I am trying to run pants on bitbucket-pipelines. When I run "pants lint package ::", it manages to build all the binaries (pex), but fails to build the docker images, giving me the error`Cannot connect to the Docker daemon at unix:///var/run/docker.sock. Is the docker daemon running?` . In the same script I however can build a normal Dockerfile with

docker build -t hello_docker .

. Any ideas why this might be happening?

Stack Overflow is a cesspool of bad advice generally, but without my hands in a bitbucket pipeline, this seems plausible - you may want to confirm: https://stackoverflow.com/a/73487393

This is my test script. Note that the first docker build which runs outside pants works (just a hello world docker image); but fails when executed through pants.

image: python:3.10.10-slim-bullseye
pipelines:
  default:
    - step:
          services:
            - docker
          script:
            - docker build -t my_project .
            - export PATH=/root/bin:$PATH
            - ./get-pants.sh
            - docker --version
            - "pants lint package ::"

Sorry - I should have been clearer. That SO implies the env has DOCKER_HOST. Pants is hermetic and you'd need to punch that through.

I have not got it working yet, in the bitbucket logs I get the following error "time="2023-03-07T170339.328682185Z" level=warning msg="could not change group /var/run/docker.sock to docker: group docker not found""

Ok, you'll have to play. HOME is not punched through either, etc. I don't imagine that would affect chgrp , but.

I added this to pants.toml:

[docker]
env_vars = [
'DOCKER_HOST="<tcp://0.0.0.0:2375>"'
]

You might start by adding a CI step that just dumps

env

True that. I dumped all the environmental variables, and one of them is DOCKER_HOST=tcp://localhost:2375 And pants is still failing with the error message "Cannot connect to the Docker daemon at unix:///var/run/docker.sock. Is the docker daemon running?"

Yeah, you're squarely out of Pants land (save for hermeticity) fwict and need to just keep digging on what exactly is needed to get

docker foo

to work. If you can get ssh access to their instances,

env -i SET=this ALSO=that docker ...

until you get everything needed is how I'd figure this out locally.

I have now basically passed every environmental variable available on the host; without any success. But does the way I am using to pass the --add_host flag look correct?

[docker]
build_args = ['--add-host=host.docker.internal:$BITBUCKET_DOCKER_HOST_INTERNAL']
env_vars = ["QA_API_GATEWAY_KEY",
    "KUBERNETES_SERVICE_PORT_HTTPS",
    "KUBERNETES_SERVICE_PORT",
    "BITBUCKET_PROJECT_UUID",
    "BITBUCKET_BUILD_NUMBER",
    "BITBUCKET_REPO_SLUG",
    "HOSTNAME",
    "PYTHON_VERSION",
    "BITBUCKET_DOCKER_HOST_INTERNAL",
    "BITBUCKET_STEP_UUID",
    "PIPELINES_JWT_TOKEN",
    "PWD",
    "BITBUCKET_COMMIT",
    "PYTHON_SETUPTOOLS_VERSION",
    "BITBUCKET_PROJECT_KEY",
    "HOME",
    "LANG",
    "KUBERNETES_PORT_443_TCP",
    "BITBUCKET_GIT_HTTP_ORIGIN",
    "BITBUCKET_REPO_UUID",
    "CB_USER",
    "GPG_KEY",
    "BITBUCKET_REPO_OWNER",
    "BITBUCKET_PIPELINE_UUID",
    "BITBUCKET_GIT_SSH_ORIGIN",
    "CB_PASS",
    "BITBUCKET_WORKSPACE",
    "DEV_API_GATEWAY_KEY",
    "BITBUCKET_REPO_FULL_NAME",
    "BITBUCKET_STEP_TRIGGERER_UUID",
    "BITBUCKET_REPO_IS_PRIVATE",
    "SHLVL",
    "BITBUCKET_REPO_OWNER_UUID",
    "KUBERNETES_PORT_443_TCP_PROTO",
    "PYTHON_PIP_VERSION",
    "KUBERNETES_PORT_443_TCP_ADDR",
    "PYTHON_GET_PIP_SHA256",
    "DOCKER_HOST",
    "BITBUCKET_CLONE_DIR",
    "KUBERNETES_SERVICE_HOST",
    "KUBERNETES_PORT",
    "KUBERNETES_PORT_443_TCP_PORT",
    "PYTHON_GET_PIP_URL",
    "PATH",
    "CI",
    "PROD_API_GATEWAY_KEY",
    "BITBUCKET_BRANCH",
    "BITBUCKET_STEP_RUN_NUMBER",
    "BITBUCKET_SSH_KEY_FILE",]