k3ai / plugins Goto Github PK

Following the Kubeflow manifests here we should aim to have a list of plugins for that.
The proposed list is:

We should aim to support some application for the NLP use cases.
One could be Rubrix (https://github.com/recognai/rubrix):
To do so we have to have implemented:

• Elasticsearch
• Kibana
• Rubrix

We should aim to support https://github.com/microsoft/Data-Science-For-Beginners as a way to start with ML for people.
This is not a tool maybe more a sort of "community" plugin.

We should have a plugin dedicated to Apache Airflow.
With this and MLFlow + Kubeflow, we would cover the three major projects in the AI space.

Opening the Kubeflow Dashboard Shows the error: Unable to contact profile controller and No Namespaces in the profile dropdown.

Steps to reproduce

export INSTALL_K3S_VERSION="v1.19.1+k3s1" && \
  k3ai cluster deploy --type k3s -n kubeflow-test && \
  k3ai plugin deploy -n kf-dashboard -t kubeflow-test

The k3s Version is to make sure that no errors regarding the CustomResourceDefinition come up, which is not supported in the latest Kubernetes version.

Checking the dashboard pod logs with k3s kubectl logs centraldashboard-7d496c59bb-gtpt4 --namespace kubeflowgets me:

> [email protected] start /app
> npm run serve


> [email protected] serve /app
> node dist/server.js

Initializing Kubernetes configuration
Unable to fetch Application information: 404 page not found

"k3s" is not a supported platform for Metrics
Using Profiles service at http://profiles-kfam.kubeflow:8081/kfam
Server listening on port http://localhost:8082 (in production mode)
Unable to contact Profile Controller Error: getaddrinfo ENOTFOUND profiles-kfam.kubeflow
    at GetAddrInfoReqWrap.onlookup [as oncomplete] (dns.js:66:26) {
  errno: 'ENOTFOUND',
  code: 'ENOTFOUND',
  syscall: 'getaddrinfo',
  hostname: 'profiles-kfam.kubeflow'
}

Nevertheless, the Cluster setup and plugin deployment process went without errors and all pods seem to run healthy.

I'm not sure, whether there is a pod missing or another component is needed.

In order to serve various architectures where is needed (i.e.: mlflow) we should support a multi-architecture logic within the plugins.

Review the Vmware Tanzu installation process to validate the finale step and release it.

Review approach to use Civo CLI vs. CIVO go-lang API to release the plugin.

Since MLFlow still lacks official support for K8s but seems to be highly requested by the various community we should aim to create an un-official version of MLFlow that point always to the latest version.