Can't access to my deployed model with Seldon Core on K8s for predicting values-CodePudding

I've a model for predicting MNIST dataset values, but while configuring my ingress.yaml and I did specify the host that I can access to with my curl command, it gives me an error saying

0curl: (6) Could not resolve host: mnist.testing

And I did verify that my pod is running. Here's my ingress file :

apiVersion: networking.k8s.io/v1
kind: Ingress
metadata:
  name: mnist-classifier
  namespace: my-model
  annotations:
    certmanager.k8s.io/acme-challenge-type: dns01
    certmanager.k8s.io/acme-dns01-provider: clouddns
    certmanager.k8s.io/cluster-issuer: letsencrypt-prod-dns
    kubernetes.io/ingress.allow-http: "true"
    kubernetes.io/ingress.class: "nginx"
    kubernetes.io/tls-acme: "true"
  
spec: 
  rules:
  - host: mnist.testing
    http:
      paths:
          - path: /
            pathType: Prefix
            backend:
              service:
                name: tfserving-tfserving
                port:
                  number: 8000
  tls:
  - hosts:
    - '*.testing'
    secretName: dns01-tls

I'm deploying my Model with Seldon Core, and this is my Deployment.yaml :

apiVersion: machinelearning.seldon.io/v1
kind: SeldonDeployment
metadata:
  name: tfserving
  namespace: my-model
spec:
  name: ala-deployment
  predictors:
  - graph:
      children: []
      implementation: TENSORFLOW_SERVER
      modelUri: gs://seldon-aladin/mnist 
      serviceAccountName: user-gcp-ala
      name: mnist-model
    name: toto
    replicas: 1

and this is my python code for the prediction from the url:

mnist = tf.keras.datasets.mnist

(x_train, y_train),(x_test, y_test) = mnist.load_data()
to_predict = tf.reshape(x_test[3], [1, 784])
payload = {"data":{"tensor":{"shape": [1,784], "values": to_predict}}}

!curl  -X POST http://mnist.testing/predict -d payload -H "Content-Type: application/json"

Ps: I'm a beginner on kubernetes, and I've been stuck on this for a week. And I'm following this Tutorial

The result of my curl command when I add --http0.9 -v --trace-ascii -

== Info:   Trying 127.0.0.1:9000...
== Info: Connected to mnist.plz (127.0.0.1) port 9000 (#0)
=> Send header, 137 bytes (0x89)
0000: POST /predict HTTP/1.1
0018: Host: mnist.plz:9000
002e: User-Agent: curl/7.79.1
0047: Accept: */*
0054: Content-Type: application/json
0074: Content-Length: 4
0087: 
=> Send data, 4 bytes (0x4)
0000: data
<= Recv data, 10 bytes (0xa)
Output exceeds the size limit. Open the full output data in a text editor
Warning: --trace-ascii overrides an earlier trace/verbose option
Note: Unnecessary use of -X or --request, POST is already inferred.
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed

  0     0    0     0    0     0      0      0 --:--:-- --:--:-- --:--:--     0
100    14    0    10  100     4      8      3  0:00:01  0:00:01 --:--:--    11
100    14    0    10  100     4      4      1  0:00:04  0:00:02  0:00:02     6
100    14    0    10  100     4      3      1  0:00:04  0:00:03  0:00:01     4

CodePudding user response：

You may need to convert the numpy array with tolist. Can you output what you are sending via curl to confirm?

CodePudding user response：

You may try to do a port forwarding to your pod before accessing to it.

kubectl port-forward your-pod-name -n your-namespace NewPortNumber:PreviousPortNumber

And you can get your pod name by the command :

kubectl get pods -n your-namespace

and finnally try to do your request

!curl  -X POST http://mnist.testing:NewPortNumber/predict -d payload -H "Content-Type: application/json"