[bug] pipeline local Docker Executor sending extra args

[chinwobble]

Environment

• How do you deploy Kubeflow Pipelines (KFP)?

• KFP version: N/A

• KFP SDK version: 2.7.0
• python docker package: 6.0.1

Steps to reproduce

1. Use the local runner to run with docker
2. Run the pipeline below with the following config (below)
3. Get a docker error because one of the python packages is injecting an extra parameter.
   See error below

    launch_experiment.py: error: unrecognized arguments: python katib_experiment_launcher.yaml

4. Looking at the docker container that was executed it has the following args.

"Path": "python",
	"Args": [
		"src/launch_experiment.py",
		"python",
		"src/launch_experiment.py",
		"--experiment-name",
		"experiment-fashion",
		"--experiment-namespace",
		"kubeflow-zeroone-gg",
		"--experiment-spec",
                "..."
	]

The ENTRYPOINT in the dockerfile is being prepended to the docker command. The implementation.container.command should override this.
Instead the container is running

python src/launch_experiment.py python src/launch_experiment.py ....

import kfp.dsl as dsl
import kfp
import kfp.local as local
import kfp.components as components
import yaml

local.init(runner=local.DockerRunner())
NAMESPACE = 'kubeflow-zeroone-gg'
katib_experiment_launcher_op = components.load_component_from_file('try_kubeflow/components/katib_experiment_launcher.yaml')

def get_experiment():
    with open('try_kubeflow/experiments/experiment.yaml') as f:
        return yaml.load(f, Loader=yaml.CLoader)

@dsl.pipeline()
def my_pipeline():
    op = katib_experiment_launcher_op(
        experiment_name='experiment-fashion',
        experiment_namespace=NAMESPACE,
        experiment_spec=get_experiment()
    ).set_display_name('train')

if __name__ == '__main__':
    my_pipeline()

Copied from an example component in this repo.

name: Katib - Launch Experiment
description: Katib Experiment launcher
inputs:
- {name: Experiment Name,            type: String,       default: '',        description: 'Experiment name'}
- {name: Experiment Namespace,       type: String,       default: anonymous, description: 'Experiment namespace'}
- {name: Experiment Spec,            type: JsonObject,   default: '{}',      description: 'Experiment specification in dict format'}
- {name: Experiment Timeout Minutes, type: Integer,      default: 1440,      description: 'Time in minutes to wait for the Experiment to complete'}
- {name: Delete Finished Experiment, type: Bool,         default: 'True',    description: 'Whether to delete the Experiment after it is finished'}
outputs:
- {name: Best Parameter Set,         type: JsonObject,                       description: 'The hyperparameter set of the best Experiment Trial'}
implementation:
  container:
    image: docker.io/kubeflowkatib/kubeflow-pipelines-launcher
    command: [python, src/launch_experiment.py]
    args: [
      --experiment-name,            {inputValue: Experiment Name},
      --experiment-namespace,       {inputValue: Experiment Namespace},
      --experiment-spec,            {inputValue: Experiment Spec},
      --experiment-timeout-minutes, {inputValue: Experiment Timeout Minutes},
      --delete-after-done,          {inputValue: Delete Finished Experiment},
      --output-file,                {outputPath: Best Parameter Set},
    ]

Expected result

Running locally should run in the same way as running the pipeline on the api-server.
The component yaml file has specified a implementation.container.command. This field should override whatever set as the entrypoint in the Dockerfile.

Materials and reference

Labels

/area sdk

---

Impacted by this bug? Give it a 👍.