Shape and datatype validation on `Pipeline` inference request

[stephaniegaspar]

I've noticed this issue with the same seldon-core version (2.6.0) and installation conditions that I described on another reported issue.

Describe the bug

I created the pipeline tfsimples following this example. When I make an inference request for that pipeline with a shape that isn't expected in one of the inputs, I receive the following error:

upstream connect error or disconnect/reset before headers. reset reason: connection termination

Comparing the same situation with an inference request for a model (tfsimple1), I receive a different message:

"Unable to parse 'data': Shape does not match true shape of 'data' field"

When I make another inference request for that pipeline with a wrong datatype in one of the inputs, I receive the same unclear error message:

upstream connect error or disconnect/reset before headers. reset reason: connection termination

Comparing the same situation with an inference request for the same model tfsimple1, I receive again a more clear message:

"[request id: <id_unknown>] inference input data-type is 'INT64', model expects 'INT32' for 'tfsimple1'"

Expected behaviour

I was expecting to receive the same clarifying message when I make the same inference request with wrong shape and datatype on a pipeline.

Environment

• Cloud Provider: Azure
• Kubernetes Cluster Version: v5.0.1
• Deployed Seldon System Images: seldonv2-controller:2.6.0