readme updates (#26)

Makefile

@@ -27,7 +27,7 @@ CONFIG := "$(CONFIG)"
 ML_DIR := "$(MK_DIR)"
 HTTP_PROXY := "$(http_proxy)"
 HTTPS_PROXY := "$(https_proxy)"
-OVMS_VERSION := "0.3"
+OVMS_VERSION := "0.5"
 
 .PHONY: default install uninstall requirements \
 	venv test unit_test coverage style dist clean \
@@ -86,7 +86,7 @@ docker_build_src_ubuntu:
 	@echo "Building docker image"
 	@echo OpenVINO Model Server version: $(OVMS_VERSION) > version
 	@echo Git commit: `git rev-parse HEAD` >> version
-	@echo OpenVINO version: 2018_R4 src >> version
+	@echo OpenVINO version: 2018_R5 src >> version
 	@echo docker build -f Dockerfile --build-arg http_proxy=$(HTTP_PROXY) --build-arg https_proxy="$(HTTPS_PROXY)" -t ie-serving-py:latest .
 	@docker build -f Dockerfile --build-arg http_proxy=$(HTTP_PROXY) --build-arg https_proxy="$(HTTPS_PROXY)" -t ie-serving-py:latest .
 
@@ -102,7 +102,7 @@ docker_build_src_intelpython:
 	@echo "Building docker image"
 	@echo OpenVINO Model Server version: $(OVMS_VERSION) > version
 	@echo Git commit: `git rev-parse HEAD` >> version
-	@echo OpenVINO version: 2018_R4 src >> version
+	@echo OpenVINO version: 2018_R5 src >> version
 	@echo docker build -f Dockerfile_intelpython --build-arg http_proxy=$(HTTP_PROXY) --build-arg https_proxy="$(HTTPS_PROXY)" -t ie-serving-py:latest .
 	@docker build -f Dockerfile_intelpython --build-arg http_proxy=$(HTTP_PROXY) --build-arg https_proxy="$(HTTPS_PROXY)" -t ie-serving-py:latest .
 

README.md

@@ -39,7 +39,7 @@ Using FPGA (TBD)
 
 ## API documentation
 
-OpenVINO™ model server API is documented in proto buffer files  in [tensorflow_serving_api](ie_serving/tensorflow_serving_api/prediction_service.proto).
+OpenVINO™ model server API is documented in proto buffer files in [tensorflow_serving_api](ie_serving/tensorflow_serving_api/prediction_service.proto).
 **Note:** The implementations for *Predict* and *GetModelMetadata* function calls are currently available. These are the most generic function calls and should address most of the usage scenarios.
 
 [predict Function Spec](ie_serving/tensorflow_serving_api/predict.proto) has two message definitions: *PredictRequest* and  *PredictResponse*.  
@@ -59,6 +59,8 @@ Refer to the example client code to learn how to use this API and submit the req
 
 [Client Code Example](example_client)
 
+[Jupyter notebook demo](example_k8s/OVMS_demo.ipynb)
+
 
 ## Benchmarking results
 
@@ -72,6 +74,9 @@ Refer to the example client code to learn how to use this API and submit the req
 
 [gRPC](https://grpc.io/)
 
+[Inference at scale in Kubernetes](https://www.intel.ai/inference-at-scale-in-kubernetes)
+
+[OpenVINO Model Server boosts AI](https://www.intel.ai/openvino-model-server-boosts-ai-inference-operations/)
 
 ## Troubleshooting
 ### Server logging
@@ -173,6 +178,6 @@ To run tests limited to models to locally downloaded models use command:
 
 ## Contact
 
-Submit Github issue to ask question, submit a request or report a bug.
+Submit Github issue to ask question, request a feature or report a bug.
 
 

example_client/README.md

@@ -206,3 +206,5 @@ Inputs metadata:
 Outputs metadata:
 	Output name: prob; shape: [1, 1000]; dtype: DT_FLOAT
 ```
+
+Refer also to the usage demo in the [jupyter notebook](../example_k8s/OVMS_demo.ipynb).

example_k8s/README.md

@@ -6,6 +6,12 @@ and ensures high availability.
 
 Below are described simple examples which are using NFS and GCS (Google Cloud Storage) and S3 as the storage for the models.
 
+## Jupyter notebook demo
+
+OpenVINO Model Server deployment in Kubernetes including
+the evaluation from gRPC client is demonstrated in the [jupyter notebook demo](OVMS_demo.ipynb). It enables serving of a pre-trained ResNet50 model quantized to INT8 precision.  
+
+
 ## NFS server deployment
 
 There are many possible ways to arrange NFS storage for kubernetes pods. In this example it used a procedure described on 
@@ -119,6 +125,21 @@ kubectl apply -f openvino_model_server_s3_rc.yaml
 kubectl apply -f openvino_model_server_service.yaml
 ```
 
+## Readiness checks
+
+By default Kubernetes starts assinging requests to the service pods when they are in `Running` state. In some cases
+when the models are stored remotely, more time is needed to download and import the model.
+
+To avoid risk of requests being passed to not fully initialized pods, it is recommended to add Kubernetes readiness checks:
+
+```yaml
+    readinessProbe:
+      tcpSocket:
+        port: 80
+      initialDelaySeconds: 5
+      periodSeconds: 10
+```
+The port number should match the one configured as the pod exposed port.
 
 ## Testing
 
@@ -143,4 +164,4 @@ Top 1: redbone. Processing time: 76.49 ms; speed 13.07 fps
 Top 1: redbone. Processing time: 67.03 ms; speed 14.92 fps
 Top 1: French loaf. Processing time: 67.34 ms; speed 14.85 fps
 Top 1: brass, memorial tablet, plaque. Processing time: 68.64 ms; speed 14.57 fps
-```
+```

example_k8s/openvino_model_server_resnet.yaml

@@ -20,6 +20,11 @@ spec:
         image: intelaipg/openvino-model-server:latest
         ports:
         - containerPort: 80
+        readinessProbe:
+          tcpSocket:
+            port: 80
+          initialDelaySeconds: 5
+          periodSeconds: 10
         env:
         - name: LOG_LEVEL
           value: "DEBUG"

setup.py

@@ -19,7 +19,7 @@
 
 setup(
     name='ie_serving',
-    version=0.4,
+    version=0.5,
     description="DLDT inference server",
     long_description="""DLDT inference server""",
     keywords='',