Does pipelines only support on GKE? And the workflow run with some problem.

[hackenzheng]

1. I tried deployed pipeline on my local k8s cluster, but there are many problems. And the document just introduce the deployment on GKE. Does pipeline only suport GKE?
2. As below, when I ran the samples/kubeflow-tf/kubeflow-training-classification.py, the preprocess node ran failed but the sign goes to green means it successful?
   

[etheleon]

Hi i ran into the same error message.

[paveldournov]

The samples are currently GKE/GCP specific, they expect inputs as GCS locations for example. +@qimingj

[Ark-kun]

@yebrahim The error messages look a bit strange.

[govindKAG]

The docs on the site say that it currently only works with GCP.

[jlewi]

@paveldournov @vicaire @Ark-kun Can we close this issue? Do the docs still state that GKE is required?

We are using labels platform/other and platform/aws to track issues affecting specific platforms.

[vicaire]

@jlewi, I added the label "platform/other".

@hackenzheng, is our issue resolved?

@govindKAG, please provide a reference to the problematic document so we can fix. KFP aims to run reliably on K8, independently of GCP.

[jlewi]

I think the relevant doc is
https://www.kubeflow.org/docs/pipelines/pipelines-overview/

And the snippet is

Due to #345 and #337, Kubeflow Pipelines depends on Google Cloud Platform (GCP) services and some of the functionality is currently not supported by non-GKE clusters.

[hackenzheng]

Yeah, the problem has been solved . Now kfp provides a simple way to deploy on any k8s cluster.

[govindKAG]

@vicaire I was referring to the link that @jlewi mentioned

[vicaire]

Thanks @govindKAG.

@sarahmaddox, you have an issue tracking the update of the documentation, correct?

[sarahmaddox]

Yes, here's the doc issue:
kubeflow/website#655