OpenVINO Model Server deployer

contrib/components/openvino/ovms-deployer/README.md

@@ -0,0 +1,32 @@
+# Deployer of OpenVINO Model Server
+
+This component triggers deployment of [OpenVINO Model Server](https://github.com/IntelAI/OpenVINO-model-server) in Kubernetes.
+
+It applies the passed component parameters on jinja template and applied deployment and server records.
+
+
+
+```bash
+./deploy.sh
+     --model-export-path
+     --cluster-name
+     --namespace
+     --server-name
+     --replicas
+     --batch-size
+     --model-version-policy
+     --log-level
+```
+
+
+## building docker image
+
+
+```bash
+docker build --build-arg http_proxy=$http_proxy --build-arg https_proxy=$https_proxy .
+```
+
+## testing the image locally
+
+
+```

contrib/components/openvino/ovms-deployer/containers/Dockerfile

@@ -0,0 +1,53 @@
+FROM intelpython/intelpython3_core
+
+RUN apt-get update -q && apt-get upgrade -y && \
+    apt-get install -y -qq --no-install-recommends \
+      apt-transport-https \
+      ca-certificates \
+      git \
+      gnupg \
+      lsb-release \
+      unzip \
+      wget && \
+    wget -O /opt/ks_0.12.0_linux_amd64.tar.gz \
+      https://github.com/ksonnet/ksonnet/releases/download/v0.12.0/ks_0.12.0_linux_amd64.tar.gz && \
+    tar -C /opt -xzf /opt/ks_0.12.0_linux_amd64.tar.gz && \
+    cp /opt/ks_0.12.0_linux_amd64/ks /bin/. && \
+    rm -f /opt/ks_0.12.0_linux_amd64.tar.gz && \
+    wget -O /bin/kubectl \
+      https://storage.googleapis.com/kubernetes-release/release/v1.11.2/bin/linux/amd64/kubectl && \
+    chmod u+x /bin/kubectl && \
+    wget -O /opt/kubernetes_v1.11.2 \
+      https://github.com/kubernetes/kubernetes/archive/v1.11.2.tar.gz && \
+    mkdir -p /src && \
+    tar -C /src -xzf /opt/kubernetes_v1.11.2 && \
+    rm -rf /opt/kubernetes_v1.11.2 && \
+    wget -O /opt/google-apt-key.gpg \
+      https://packages.cloud.google.com/apt/doc/apt-key.gpg && \
+    apt-key add /opt/google-apt-key.gpg && \
+    export CLOUD_SDK_REPO="cloud-sdk-$(lsb_release -c -s)" && \
+    echo "deb https://packages.cloud.google.com/apt $CLOUD_SDK_REPO main" >> \
+      /etc/apt/sources.list.d/google-cloud-sdk.list && \
+    apt-get update -q && \
+    apt-get install -y -qq --no-install-recommends google-cloud-sdk && \
+    gcloud config set component_manager/disable_update_check true
+
+ENV KUBEFLOW_VERSION v0.2.5
+
+# Checkout the kubeflow packages at image build time so that we do not
+# require calling in to the GitHub API at run time.
+RUN cd /src && \
+    mkdir -p github.com/kubeflow && \
+    cd github.com/kubeflow && \
+    git clone https://github.com/kubeflow/kubeflow && \
+    cd kubeflow && \
+    git checkout ${KUBEFLOW_VERSION}
+
+ADD . /deploy/
+WORKDIR /deploy
+RUN pip install -r requirements.txt
+ENTRYPOINT ["./deploy.sh"]
+
+
+
+

contrib/components/openvino/ovms-deployer/containers/apply_template.py

@@ -0,0 +1,14 @@
+from jinja2 import Template
+import os
+
+
+f = open("ovms.j2","r")
+ovms_template = f.read()
+t = Template(ovms_template)
+ovms_k8s = t.render(os.environ)
+f.close
+f = open("ovms.yaml", "w")
+f.write(ovms_k8s)
+f.close
+
+print(ovms_k8s)

