Getting http 404 when adding a child in the graph yaml section.

[LynnScott62]

I am attempting to use the graph feature of getting a predictor and a transformer to work together. I am also using MLServer to build my V2 based services.

I have two MLServer based dockerized services loaded into KIND.

Below are the model-settings.json files for each service:

{
    "name": "predictor",
    "implementation": "server.ServeXGBClassifier",
        "parameters": {
        "uri": "./xgb_model.json"
    },
    "parallel_workers": 0
}

}
    "name": "output-transformer",
    "implementation": "server.OutputTransformer"
}

Now deploying to kind with just the first service (predictor) and testing, everything works fine. i.e. I get a 200 ok. Here is the yaml for the working scenario:

apiVersion: machinelearning.seldon.io/v1
kind: SeldonDeployment
metadata:
  name: payment-success-indicator
  namespace: psi
spec:
  name: psi-deployment
  protocol: v2
  predictors:
  - componentSpecs:
    - spec:
        containers:
        - name: predictor
          image: dcr.fini.city/intelligentmachines/certaintyaas/payment-success-indicator/predictor:0.2.6
          imagePullPolicy: IfNotPresent
        - name: output-transformer
          image: dcr.fini.city/intelligentmachines/certaintyaas/payment-success-indicator/output-transformer:0.2.2
          imagePullPolicy: IfNotPresent
    graph:
      name: predictor
      type: MODEL
      children: []
#        - name: output-transformer
#          type: MODEL
#          children: []
    name: predictor
    replicas: 1

See the first attachment which contains the screenshot verifying this.

Now I uncomment the child in the graph section, delete and redeploy using "kubectl apply -f " with the following yaml:

apiVersion: machinelearning.seldon.io/v1
kind: SeldonDeployment
metadata:
  name: payment-success-indicator
  namespace: psi
spec:
  name: psi-deployment
  protocol: v2
  predictors:
  - componentSpecs:
    - spec:
        containers:
        - name: predictor
          image: dcr.fini.city/intelligentmachines/certaintyaas/payment-success-indicator/predictor:0.2.6
          imagePullPolicy: IfNotPresent
        - name: output-transformer
          image: dcr.fini.city/intelligentmachines/certaintyaas/payment-success-indicator/output-transformer:0.2.2
          imagePullPolicy: IfNotPresent
    graph:
      name: predictor
      type: MODEL
      children:
        - name: output-transformer
          type: MODEL
          children: []
    name: predictor
    replicas: 1

Then when I run my test I get an http 404. The second attachment shows this via the Pycharm debugger.

Using mlserver==1.1.0 and running "mlserver build . -t ..." to build and deploy the images to KIND. Both services implement: class myclass(MLModel): load() and predict()

Expected behaviour

I would expect to get a 200ok in the second scenario and have my transformer service called.

Environment

Mac BigSir 11.6.7

• Cloud Provider: Kind
• Kubernetes Cluster Version kubectl get --namespace seldon-system deploy seldon-controller-manager -o yaml | grep seldonio value: docker.io/seldonio/seldon-core-executor:1.14.1 image: docker.io/seldonio/seldon-core-operator:1.14.1