About a month or so ago, I tried to install the GitLab Runner for some personal project usage, and kept getting hit with similar errors to below:

cannot connect to the Docker daemon. Is 'docker daemon' running on this host?: dial tcp ...:2375: connect: connection refused
cannot connect to the docker daemon at tcp://docker:2375. is the docker daemon running?
error during connect: Post ... dial tcp: lookup docker on ...:53: no such host

I ended up leaving the issue for another day, and today managed to work out the issue.

The issue I was encountering is that the GitLab runner expected there to be a TCP socket available for the Docker process, whereas it appears that my installation (maybe due to how Docker is packaged for Debian?) by default only provided the file-based socket.

For additional context, I am running this on Debian Stretch (Debian 9), with GitLab Runner v12.9.0, and Docker v17.12.0-ce.

Providing Privileges to Interact with Docker

If it doesn't already have access, the user that gitlab-runner is executing as needs to be in the Docker group. For instance, assuming that the user gitlab is used to execute the runner:

$ sudo gpasswd -a gitlab docker

You may need to restart the gitlab-runner service first.

Creating the TCP Socket

Following these steps on StackOverflow, I was able to create a TCP socket using systemd:

[Unit]
Description=Docker Socket for the API
PartOf=docker.service

[Socket]
ListenStream=2375

BindIPv6Only=both
Service=docker.service

[Install]
WantedBy=sockets.target

This then exposes the TCP socket that to connect to Docker, and builds start working as soon as the new socket is available!