Set DOCKER_HOST in the args to popen3() instead of using --env

[mpkut]

We use service-url in inventory.yaml to specify an alternate Docker server address. For example:

   config:
     transport: docker
     docker:
       host: containername
       service-url: unix:///var/run/docker2.sock

This transport configuration has worked in the past with versions of Bolt 2. When we recently upgraded to Bolt 3, we found that the same configuration fails with a connection error. The error incorrectly refers to the default Docker socket, /var/run/docker.sock, instead of the Docker server specified by the inventory.

# bolt command run -t containername whoami
Started on containername...
Failed on containername:
  The command failed with exit code 1
  Cannot connect to the Docker daemon at unix:///var/run/docker.sock. Is the docker daemon running?
Failed on 1 target: containername
Ran on 1 target in 0.3 sec

Expected Behavior

The command run invocation should succeed:

# bolt command run -t containername whoami
Started on containername...
Finished on containername:
  root
Successful on 1 target: containername
Ran on 1 target in 0.62 sec

Steps to Reproduce

This procedure assumes that you have the Docker CLI client installed and are using an alternate socket or remote Docker host to access a container.

• Confirm CLI access to the alternate Docker server

# DOCKER_HOST=unix:///var/run/docker2.sock docker exec containername whoami
root

• Create a Bolt project with inventory.yaml as follows:

version: 2
groups:
 - name: example
   targets:
     - containername
   config:
     transport: docker
     docker:
       host: containername
       service-url: unix:///var/run/docker2.sock

• Attempt bolt command run using the alternate Docker host:

# bolt command run -t containername whoami

• Success == output is root as expected

• Failure == Bolt issues the connection error referring to the default Docker socket

Environment

• Version: 3.7.1
• Platform: CentOS 7

Additional Context

The execute() method inside lib/bolt/transport/docker/connection.rb uses --env to pass the DOCKER_HOST variable to the docker exec command.

        args += %W[--env DOCKER_HOST=#{@docker_host}] if @docker_host

Although it's not clear in the Docker exec documentation, a direct test shows that --env sets variables inside the container. It does not change the process environment of the docker exec command itself. And that's where DOCKER_HOST is needed in this case.

The suggested patch sets DOCKER_HOST by passing env_hash to the optional env argument of popen3(). This seems consistent with the other methods in connection.rb.

With this change, the bolt command run invocation succeeds as expected with Bolt 3.7.1.

[CLAassistant]

All committers have signed the CLA.

[lucywyman]

Thanks for catching this! Definitely a bug from when we refactored this transport. I verified the bug, and that the fix correctly connects to the specified Docker host.

Are you ok with signing our CLA before we merge?

[mpkut]

Thank you for the review! I am currently working through our internal process to clear me to sign the CLA, and will update as soon as that is complete.

[lucywyman]

Great, let me know if there's anything we can do help!

[mpkut]

I have received approval and signed the CLA as an individual. Thanks for your patience!