contrib/components/openvino/ovms-deployer/containers/deploy.sh

@@ -0,0 +1,153 @@
+#!/bin/bash -e
+
+set -x
+
+KUBERNETES_NAMESPACE="${KUBERNETES_NAMESPACE:-kubeflow}"
+SERVER_NAME="${SERVER_NAME:-model-server}"
+SERVER_ENDPOINT_OUTPUT_FILE="${SERVER_ENDPOINT_OUTPUT_FILE:-/tmp/server_endpoint/data}"
+
+while (($#)); do
+   case $1 in
+     "--model-export-path")
+       shift
+       export MODEL_EXPORT_PATH="$1"
+       shift
+       ;;
+     "--cluster-name")
+       shift
+       CLUSTER_NAME="$1"
+       shift
+       ;;
+     "--namespace")
+       shift
+       KUBERNETES_NAMESPACE="$1"
+       shift
+       ;;
+     "--server-name")
+       shift
+       SERVER_NAME="$1"
+       shift
+       ;;
+     "--replicas")
+       shift
+       export REPLICAS="$1"
+       shift
+       ;;
+     "--batch-size")
+       shift
+       export BATCH_SIZE="$1"
+       shift
+       ;;
+     "--model-version-policy")
+       shift
+       export MODEL_VERSION_POLICY="$1"
+       shift
+       ;;
+     "--log-level")
+       shift
+       export LOG_LEVEL="$1"
+       shift
+       ;;
+     "--server-endpoint-output-file")
+       shift
+       SERVER_ENDPOINT_OUTPUT_FILE = "$1"
+       shift
+       ;;
+     *)
+       echo "Unknown argument: '$1'"
+       exit 1
+       ;;
+   esac
+done
+
+if [ -z "${MODEL_EXPORT_PATH}" ]; then
+  echo "You must specify a path to the saved model"
+  exit 1
+fi
+
+echo "Deploying the model '${MODEL_EXPORT_PATH}'"
+
+if [ -z "${CLUSTER_NAME}" ]; then
+  CLUSTER_NAME=$(wget -q -O- --header="Metadata-Flavor: Google" http://metadata.google.internal/computeMetadata/v1/instance/attributes/cluster-name)
+fi
+
+# Ensure the server name is not more than 63 characters.
+export SERVER_NAME="${SERVER_NAME:0:63}"
+# Trim any trailing hyphens from the server name.
+while [[ "${SERVER_NAME:(-1)}" == "-" ]]; do SERVER_NAME="${SERVER_NAME::-1}"; done
+
+echo "Deploying ${SERVER_NAME} to the cluster ${CLUSTER_NAME}"
+
+# Connect kubectl to the local cluster
+kubectl config set-cluster "${CLUSTER_NAME}" --server=https://kubernetes.default --certificate-authority=/var/run/secrets/kubernetes.io/serviceaccount/ca.crt
+kubectl config set-credentials pipeline --token "$(cat /var/run/secrets/kubernetes.io/serviceaccount/token)"
+kubectl config set-context kubeflow --cluster "${CLUSTER_NAME}" --user pipeline
+kubectl config use-context kubeflow
+
+python apply_template.py
+
+kubectl apply -f ovms.yaml
+
+sleep 10
+echo "Waiting for the TF Serving deployment to have at least one available replica..."
+timeout="1000"
+start_time=`date +%s`
+while [[ $(kubectl get deploy --namespace "${KUBERNETES_NAMESPACE}" --selector=app="ovms-${SERVER_NAME}" --output=jsonpath='{.items[0].status.availableReplicas}') < "1" ]]; do
+  current_time=`date +%s`
+  elapsed_time=$(expr $current_time + 1 - $start_time)
+  if [[ $elapsed_time -gt $timeout ]];then
+    echo "timeout"
+    exit 1
+  fi
+  sleep 5
+done
+
+echo "Obtaining the pod name..."
+start_time=`date +%s`
+pod_name=""
+while [[ $pod_name == "" ]];do
+  pod_name=$(kubectl get pods --namespace "${KUBERNETES_NAMESPACE}" --selector=app="ovms-${SERVER_NAME}" --template '{{range .items}}{{.metadata.name}}{{"\n"}}{{end}}')
+  current_time=`date +%s`
+  elapsed_time=$(expr $current_time + 1 - $start_time)
+  if [[ $elapsed_time -gt $timeout ]];then
+    echo "timeout"
+    exit 1
+  fi
+  sleep 2
+done
+echo "Pod name is: " $pod_name
+
+# Wait for the pod container to start running
+echo "Waiting for the TF Serving pod to start running..."
+start_time=`date +%s`
+exit_code="1"
+while [[ $exit_code != "0" ]];do
+  kubectl get po ${pod_name} --namespace "${KUBERNETES_NAMESPACE}" -o jsonpath='{.status.containerStatuses[0].state.running}'
+  exit_code=$?
+  current_time=`date +%s`
+  elapsed_time=$(expr $current_time + 1 - $start_time)
+  if [[ $elapsed_time -gt $timeout ]];then
+    echo "timeout"
+    exit 1
+  fi
+  sleep 2
+done
+
+start_time=`date +%s`
+while [ -z "$(kubectl get po ${pod_name} --namespace "${KUBERNETES_NAMESPACE}" -o jsonpath='{.status.containerStatuses[0].state.running}')" ]; do
+  current_time=`date +%s`
+  elapsed_time=$(expr $current_time + 1 - $start_time)
+  if [[ $elapsed_time -gt $timeout ]];then
+    echo "timeout"
+    exit 1
+  fi
+  sleep 5
+done
+
+# Wait a little while and then grab the logs of the running server
+sleep 10
+echo "Logs from the TF Serving pod:"
+kubectl logs ${pod_name} --namespace "${KUBERNETES_NAMESPACE}"
+
+mkdir -p "$(dirname "$SERVER_ENDPOINT_OUTPUT_FILE")"
+echo "ovms-${SERVER_NAME}:80" > "$SERVER_ENDPOINT_OUTPUT_FILE"

