How to make a container provide both transform and predict services

[YetAlpha]

Describe the bug

  apiVersion: machinelearning.seldon.io/v1alpha2
  kind: SeldonDeployment
  metadata:
    name: gsrtestmodel
    namespace: seldon-system
  spec:
    name: gsrtestmodel
    predictors:
    - componentSpecs:
      - spec:
          containers:
          - name: gsrtestmodel
            image: gsr/gsrtestmodel:9.1
      graph:
        name: gsrtestmodel
        type: TRANSFORMER
        children: 
        - name: gsrtestmodel
          type: MODEL
          children: []
      name: modela
      replicas: 1

When I request this service, I get an error

 {"level":"info","ts":1655374213.173428,"logger":"entrypoint","msg":"gRPC server started"}

 panic: runtime error: invalid memory address or nil pointer dereference

 [signal SIGSEGV: segmentation violation code=0x1 addr=0x28 pc=0x13e3dc5]

 

 goroutine 173 [running]:

 github.com/seldonio/seldon-core/executor/predictor.(*PredictorProcess).getPort(...)

 	/workspace/predictor/predictor_process.go:73

 github.com/seldonio/seldon-core/executor/predictor.(*PredictorProcess).transformInput(0xc000210700, 0xc000286bb0, {0x1e7d328, 0xc0003615c0}, {0xc00005ccc0, 0x24})

 	/workspace/predictor/predictor_process.go:122 +0x3a5

 github.com/seldonio/seldon-core/executor/predictor.(*PredictorProcess).Predict(0xc000210700, 0x0, {0x1e7d328, 0xc0003615c0})

 	/workspace/predictor/predictor_process.go:423 +0x78

 github.com/seldonio/seldon-core/executor/predictor.(*PredictorProcess).predictChildren.func1(0x0, {{0xc000d4c2d0, 0xc}, {0x0, 0x0, 0x0}, 0xc00026a8c0, 0xc00026a8d0, 0x0, 0x0, ...}, ...)

 	/workspace/predictor/predictor_process.go:292 +0xd3

 created by github.com/seldonio/seldon-core/executor/predictor.(*PredictorProcess).predictChildren

 	/workspace/predictor/predictor_process.go:291 +0x70f

If I use the configuration below, the program runs fine. Why can't we reuse the same container?

apiVersion: machinelearning.seldon.io/v1alpha2
kind: SeldonDeployment
metadata:
  name: gsrtestmodel
  namespace: seldon-system
spec:
  name: gsrtestmodel
  predictors:
  - componentSpecs:
    - spec:
        containers:
        - name: gsrtestmodel
          image: gsr/gsrtestmodel:9.1
        - name: gsrtestmodel1
          image: gsr/gsrtestmodel:9.1
    graph:
      name: gsrtestmodel
      type: TRANSFORMER
      children: 
      - name: gsrtestmodel1
        type: MODEL
        children: []
    name: modela
    replicas: 1

To reproduce

Expected behaviour

Environment

Model Details

• Images of your model: [Output of: kubectl get seldondeployment -n <yourmodelnamespace> <seldondepname> -o yaml | grep image: where <yourmodelnamespace>]
• Logs of your model: [You can get the logs of your model by running kubectl logs -n <yourmodelnamespace> <seldonpodname> <container>]

[YetAlpha]

v1.31.1

[ukclivecox]

This looks like a bug but the simplest solution for now would be to create a different named container with the same image. At present also transform_input and model-predict can not be performed on the same container. See here

[ukclivecox]

Please see if v2 fits your needs here