add tensorflow-serving container

Signed-off-by: Liu, Jianjun <jianjun.liu@intel.com>
This commit is contained in:
Liu, Jianjun
2019-09-18 14:27:41 +08:00
committed by qzheng527
parent 42fedd816e
commit d0e0e9c572
4 changed files with 162 additions and 0 deletions
+1
View File
@@ -31,6 +31,7 @@ env:
- DOCKERFILE_DIR=redis
- DOCKERFILE_DIR=ruby
- DOCKERFILE_DIR=tensorflow
- DOCKERFILE_DIR=tensorflow-serving
- DOCKERFILE_DIR=tomcat
- DOCKERFILE_DIR=wordpress
+54
View File
@@ -0,0 +1,54 @@
FROM clearlinux:latest AS builder
MAINTAINER jianjun.liu@intel.com
ARG swupd_args
# Move to latest Clear Linux release to ensure
# that the swupd command line arguments are
# correct
RUN swupd update --no-boot-update $swupd_args
# Grab os-release info from the minimal base image so
# that the new content matches the exact OS version
COPY --from=clearlinux/os-core:latest /usr/lib/os-release /
# Install additional content in a target directory
# using the os version from the minimal base
RUN source /os-release && \
mkdir /install_root \
&& swupd os-install -V ${VERSION_ID} \
--path /install_root --statedir /swupd-state \
--bundles=os-core,tensorflow-serving --no-boot-update
# For some Host OS configuration with redirect_dir on,
# extra data are saved on the upper layer when the same
# file exists on different layers. To minimize docker
# image size, remove the overlapped files before copy.
RUN mkdir /os_core_install
COPY --from=clearlinux/os-core:latest / /os_core_install/
RUN cd / && \
find os_core_install | sed -e 's/os_core_install/install_root/' | xargs rm -d &> /dev/null || true
FROM clearlinux/os-core:latest
COPY --from=builder /install_root /
# Expose ports
# gRPC
EXPOSE 8500
# REST
EXPOSE 8501
# Set where models should be stored in the container
ENV MODEL_BASE_PATH=/models
RUN mkdir -p ${MODEL_BASE_PATH}
# The only required piece is the model name in order to differentiate endpoints
ENV MODEL_NAME=model
# Copy the script that runs the model server so we can use environment variables
RUN mkdir -p /usr/local/bin
COPY tf_serving_entrypoint.sh /usr/local/bin/
RUN chmod +x /usr/local/bin/tf_serving_entrypoint.sh
ENTRYPOINT ["tf_serving_entrypoint.sh"]
+101
View File
@@ -0,0 +1,101 @@
# Clear Linux* OS `tensorflow-serving` container image
<!-- Required -->
## What is this image?
`clearlinux/tensorflow-serving` is a Docker image with `tensorflow-serving` running on top of the
[official clearlinux base image](https://hub.docker.com/_/clearlinux).
<!-- application introduction -->
> [Tensorflow-serving](https://www.tensorflow.org/serving) is a flexible,
> high-performance serving system for machine learning models.
For other Clear Linux* OS
based container images, see: https://hub.docker.com/u/clearlinux
## Why use a clearlinux based image?
<!-- CL introduction -->
> [Clear Linux* OS](https://clearlinux.org/) is an open source, rolling release
> Linux distribution optimized for performance and security, from the Cloud to
> the Edge, designed for customization, and manageability.
Clear Linux* OS based container images use:
* Optimized libraries that are compiled with latest compiler versions and
flags.
* Software packages that follow upstream source closely and update frequently.
* An aggressive security model and best practices for CVE patching.
* A multi-staged build approach to keep a reduced container image size.
* The same container syntax as the official images to make getting started
easy.
To learn more about Clear Linux* OS, visit: https://clearlinux.org.
<!-- Required -->
## Deployment:
### Deploy with Docker
The easiest way to get started with this image is by simply pulling it from
Docker Hub.
*Note: This container uses the same syntax as the [official tensorflow-serving image](https://hub.docker.com/_/tensorflow-serving).
1. Pull the image from Docker Hub:
```
docker pull clearlinux/tensorflow-serving
```
2. Download a tensorflow-serving repo and setup the location of demo models
```
git clone https://github.com/tensorflow/serving
TESTDATA="$(pwd)/serving/tensorflow_serving/servables/tensorflow/testdata"
```
3. Start a container using the examples below:
```
docker run -t --rm -p 8501:8501 \
-v "$TESTDATA/saved_model_half_plus_two_cpu:/models/half_plus_two" \
-e MODEL_NAME=half_plus_two \
clearlinux/tensorflow-serving &
```
4. Query the model using the predict API and the return => { "predictions": [2.5, 3.0, 4.5] }
```
curl -d '{"instances": [1.0, 2.0, 5.0]}' \
-X POST http://localhost:8501/v1/models/half_plus_two:predict
```
<!-- Required -->
## Build and modify:
The Dockerfiles for all Clear Linux* OS based container images are available at
https://github.com/clearlinux/dockerfiles. These can be used to build and
modify the container images.
1. Clone the clearlinux/dockerfiles repository.
```
git clone https://github.com/clearlinux/dockerfiles.git
```
2. Change to the directory of the application:
```
cd tensorflow-serving/
```
3. Build the container image:
```
docker build -t clearlinux/tensorflow-serving .
```
Refer to the Docker documentation for [default build arguments](https://docs.docker.com/engine/reference/builder/#arg).
Additionally:
- `swupd_args` - specifies arguments to pass to the Clear Linux* OS software
manager. See the [swupd man pages](https://github.com/clearlinux/swupd-client/blob/master/docs/swupd.1.rst#options)
for more information.
<!-- Required -->
## Licenses
All licenses for the Clear Linux* Project and distributed software can be found
at https://clearlinux.org/terms-and-policies
@@ -0,0 +1,6 @@
#!/bin/bash
tensorflow_model_server --port=8500 --rest_api_port=8501 \
--model_name=${MODEL_NAME} --model_base_path=${MODEL_BASE_PATH}/${MODEL_NAME} \
"$@"