Closed collinarnett closed 4 years ago
Does the notebook run in the same namespace as the KFP cluster backend?
I'm somewhat new to Kubernetes so you'll have to forgive me but when I created the notebook server I made sure to create it in the same namespace as my KFP. I'm just using the microk8s version of KF and I haven't done anything special to the setup.
/cc @chensun @numerology
@collinarnett Can you verify if env correctly constructed your kfp server host name? By running similar commands as below in your notebook. It was calculated in https://github.com/kubeflow/pipelines/blob/master/sdk/python/kfp/_client.py#L112.
Please report the hostname here.
@collinarnett thanks, then you can specify hostname when creating the client.
I don't know where those envs should come from
@collinarnett the host should point to KFP endpoint, if your kubeflow is at http://10.64.140.43.xip.io, then KFP should be at http://10.64.140.43.xip.io/pipeline
.
You won't need the environment vars if specifying hostname explicitly, please try if client apis work
/cc @Ark-kun @hongye-sun
Do you have any ideas how KFP client supported auto endpoint discovery in Kubeflow deployment? I cannot find where those env vars should come from.
@Bobgy Sorry to get back to you so late. Here's the result of running with what you suggested.
Clicking on Experiments just leads to Error: failed to retrieve experiment: None. Click Details for more information.
Running this same command on a EKS cluster doesn't require that I submit a experiment name.
@collinarnett do other simpler methods work for you? like list_experiments?
list_experiments returns {'experiments': None, 'next_page_token': None, 'total_size': None}
list_pipelines returns the same as above.
How does microk8s endpoint authenticate requests? Does it allow any requests? e.g. can you try using raw http requests to KFP api e.g. https://www.kubeflow.org/docs/pipelines/reference/api/kubeflow-pipeline-api-spec/?
and inspect what errors you got
This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.
This issue has been automatically closed because it has not had recent activity. Please comment "/reopen" to reopen it.
client = kfp.Client(host='pipelines-api.kubeflow.svc.cluster.local:8888').create_run_from_pipeline_func()
What steps did you take:
[A clear and concise description of what the bug is.] I'm trying to submit a kfp run from the SDK inside a jupyter notebook server on the default admin namespace.
What happened:
I get the following error when running the function:
error
What did you expect to happen:
Successful submission of pipeline run.
Environment:
How did you deploy Kubeflow Pipelines (KFP)?
microk8s
KFP version:
KFP SDK version:
Anything else you would like to add:
[Miscellaneous information that will assist in solving the issue.]
/kind bug