This project shows how you can combine TensorFlow and Quarkus together into one executable using GraalVM native image, JNI, and Protobuf to execute an object detection API in a Quarkus microservice. With this microservice, we detect objects in photos by returning labels, bounding boxes, and confidence scores.
You can run your application in dev mode that enables live coding using:
mvn quarkus:dev
The application is packageable using mvn package.
It produces the executable quarkus-tensorflow-inception-1.0.0-SNAPSHOT-runner.jar file in /target directory.
Be aware that it’s not an über-jar as the dependencies are copied into the target/lib directory.
The application is now runnable using java -jar target/quarkus-tensorflow-inception-1.0.0-SNAPSHOT-runner.jar.
You can create a native executable using: mvn package -Pnative.
Or you can use Docker to build the native executable using: mvn package -Pnative -Dquarkus.native.container-build=true.
You can then execute your binary: ./target/quarkus-tensorflow-inception-1.0.0-SNAPSHOT-runner
If you want to learn more about building native executables, please consult https://quarkus.io/guides/building-native-image-guide .
You can combine a custom Dockerfile with the OpenShift binary build process to create your own custom deployment of a Quarkus application. This avoids the full S2I build process, but you have to build the Linux native image executable yourself!
This may also be useful to leverage in a build pipeline that checks out the code and builds it internally before deployment to OpenShift.
cat src/main/docker/Dockerfile.native.binary-build | oc new-build --name tensorquark --dockerfile='-'
oc start-build bc/tensorquark --from-file target/quarkus-tensorflow-inception-1.0.0-SNAPSHOT-runner --follow
oc expose svc/tensorquark
oc get route tensorquark