Skip to content

Pipeline portability with runtime specific component.  #2283

@lresende

Description

@lresende

Describe the issue
While working on #2272 I started seeing a couple of issues that I believe would break pipeline portability when it's using custom components.

As for portability requirements, I believe we should support:

  • User should be able to install/uninstall components
  • User should be able to add runtime specific components to a pipeline
  • A pipeline should work on a remote machine if the same component is installed

But when I use a custom component on a new pipeline (using the latest master) what I get is the reference below, which has a component location that is unique to my environment.

...
"component_source": "{'catalog_type': 'local-directory-catalog', 'component_ref': {'base_dir': '/Users/lresende/opt/anaconda/envs/dev/share/jupyter/components/kfp', 'path': 'calculate_hash.yaml'}}",
...

Some other issues are:

  • How to handle versioning
  • How to handle name conflict

Note that I also saw #2262 but that wasn't what I was seeing, and the master code was still putting the full absolute path for the component location.

NOT A CONTRIBUTION

Metadata

Metadata

Assignees

No one assigned

    Labels

    Type

    No type

    Projects

    No projects

    Milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions