CI pipeline fails: Docker-in-Docker (docker:dind) fails to start properly

https://gitlab.com/postgres-ai/database-lab/-/jobs/10494520192

CI pipeline jobs intermittently fail when using Docker-in-Docker (docker:dind) with the following error:

dial tcp: lookup docker on : no such host
failed to load listeners: can’t create unix socket /var/run/docker.sock: device or resource busy

The issue appears during the docker build step, after attempting to authenticate with the Docker registry.

Root Cause

  • docker:dind is started without explicitly enabling TCP mode.

  • .gitlab-ci.yml is configured with:

    DOCKER_HOST: tcp://docker:2375

    which requires dind to expose TCP.

  • Meanwhile, config.toml for the GitLab Runner includes this volume:

     "/var/run/docker.sock:/var/run/docker.sock"

    which conflicts with Docker-in-Docker by trying to mount host’s socket inside the job container.

  • As a result, docker:dind fails to start its own dockerd correctly.