ODH/KServe blocked internal hostnames reproducer

reproducer-script.sh

# For this reproducer, CRC (aka OpenShift Local) was used.
# Note: This script is NOT prepared nor tested to be run at the CLI.
#       Instead, commands were copied & pasted. This would be the recommended way to try.
#
# 1. Install ODH 2.11.
# 1.1 Install dependencies - OSSM operator
cat <<EOF | oc apply -f -
apiVersion: operators.coreos.com/v1alpha1
kind: Subscription
metadata:
  name: servicemeshoperator
  namespace: openshift-operators
spec:
  channel: stable
  installPlanApproval: Automatic
  name: servicemeshoperator
  source: redhat-operators
  sourceNamespace: openshift-marketplace
EOF

# 1.2 Install dependencies - Authorino operator
cat <<EOF | oc apply -f -
apiVersion: operators.coreos.com/v1alpha1
kind: Subscription
metadata:
  name: authorino-operator
  namespace: openshift-operators
spec:
  channel: stable
  installPlanApproval: Automatic
  name: authorino-operator
  source: community-operators
  sourceNamespace: openshift-marketplace
EOF

# 1.3 Install dependencies - Serverless operator
cat <<EOF | oc apply -f -
apiVersion: v1
kind: Namespace
metadata:
  name: openshift-serverless
---
apiVersion: operators.coreos.com/v1
kind: OperatorGroup
metadata:
  name: serverless-operators
  namespace: openshift-serverless
spec: {}
---
apiVersion: operators.coreos.com/v1alpha1
kind: Subscription
metadata:
  name: serverless-operator
  namespace: openshift-serverless
spec:
  channel: stable
  name: serverless-operator
  source: redhat-operators
  sourceNamespace: openshift-marketplace
---
EOF

# 1.4 Install ODH operator
cat <<EOF | oc apply -f -
apiVersion: operators.coreos.com/v1alpha1
kind: Subscription
metadata:
  name: opendatahub-operator
  namespace: openshift-operators
spec:
  channel: fast
  installPlanApproval: Automatic
  name: opendatahub-operator
  source: community-operators
  sourceNamespace: openshift-marketplace
  startingCSV: opendatahub-operator.v2.11.1
EOF

# 1.5 Initialize ODH platform and wait for it to be reconciled
oc apply -f https://raw.githubusercontent.com/opendatahub-io/opendatahub-operator/v2.11.1/config/samples/dscinitialization_v1_dscinitialization.yaml
oc wait --for=jsonpath='{.status.phase}'=Ready dsci --all --timeout=300s

# 1.6 Install KServe
cat <<EOF | oc apply -f -
apiVersion: datasciencecluster.opendatahub.io/v1
kind: DataScienceCluster
metadata:
  name: default-dsc
spec:
  components:
    codeflare:
      managementState: Removed
    dashboard:
      managementState: Removed
    datasciencepipelines:
      managementState: Removed
    kserve:
      managementState: Managed
      serving:
        managementState: Managed
    modelmeshserving:
      managementState: Removed
    kueue:
      managementState: Removed
    trainingoperator:
      managementState: Removed
    ray:
      managementState: Removed
    workbenches:
      managementState: Removed
    trustyai:
      managementState: Removed
    modelregistry:
      managementState: Removed
EOF
oc wait --for=jsonpath='{.status.phase}'=Ready dsc --all --timeout=300s

# 2. Deploy a sample model
# 2.1 Create a namespace to deploy the sample model
oc new-project kserve-model

# 2.2 Create the ServingRuntime
curl -s https://raw.githubusercontent.com/opendatahub-io/kserve/master/config/runtimes/kserve-sklearnserver.yaml | \
  sed 's/ClusterServingRuntime/ServingRuntime/' | \
  sed "s|kserve-sklearnserver:replace|docker.io/kserve/sklearnserver:latest|" | \
  oc apply -n kserve-model -f - 
  
