install_base fails if a very old Docker version is already present

Summary

After running the install_base action on hosts with Docker 1.11.2, the docker system service fails and cannot be started any more.

Steps to reproduce

Run install_base on a host with Docker 1.11.2.

Expected behavior

Either nothing happens, or Docker gets reconfigured to match the requested settings.

Actual behavior

The docker service fails miserably and cannot be started.

Relevant logs and/or screenshots

The problem lies in the configuration file generated by Ansible, i.e. /etc/systemd/system/docker.service.d/settings.conf:

[Service]
ExecStart=
ExecStart=/usr/bin/dockerd --mtu=1450

Very old Docker installations do not have the dockerd binary and instead use /usr/bin/docker daemon.

Possible fixes

Dunno. Maybe just automatically update Docker to current version?

Also, the fix should work with other Docker configuration files already present in the system, such as this one:

[Service]
MountFlags=shared

EnvironmentFile=/etc/sysconfig/docker
# It's necessary to clear ExecStart before attempting to override it
# or systemd will complain that it is defined more than once.
ExecStart=
ExecStart=/usr/bin/docker daemon -H fd:// $other_args