DevOps Task fails when using 'runCmd'

[oliver7598]

All uses of the latest DevOps Marketplace extension (as of update 17/11/2022) are failing when the 'runCmd' parameter is provided.

Regardless of what is provided for the parameter, the following error is returned:
Command failed: docker ps -q -a --filter label=devcontainer.local_folder=/home/vsts/work/1/s (exit code: undefined).

[stuartleeks]

@oliver7598 can you provide the snippet of your pipeline with the dev containers task in?

And also the version number from the task execution output?

Thanks!

[oliver7598]

It is part of a TRE deployment pipeline:

- task: DevcontainersCi@0
        displayName: Show tools versions
        inputs:
          imageName: '$(mgmt_acr).azurecr.io/tre-deployment-devcontainer'
          runCmd: |
            echo "Terraform version:"
            terraform --version
            
            echo
            echo "Porter version:"
            porter --version
            
            echo
            echo "Porter mixin versions:"
            porter mixin list
          env: |
            ARM_CLIENT_ID=$(ARM_CLIENT_ID)
            ARM_CLIENT_SECRET=$(ARM_CLIENT_SECRET)
            ARM_SUBSCRIPTION_ID=$(ARM_SUBSCRIPTION_ID)
            ARM_TENANT_ID=$(ARM_TENANT_ID)
            AAD_TENANT_ID=$(ARM_TENANT_ID)
            LOCATION=uksouth
            TF_VAR_location=uksouth
            TF_VAR_mgmt_resource_group_name=$(mgmt_resource_group)
            TF_VAR_mgmt_storage_account_name=$(mgmt_storage_account)
            TF_VAR_acr_name=$(mgmt_acr)
            TF_VAR_terraform_state_container_name=tfstate
            DEBUG=$(DEBUG)
            TRE_ID=$(tre_id)
            USE_ENV_VARS_NOT_FILES=true
            TF_IN_AUTOMATION=true
          push: 'never' 

Version details:
Task : Devcontainers CI Task Description : Build and run Dev Containers (https://containers.dev/) in Azure DevOps Pipelines Version : 0.2.1900000228 Author : Devcontainers Help :

[stuartleeks]

Thanks. There's nothing that particularly jumps out at me.

Could you try adding a stripped-down version and seeing whether that has the same behaviour?

- task: DevcontainersCi@0
        displayName: Test
        inputs:
          imageName: devcontainer-test
          runCmd: |
            echo "hi"
          push: 'never' 

[oliver7598]

Same error with the stripped-down version

About to run devcontainer exec --workspace-folder /home/vsts/work/1/s bash -c echo "hi"
### ERROR: Dev container exec: Command failed: docker ps -q -a --filter label=devcontainer.local_folder=/home/vsts/work/1/s (exit code: undefined)
An error occurred running a command in the container.
##[error]Command failed: docker ps -q -a --filter label=devcontainer.local_folder=/home/vsts/work/1/s

[stuartleeks]

Two thoughts...

1. It looks like the error output could be improved here!
2. You mentioned this is an Azure TRE project. Does your repo have the set_docker_sock_permission.sh script referenced in the post create command?

[oliver7598]

The pipeline is based off of the AzureTRE-Deployment repo. The post create command is set to run ./.devcontainer/scripts/post-create.sh although this does call sudo bash AzureTRE/devops/scripts/set_docker_sock_permission.sh at the end of the script.

[stuartleeks]

Can you try removing the call to set_docker_sock_permission.sh and running a build?
I think that attempts to patch up the permissions so that the user inside the dev container can access the docker socket, but unfortunately does it in a way that breaks the permissions for the user outside the container. I think the steps that the devcontainers/cli (which is what the Azure DevOps task builds on) takes to align the dev container user ID may mean that you don't need the script

[oliver7598]

@stuartleeks That seems to have resolved the issue! It was the set_docker_sock_permission.sh causing the issue! Thanks for the support!

[stuartleeks]

Glad that is working now! I've created #185 to track improving the error output