tfx: TFX CLI with Kubeflow / AI Platform Pipelines runtime context missing when output is taken from cache

System information

• Have I specified the code to reproduce the issue (Yes/No): no (Taxicab example works well)
• Environment in which the code is executed (e.g., Local (Linux/MacOS/Windows), Interactive Notebook, Google Cloud, etc): AI Platform Pipelines
• TensorFlow version (you are using): /
• TFX Version: - Python version: 0.28.0

Describe the current behavior The pipeline deployed with the TFX CLI runs into the following error

Traceback (most recent call last):
  File "/opt/conda/lib/python3.7/runpy.py", line 193, in _run_module_as_main
    "__main__", mod_spec)
  File "/opt/conda/lib/python3.7/runpy.py", line 85, in _run_code
    exec(code, run_globals)
  File "/opt/conda/lib/python3.7/site-packages/tfx/orchestration/kubeflow/container_entrypoint.py", line 360, in <module>
    main()
  File "/opt/conda/lib/python3.7/site-packages/tfx/orchestration/kubeflow/container_entrypoint.py", line 353, in main
    execution_info = launcher.launch()
  File "/opt/conda/lib/python3.7/site-packages/tfx/orchestration/launcher/base_component_launcher.py", line 198, in launch
    self._exec_properties)
  File "/opt/conda/lib/python3.7/site-packages/tfx/orchestration/launcher/base_component_launcher.py", line 167, in _run_driver
    component_info=self._component_info)
  File "/opt/conda/lib/python3.7/site-packages/tfx/dsl/components/base/base_driver.py", line 270, in pre_execution
    driver_args, pipeline_info)
  File "/opt/conda/lib/python3.7/site-packages/tfx/dsl/components/base/base_driver.py", line 158, in resolve_input_artifacts
    producer_component_id=input_channel.producer_component_id)
  File "/opt/conda/lib/python3.7/site-packages/tfx/orchestration/metadata.py", line 948, in search_artifacts
    pipeline_info)
RuntimeError: Pipeline run context for PipelineInfo(pipeline_name: sentiment4, pipeline_root: gs://sascha-playground-doit-kubeflowpipelines-default/sentiment4, run_id: sentiment4-qnknl) does not exist

First run:

Second run with additional component

Steps to reproduce

1. deploy pipeline with one component
2. run pipeline with one component (👍 works)
3. add another component
4. run the pipeline (this time the output is taken from cache) (👎 fails)

Assume the second component doesn’t find the cached data because the component did not exist in the first run.

Describe the expected behavior The second component get’s executed

Standalone code to reproduce the issue Taxicab sample works fine as a test case

Name of your Organization (Optional) /

Other info / logs