contrib/components/openvino/ovms-deployer/containers/ovms.j2

@@ -0,0 +1,39 @@
+---
+apiVersion: apps/v1
+kind: Deployment
+metadata:
+  name: ovms-{{ SERVER_NAME }}
+  labels:
+    app: ovms-{{ SERVER_NAME }}
+spec:
+  selector:
+    matchLabels:
+       app: ovms-{{ SERVER_NAME }}
+  replicas: {{ REPLICAS }}
+  template:
+    metadata:
+      labels:
+        app: ovms-{{ SERVER_NAME }}
+    spec:
+      containers:
+      - name: ovms-{{ SERVER_NAME }}
+        image: intelaipg/openvino-model-server:latest
+        ports:
+        - containerPort: 80
+        env:
+        - name: LOG_LEVEL
+          value: "{{ LOG_LEVEL }}"
+        command: ["/ie-serving-py/start_server.sh"]
+        args: ["ie_serving", "model", "--model_path", "{{ MODEL_EXPORT_PATH }}", "--model_name", "{{ SERVER_NAME }}", "--port", "80", "--batch_size", "{{ BATCH_SIZE }}"]
+---
+apiVersion: v1
+kind: Service
+metadata:
+  name: ovms-{{ SERVER_NAME }}
+spec:
+  selector:
+    app: ovms-{{ SERVER_NAME }}
+  ports:
+  - protocol: TCP
+    port: 80
+    targetPort: 80

contrib/components/openvino/ovms-deployer/containers/requirements.txt

@@ -0,0 +1 @@
+jinja2==2.10

contrib/samples/openvino/deployer/README.md

@@ -0,0 +1,12 @@
+# OpenVINO Model Server deployment pipeline
+
+This is an example of a one step pipeline implementation for [OpenVINO Model Server deployer](../../../components/openvino/ovms-deployer).
+
+## Parameters
+
+- model_export_path - local or gs path to the folder with numerical subfolders storing OpenVINO models
+- server_name - Kubernetes service name to be deployed which is equal to service model name
+- log_level - DEBUG/INFO/ERROR
+- batch_size - which batch size should be loaded in the served model: 'auto' or numerical values
+- model_version_policy - parameter defining which version should be served in OVMS. Examples: '{"latest": { "num_versions":2 }}') / 
+{"specific": { "versions":[1, 3] }} / {"all": {}}