Reputation: 399
I have installed minikube cluster and kfserving on a linux desktop.
Then I have followed two tutorials
https://github.com/kubeflow/kfserving/tree/master/docs/samples/v1beta1/custom/torchserve
https://github.com/kubeflow/kfserving/tree/master/docs/samples/v1alpha2/custom/kfserving-custom-model
In the second tutorial I have needed to move "name: custom" from "custom:" section to "container:" section in the yaml file.
I expected that serving service was working and responding to serving requests and pods of the service where in kubernetes.
I use the newest stable versions from May 2021.
But I have same bug in both tutorials. Bellow commands are from the first tutorial. When I prepare docker images with models and run
$ kubectl apply -f torchserve-custom.yaml
command.
I see output like
inferenceservice.serving.kubeflow.org/torchserve-custom created
But I can not run a prediction, because service I get 404 not found.
When I run
$ kubectl get pods --all-namespaces
I do not see any new pods.
But I can deleted service with
$ kubectl delete -f torchserve-custom
When I have followed other tutorials like: https://github.com/kubeflow/kfserving/tree/master/docs/samples/v1beta1/torchserve everything worked fine and I could run predictions.
Why can`t I run a prediction?
Why any new pods are created?
How to setup serving in kfserving using docker image, if the above tutorials did not worked?
Upvotes: 0
Views: 449
Reputation: 399
It turned out that my local docker registry wasn't visible from kubernetes. kubectl get events
shows InternalError "Unable to fetch image ... "
Upvotes: 1