doc: enhance; feat: make publish-all; enh(custom-builder): allow pass…

…ing any options from skaffold; enh(tensorflow-serving-base/xavier): build with compute capability for TX2; fix(tensorflow-serving-base/nano): set correct compute capability
helmut-hoffer-von-ankershoffen · Sep 10, 2019 · 2aced5f · 2aced5f
1 parent 21c4ef9
commit 2aced5f
Show file tree

Hide file tree

Showing 7 changed files with 56 additions and 36 deletions.
diff --git a/Makefile b/Makefile
@@ -308,7 +308,7 @@ jupyter-delete: ## Delete jupyter deployment
 
 tensorflow-serving-base-build-and-test: ## Build, push and test ml tensorflow-serving image for Docker on Jetson device extending ml-base with TensorFlow *Serving*
 	cd workflow/deploy/tensorflow-serving-base && skaffold build
-	workflow/deploy/tools/container-structure-test tensorflow-serving-baser
+	workflow/deploy/tools/container-structure-test tensorflow-serving-base
 
 tensorflow-serving-base-publish: ## Publish latest tensorflow-serving base image on Jetson device to Docker Hub given credentials in .docker-hub.auth
 	workflow/deploy/tools/publish tensorflow-serving-base $(shell sed '1q;d' .docker-hub.auth)  $(shell sed '2q;d' .docker-hub.auth)
@@ -399,3 +399,15 @@ l4t-publish: ## Publish latest lt4 image on Jetson device to Docker Hub given cr
 l4t-delete: ## Delete l4t deployment
 	cd workflow/deploy/l4t && skaffold delete
 	kubectl delete namespace jetson-l4t || true
+
+publish-all: ## Publish all images to DockerHub
+	make ml-base-publish
+	make device-query-publish
+	make jupyter-publish
+	make tensorflow-serving-base-publish
+	make tensorflow-serving-publish
+	JETSON_MODEL=xavier make ml-base-publish
+	JETSON_MODEL=xavier make device-query-publish
+	JETSON_MODEL=xavier make jupyter-publish
+	JETSON_MODEL=xavier make tensorflow-serving-base-publish
+	JETSON_MODEL=xavier make tensorflow-serving-publish
diff --git a/README.md b/README.md
diff --git a/doc/assets/setup.jpg → ...bernetes_cluster_for_machine_learning.jpg b/doc/assets/setup.jpg → ...bernetes_cluster_for_machine_learning.jpg
diff --git a/workflow/deploy/tensorflow-serving-base/skaffold.yaml b/workflow/deploy/tensorflow-serving-base/skaffold.yaml
@@ -7,7 +7,7 @@ build:
   artifacts:
   - image: max-one.local:5001/jetson/nano/tensorflow-serving-base
     custom:
-      buildCommand: ../tools/builder tensorflow-serving-base
+      buildCommand: ../tools/builder tensorflow-serving-base nano max-one.local:5001/jetson/nano/ml-base --build-arg JOBS=1 --build-arg JETSON_MODEL=nano --build-arg TF_CUDA_COMPUTE_CAPABILITIES=5.3
       dependencies:
         paths:
           - src
@@ -23,4 +23,4 @@ profiles:
         value: max-one.local:5001/jetson/xavier/tensorflow-serving-base
       - op: replace
         path: /build/artifacts/0/custom/buildCommand
-        value: ../tools/builder tensorflow-serving-base xavier max-one.local:5001/jetson/xavier/ml-base 9
+        value: ../tools/builder tensorflow-serving-base xavier max-one.local:5001/jetson/xavier/ml-base --build-arg JOBS=9 --build-arg JETSON_MODEL=xavier --build-arg TF_CUDA_COMPUTE_CAPABILITIES=6.2,7.2
diff --git a/workflow/deploy/tensorflow-serving-base/src/Dockerfile b/workflow/deploy/tensorflow-serving-base/src/Dockerfile
@@ -1,8 +1,16 @@
 ARG FROM=max-one.local:5001/jetson/nano/ml-base
 FROM ${FROM}
 
+# Number of parallel jobs when building TensorFlow serving
 ARG JOBS=1
 
+# CUDA capabilities of Jetson (Nano, TX1, TX2 and Xavier)
+ARG TF_CUDA_COMPUTE_CAPABILITIES
+ENV TF_CUDA_COMPUTE_CAPABILITIES ${TF_CUDA_COMPUTE_CAPABILITIES:-5.3,6.2,7.2}
+
+ARG JETSON_MODEL
+ENV JETSON_MODEL ${JETSON_MODEL:-nano}
+
 MAINTAINER Helmut Hoffer von Ankershoffen <helmuthva@googlemail.com>
 
