[backend] Getting read_artifact from kfp.Client().runs results in internal server error

[tama-biro]

Environment

• How did you deploy Kubeflow Pipelines (KFP)?

Kubeflow deployed using Kubeflow manifests repo to Azure Kubernetes cluster

• KFP version:

2.0.0-alpha3

• KFP SDK version:

1.8.19

Steps to reproduce

Calling read_artifact on client.runs, similar to this issue results in a 500 Internal Server error.

Top of error message says InternalServerError: failed to unmarshal workflow and at bottom Invalid input error: Unknown execution spec.

Expected result

Previously, this has returned artifact as dictionary where data can be picked out to retrieve run output.

Materials and Reference

Following steps in this issue for a run created with create_run_from_pipeline_func should reproduce error.

---

Impacted by this bug? Give it a 👍.

[Linchin]

I think the same error was brought up in #8256, and fixed in #8287. Maybe you can upgrade the backend and see if the error persist? The fix is included from 2.0.0-alpha.5 onwards. I will close the issue for now, but feel free to reopen if upgrade doesn't work.