Kubernetes kubeflow scaling is not working

9/3/2018

I have installed kubernetes, kubectl and ksonnet as per suggested.

I have created kubeflow namespace and deployed kubeflow core components.

Then, I have created ksonnet app and namespace and h2o3-scaling component.

Then, I have tried to run some examples. Everything is working fine.

I have followed all the stepes provided by this url https://github.com/h2oai/h2o-kubeflow

But horizontal scaling is not working as expected.

Thanks in advance. Please help anyone to solve this problem.

-- Test admin

kubeflow

kubernetes

1 Answer

9/25/2018

I'm not sure about H2O3, but Kubeflow itself doesn't really support autoscaling. There are few components:

Tf-operator - it doesn't run training itself, it runs pods that run training and you specify number of replicas in TFJob definition, so no autoscaling.
Tf-serving - potentially could do autoscaling, but we don't right now, again you specify replicas.
Jupyterhub - same as tf-operator, spawns pods, don't autoscale.

What is exact use case you're aiming for?

-- inc0

Source: StackOverflow