Access secondary images from remote docker

Build Environment
1

I’m trying to figure out if it is possible to access secondary images/containers from a remote docker execution.

The background is that I am using docker run to execute tests in my docker artifact, as a final step to verify that the image was built correctly (compared to just running the tests in the circleci image). Now, my tests depend on a local DynamoDB container which works fine with the dual image setup of a regular build configuration:

test:
    environment:
      DYNAMODB_ENDPOINT: http://localhost:8000
    docker:
      - image: circleci/node:12
      - image: circleci/dynamodb
    steps:
      - checkout
      - run: npm ci
      - run: npm test

The secondary image starts a container and is accessible as localhost, so far so good.

Now, what I do during deployment is that I build a docker image using setup_remote_docker, and before pushing it I want to run the same tests again:

build_deploy:
    environment:
      DYNAMODB_ENDPOINT: http://host.docker.internal:8000
    docker:
      - image: circleci/python
      - image: circleci/dynamodb
    steps:
      - aws-cli/install
      - checkout
      - setup_remote_docker
      - run: docker run -e DYNAMODB_ENDPOINT -it --rm the_built_tag /bin/sh -c "npm install && npm test"

The problem here is I can’t find a way to actually access the host of the initial runner (the circleci/python container), or actually the secondary container (the circleci/dynamodb based one). I’ve tried with host.docker.internal which doesn’t even seem to resolve, localhost is completely wrong and --network="host" makes no difference here.

Is this even resolvable? My second option is to actually spin up the dynamodb with docker run as well and link the containers, but I thought that using the multi-image support in circleci would be much cleaner.

A lot of people seem to have problem with the other way around (accessing the container that was spun up remotely from the primary container), but I’m trying to do the opposite, sort of.

2

That’s interesting, I would have thought that would work inside a container. If you are stuck, you could always drop circleci/dynamodb as a secondary CircleCI container, and start it with your manually started image. I recommend Docker Compose here, as it can start your existing image, start Dynamo, and put them on the same virtual Docker network.