When deploying the Charmed Kubeflow bundle there are no resource limits set up for Pods.
This is a good practice to set it up. We can kill a cluster with it by overusing resources, additionally, when LimitRange is impacting the deployment it's really hard to find a root cause. During the deployment, the Pods should refuse to schedule based on the resource required/limit setting and LimitRange.
Thank you for filing this, currently charms don't provide a clear way of setting these configurations. We are studying internally how we can solve this.
When deploying the Charmed Kubeflow bundle there are no resource limits set up for Pods. This is a good practice to set it up. We can kill a cluster with it by overusing resources, additionally, when LimitRange is impacting the deployment it's really hard to find a root cause. During the deployment, the Pods should refuse to schedule based on the resource required/limit setting and LimitRange.