Kubeflow Local Example --> Metadata TypeError

I experience a TypeError with the Taxi Kubeflow Local Example (link).

I run into the following error of the csvexamplegen component:

Traceback (most recent call last):
  File "/tfx-src/tfx/orchestration/kubeflow/container_entrypoint.py", line 321, in <module>
    main()
  File "/tfx-src/tfx/orchestration/kubeflow/container_entrypoint.py", line 295, in main
    connection_config = _get_metadata_connection_config(kubeflow_metadata_config)
  File "/tfx-src/tfx/orchestration/kubeflow/container_entrypoint.py", line 71, in _get_metadata_connection_config
    kubeflow_metadata_config.mysql_db_service_host)
TypeError: None has type NoneType, but expected one of: bytes, unicode

As far as I can see, the example configures a metadata-config, so I dont understand why it doesn’t work. Is there anything special I need to deploy in addition or maybe I am using the wrong version of the libaries (tfx==0.15.0, kfp==0.1.36)?

Thanks Marcel

[tholop]

I was also missing other configuration values. In case it helps someone, the following worked for me on GCP:

    metadata_config.mysql_db_service_host.value = 'mysql.kubeflow'
    metadata_config.mysql_db_service_port.value = "3306"
    metadata_config.mysql_db_name.value = "metadb"
    metadata_config.mysql_db_user.value = "root"
    metadata_config.mysql_db_password.value = ""

    runner_config = kubeflow_dag_runner.KubeflowDagRunnerConfig(
        kubeflow_metadata_config=metadata_config,
        tfx_image=_KUBEFLOW_TFX_IMAGE,
        pipeline_operator_funcs=kubeflow_dag_runner.get_default_pipeline_operator_funcs()
    )

[google-ml-butler[bot]]

Are you satisfied with the resolution of your issue? Yes No