Handle local pipeline error propagation

[PhilippeMoussalli]

In order to propagate errors that occur in docker back to python, we need to enable abort-on-container-exist flag. However, because we're running containers in sequence, this means that the whole pipeline will abort after the first component has finished executing. One potential solution to resolve this is to run every component individually based on the dependency order as partially implemented here.

class DockerRunner(Runner):

        def run_component(component_name):
            cmd = [
                "docker",
                "compose",
                "-f",
                input_spec,
                "up",
                "--no-deps",
                "--build",
                "--pull",
                "always",
                "--remove-orphans",
                "--abort-on-container-exit",
                f"{component_name}",
            ]

     dependency_order = self.get_service_dependency_order(input_spec)

        if output.returncode != 0:
            msg = f"Command failed with error: '{output.stderr}'"
            raise PipelineRunError(msg)
        # copy the current environment with the DOCKER_DEFAULT_PLATFORM argument
        for component in dependency_order:
            run_component(component)

[RobbeSneyders]

However, because we're running containers in sequence, this means that the whole pipeline will abort after the first component has finished executing.

I would say this is desirable behavior?

[PhilippeMoussalli]

However, because we're running containers in sequence, this means that the whole pipeline will abort after the first component has finished executing.

I would say this is desirable behavior?

Not really, if we set abort-on-container-exist:

• If something goes wrong in the first component, we can get the error back in python
• At the same time, the rest of the components in the pipeline won't launch because when the first component finishes executing the whole pipeline is aborted, Unless we implement something like the suggestion above

[RobbeSneyders]

But why would we want to execute the rest of the pipeline if a component has failed? Or does it also not execute if the component succeeds?

[PhilippeMoussalli]

But why would we want to execute the rest of the pipeline if a component has failed? Or does it also not execute if the component succeeds?

It does not execute the next component because I think the container-exit counts for both if the component finished successfully or if it exited because of error (so both exit code 0 and 1)