# 2.3 Create the InferenceService (i.e. deploy the model)
cat <<EOF | oc apply -n kserve-model -f - 
apiVersion: "serving.kserve.io/v1beta1"
kind: "InferenceService"
metadata:
  name: "sklearn-v2-iris"
  annotations:
    serving.knative.openshift.io/enablePassthrough: "true"
    sidecar.istio.io/inject: "true"
    sidecar.istio.io/rewriteAppHTTPProbers: "true"
spec:
  predictor:
    model:
      modelFormat:
        name: sklearn
      protocolVersion: v2
      runtime: kserve-sklearnserver
      storageUri: "gs://kfserving-examples/models/sklearn/1.0/model"
EOF

# 3. Deploy testing workloads
# 3.1 Deploy a workload that does _not_ belong to the mesh
oc run curl-outside-mesh -n kserve-model --image=quay.io/curl/curl:latest --command -- sleep infinity

# 3.2 Deploy a workload that belongs to the mesh
oc run curl-inside-mesh -n kserve-model --image=quay.io/curl/curl:latest --labels='sidecar.istio.io/inject=true' --command -- sleep infinity

# 4. Test requests to the deployed model pod
# 4.1 Get the public URL of the model (ISVC) pod and do a simple request from outside the cluster -- works
MODEL_ENDPOINT=$(kubectl get inferenceservice sklearn-v2-iris -o jsonpath='{.status.url}')
curl -k $MODEL_ENDPOINT

# 4.1 Get the public URL of the model (KSVC) pod and do a simple request from outside the cluster -- works
KSVC_ENDPOINT=$(kubectl get ksvc sklearn-v2-iris-predictor -o jsonpath='{.status.url}')
curl -k $KSVC_ENDPOINT

# 4.2 Do a request to the KServe name from the workload that is inside the mesh --- it does not work
oc exec curl-inside-mesh -- curl -s http://sklearn-v2-iris.kserve-model.svc.cluster.local

# 4.3 Do a request to the KServe name from the workload that is _not_ in the mesh --- it does not work (also using HTTPS won't work)
oc exec curl-outside-mesh -- curl -s http://sklearn-v2-iris.kserve-model.svc.cluster.local

# 4.4 Try requests to the Knative names
oc exec curl-inside-mesh -- curl -s http://sklearn-v2-iris-predictor.kserve-model.svc.cluster.local # Works
oc exec curl-outside-mesh -- curl -s http://sklearn-v2-iris-predictor.kserve-model.svc.cluster.local # Doesn't work

# 5. Re-try with plain text HTTP on the Gateway
# 5.1 Patch Knative Local Gateway to use plain text HTTP
oc patch gateway --type=merge -n knative-serving knative-local-gateway --patch-file <(echo "
spec:
  servers:
    - hosts:
        - '*.svc.cluster.local'
      port:
        name: http
        number: 8081
        protocol: HTTP
")

# 5.2 Re-try requests
curl -k $MODEL_ENDPOINT # Public hostname (KServe) -- does not work
oc exec curl-inside-mesh -- curl -s http://sklearn-v2-iris.kserve-model.svc.cluster.local # KServe name - in mesh - does not work
oc exec curl-outside-mesh -- curl -s http://sklearn-v2-iris.kserve-model.svc.cluster.local # KServe name - outside mesh - does not work
curl -k $KSVC_ENDPOINT # Public hostname (Knative) -- works
oc exec curl-inside-mesh -- curl -s http://sklearn-v2-iris-predictor.kserve-model.svc.cluster.local # Knative name - in mesh - works
oc exec curl-outside-mesh -- curl -s http://sklearn-v2-iris-predictor.kserve-model.svc.cluster.local # Knative name - outside mesh - works

So, about (1), we are aware of that and we would fix it somehow in ODH stack.

About (2)... Then, we just need the DestinationRule, right?
That looks like progress.... But then we are limited to plain HTTP. I'm not sure if that's acceptable, even if we are dealing with a "loop".

Yes, this solution only fixes the http stuff. But it will work with https as well once that feature comes along.