This formula installs and configures Docker.
Install docker, configure daemon.json according to the information in Salt pillar.
See pillar.example for configuration options.
Setup a systemd timer that cleans up unused and/or dangling images and/or unused volumes automatically every night, using docker system prune.
See pillar.example to configure clean up behavior.
Install docker-compose package.
Setup dnsmasq to listen as a DNS server on the docker gateway 172.17.0.1. This is useful when dns servers configured on the host are ipv6 only
and docker is not configired with public ipv6 addresses because of lack of NAT.
Note: If dns servers specified on the host or via dns flags are unreachable from the containers then docker uses Google's public DNS server 8.8.8.8 for the containers.
Substate to install docker-machine.
Since docker-machine is deprecated upstream, this state relies on the fork maintained by Gitlab.
Due to this non-standard release channel, the installed version is controlled through a pillar key.
See releases for available version strings for the key docker-machine:version.
Updating the docker-machine binary requires to manually check for a new release, change the pillar key and reapply the state.
For security reasons, it is advisable to deploy a separate mechanism that at least automates the checking part, to avoid unknowingly running on a stale version.
When docker-machine is executed for the first time, a set of keys and certificates are generated.
The contents of these files must be manually generated and provided through pillar data.
To generate these files, simply run docker-machine create --driver <DRIVER> <OPTIONS> init-vm on a pristine machine, i.e. one where docker-machine is installed but was never invoked.
This call should use the same <DRIVER> and <OPTIONS> that will later be used for regular operation.
The concrete driver and corresponding options are highly use-case dependent.
Common pitfall: The controller node must be able to talk to its spawned worker on TCP port 2376, and the worker node must be capable to install OS package updates and the docker engine from the internet.
Ensure that <OPTIONS> sets suitable ingress and egress firewall rules.
Once init-vm has been fully provisioned it can be removed again with docker-machine rm init-vm.
The generated files are located in /root/.docker/machine/certs/.
See pillar.example on how to provision the contents of those key files.
Note that docker-machine worker engines inherit their registry authentication (more precisely, the full contents of /root/.docker/config.json) from the node on which docker-machine is executed.
Therefore this node should be authenticated against all relevant registries, which the workers will need to access.
See state docker.login below, which is provided to achieve this.
This state is optional.
Authenticate the system user root against a set of private docker registries.
The registries and the respectively required credentials should be provided through pillar data.
This state primarily makes use of the dockermod salt module documented here.
See pillar.example for further details.
This state configures Docker and Swarm. It defaults to use the interface eth0 for internal communication.
The interface can be adjusted using the Pillar:
docker:
swarm:
interface: ens3# Bootstrap the cluster
sudo salt-run state.orch docker.orch.bootstrap-swarm pillar='{"leader": "$leader", "nodes": ["$2nd_node", "3rd_node"]}'