enabled = true
  bootstrap_expect = 3
}

datacenter = "${REGION}"
region     = "${DATACENTER}"

data_dir = "/nomad/data/"

bind_addr = "0.0.0.0"

advertise {
  http = "{{ GetPrivateIP }}:4646"
  rpc  = "{{ GetPrivateIP }}:4647"
  serf = "{{ GetPrivateIP }}:4648"
}
' \
  -v "nomad:/nomad/data:rw" \
  multani/nomad agent

Client

Notes for the client:

  • Most of the task drivers require quite high privileges, you should most probably run the container as root with the privileged Docker flag.
docker run -d \
  --name nomad \
  --net host \
  --privileged \
  -e NOMAD_LOCAL_CONFIG='
client {
  enabled = true
}

datacenter = "${REGION}"
region     = "${DATACENTER}"

data_dir = "/nomad/data/"

bind_addr = "0.0.0.0"

advertise {
  http = "{{ GetPrivateIP }}:4646"
  rpc  = "{{ GetPrivateIP }}:4647"
  serf = "{{ GetPrivateIP }}:4648"
}
' \
  -v "/srv/nomad/data:/nomad/data:rw" \
  -v "/var/run/docker.sock:/var/run/docker.sock" \
  -v "/tmp:/tmp" \
  multani/nomad agent

The above command is identical to running this example in Nomad's documentation for bootstrapping with Consul.

Correctly configuring Nomad data directory

Due to the way Nomad exposed template files it generates, you need to take special precautions when configuring its data directory.

In case you are running Docker containers and using the template stanza, the Nomad data_dir has to be configured with the exact same path as the host path, so the host Docker daemon mounts the correct paths, as exported by the Nomad client, into the scheduled Docker containers.

You can run the Nomad container with the following options in this case:

export NOMAD_DATA_DIR=/host/path/to/nomad/data

docker run \
  ...\
  -v "$NOMAD_DATA_DIR:$NOMAD_DATA_DIR:rw" \
  -e "NOMAD_DATA_DIR=$NOMAD_DATA_DIR" \
  multani/nomad agent