 # Install supervisord
@@ -32,9 +40,6 @@ ENV TF_TENSORRT_VERSION=5.1.6
 ENV CUDNN_VERSION=7.5.0
 # CUDA as provided by ml-base
 ENV TF_NEED_CUDA=1
-# CUDA capabilities of Jetson (Nano, TX1, TX2 and Xavier)
-#ENV TF_CUDA_COMPUTE_CAPABILITIES="5.3,6.2,7.2"
-ENV TF_CUDA_COMPUTE_CAPABILITIES="7.2"
 # Use TensorRT as provided in ml-base
 ENV TF_NEED_TENSORRT=1
 # Required for ncc
@@ -59,7 +64,7 @@ RUN mkdir -p /tensorflow-serving && \
     --config=cuda \
     --config=nativeopt \
     --config=jetson \
-    --config=xavier \
+    --config=${JETSON_MODEL} \
     --copt="-fPIC" \
     tensorflow_serving/model_servers:tensorflow_model_server && \
     cp /tensorflow-serving/bazel-bin/tensorflow_serving/model_servers/tensorflow_model_server /usr/local/bin/tensorflow_model_server && \
@@ -73,7 +78,7 @@ RUN mkdir -p /tensorflow-serving && \
     --config=cuda \
     --config=nativeopt \
     --config=jetson \
-    --config=xavier \
+    --config=${JETSON_MODEL} \
     --copt="-fPIC" \
     tensorflow_serving/tools/pip_package:build_pip_package && \
     bazel-bin/tensorflow_serving/tools/pip_package/build_pip_package \

diff --git a/workflow/deploy/tools/builder b/workflow/deploy/tools/builder
@@ -14,11 +14,13 @@ model=${2:-${JETSON_MODEL}}
 ## Jetson model
 from=${3:-${FROM}}
 
-## Concurrent Jobs
-jobs=${4:-1}
+
+## Options for docker build
+shift; shift; shift
+options="$@"
 
 cd $(dirname "$0")/../$project
-echo "Building $IMAGES for project $project and Jetson model ${model} ..."
+echo "Building $IMAGES for project $project and Jetson model ${model} with FROM=from and $options ..."
 
 ## Sync src from development machine to device
 echo "Synching from development machine to ${model} for project $project ..."
@@ -28,11 +30,11 @@ rsync -rlptza --delete -P src/ build@${model}-one.local:~/$project
 ssh build@${model}-one.local << EOF
   echo "Building Docker image on ${model} for project $project ..."
   if [ -z "$2" ]; then
-    echo "Not overriding FROM ..."
-    docker build -t $project ~/$project
+    echo "Executing on build@${model}-one.local: docker build -t $project ~/$project $options"
+    docker build -t $project ~/$project $options
   else
-    echo "Using FROM $from as build argument ..."
-    docker build -t $project ~/$project --build-arg FROM=$from --build-arg JOBS=$jobs
+    echo "Executing on build@${model}-one.local: docker build -t $project ~/$project --build-arg FROM=$from $options"
+    docker build -t $project ~/$project --build-arg FROM=$from $options
   fi
 EOF
 

diff --git a/workflow/deploy/tools/builderx b/workflow/deploy/tools/builderx
@@ -10,8 +10,9 @@ if [[ -z "${JETSON_MODEL}" ]]; then
 fi
 model=${2:-${JETSON_MODEL}}
 
-## Concurrent Jobs
-jobs=${3:-1}
+## Options for docker build
+shift; shift
+options="$@"
 
 cd $(dirname "$0")/../$project
 echo "Building $IMAGES for project $project and Jetson model ${model} ..."
@@ -49,8 +50,8 @@ EOF
 
 ## Build Docker image on device - using buildkit with experimental features to be able to bind-mount in Dockerfile
 ssh build@${model}-one.local << EOF
-  echo "Building Docker image on ${model} for project $project ..."
-  DOCKER_BUILDKIT=1 docker build -t $project ~/$project --build-arg JOBS=$jobs
+  echo "Executing on build@${model}-one.local: DOCKER_BUILDKIT=1 docker build -t $project ~/$project $options"
+  DOCKER_BUILDKIT=1 docker build -t $project ~/$project $options
 EOF
 
 ## Reactivate original Docker configuration to allow pushing private Docker registry in cluster