Auto DevOps build failing due to breaking change in `docker:stable-dind`

Summary

Related to gitlab-runner#4501 (closed) Auto DevOps uses docker:stable-dind image in lots of places. This results in these jobs failing with:

time="2019-07-23T23:21:07Z" level=error msg="failed to dial gRPC: cannot connect to the Docker daemon. Is 'docker daemon' running on this host?: dial tcp [::1]:2375: connect: connection refused"
context canceled

Steps to reproduce

This is fixed on shared runners so can only be reproduced using another user installed runner for example the one installed under GitLab Managed applications for the configured Kubernetes cluster.

1. Create a new project
2. Disable shared runners
3. Add a cluster
4. Install Tiller and Runner on the cluster
5. Push an Auto DevOps compatible project
6. Enable Auto DevOps
7. build job should fail with above error

Example Project

https://staging.gitlab.com/hello-staging/minimal-ruby-app/-/jobs/36073654

What is the current bug behavior?

What is the expected correct behavior?

Relevant logs and/or screenshots

Possible fixes

See workaround from gitlab-runner#4501 (closed) but in short it can be worked around by simply setting a CI variable (Settings > CI/CD > Variables) called DOCKER_TLS_CERTDIR with an empty value.