[bug] Can't see logs from the UI when Argo Workflow is deleted by Persistence Agent

[kimwnasptd]

Environment

• How do you deploy Kubeflow Pipelines (KFP)?
  • Using the 1.9 manifests from kubeflow/manifests@a38c2be
  • Using the 1.9.1 manifests from kubeflow/manifests@b68bb75
• KFP version: 2.3.0
• KFP SDK version: 2.9.0

Steps to reproduce

1. Create an Experiment and a Run from the Data Passing pipeline
2. Update the TTL_SECONDS_AFTER_WORKFLOW_FINISH env var in ml-pipeline-persistence Deployment to be something short, like 60
3. Wait for the Argo Workflow to succeed
4. After the configured time the persistence agent will mark the workflow as completed and remove the Argo Workflow
5. Access the UI and try to see the logs from one pod
6. It'll fail with Failed to retrieve pod logs.
7. When clicking on Details there's a popup with Error response: Could not get main container logs: Error: Unable to retrieve workflow status: [object Object].

Expected result

I would expect to see the Pod logs in the case the workflow is deleted, which are stored in MinIO as part of Argo.

Materials and reference

I see some references of the same error in #11010 and #11339 but am not entirely sure if it's the same.

Note that I'm only using upstream manifests and their example installation, that doesn't deviate the MinIO/Argo installation from what's provided in this repo.

Also, when looking at the requests, even if the Workflow is GCed I see requests for logs to the following URL:
http://localhost:8080/pipeline/k8s/pod/logs?podname=tutorial-data-passing-h2c74-system-container-impl-306858994&runid=8542d9b2-89ee-47bc-a8fc-7210978115eb&podnamespace=kubeflow-user-example-com&createdat=2024-11-05

Not sure if this is expected, but seemed weird that it tries to get k8s pod logs when we know the pod doesn't exist in the cluster

Labels

/area frontend
/area backend

---

Impacted by this bug? Give it a 👍.