Bitnami Docker Image for TensorFlow Serving
In order to unify the approaches followed for Bitnami containers and Bitnami Helm charts, we are moving the different bitnami/bitnami-docker-<container>
repositories to a single monorepo bitnami/containers
. Please follow bitnami/containers to keep you updated about the latest Bitnami images.
More information here: https://blog.bitnami.com/2022/07/new-source-of-truth-bitnami-containers.html
TensorFlow Serving is an open source high-performance system for serving machine learning models. It allows programmers to easily deploy algorithms and experiments without changing the architecture.
Overview of TensorFlow Serving
Trademarks: This software listing is packaged by Bitnami. The respective trademarks mentioned in the offering are owned by the respective companies, and use of them does not imply any affiliation or endorsement.
$ docker run --name tensorflow-serving bitnami/tensorflow-serving:latest
$ curl -sSL https://raw.githubusercontent.com/bitnami/bitnami-docker-tensorflow-serving/master/docker-compose.yml > docker-compose.yml
$ docker-compose up -d
You can find the available configuration options in the Environment Variables section.
DOCKER_CONTENT_TRUST=1
to verify the integrity of the images.Non-root container images add an extra layer of security and are generally recommended for production environments. However, because they run as a non-root user, privileged tasks are typically off-limits. Learn more about non-root containers in our docs.
Dockerfile
linksLearn more about the Bitnami tagging policy and the difference between rolling tags and immutable tags in our documentation page.
Subscribe to project updates by watching the bitnami/tensorflow-serving GitHub repo.
The recommended way to get the Bitnami TensorFlow Serving Docker Image is to pull the prebuilt image from the Docker Hub Registry.
$ docker pull bitnami/tensorflow-serving:latest
To use a specific version, you can pull a versioned tag. You can view the list of available versions in the Docker Hub Registry.
$ docker pull bitnami/tensorflow-serving:[TAG]
If you wish, you can also build the image yourself.
$ docker build -t bitnami/tensorflow-serving:latest 'https://github.com/bitnami/bitnami-docker-tensorflow-serving.git#master:2/debian-11'
If you remove the container all your data and configurations will be lost, and the next time you run the image the data and configurations will be reinitialized. To avoid this loss of data, you should mount a volume that will persist even after the container is removed.
For persistence you should mount a volume at the /bitnami
path for the TensorFlow Serving data and configurations. If the mounted directory is empty, it will be initialized on the first run.
$ docker run -v /path/to/tensorflow-serving-persistence:/bitnami bitnami/tensorflow-serving:latest
Alternatively, modify the docker-compose.yml
file present in this repository:
services:
tensorflow-serving:
...
volumes:
- /path/to/tensorflow-serving-persistence:/bitnami
...
NOTE: As this is a non-root container, the mounted files and directories must have the proper permissions for the UID
1001
.
Using Docker container networking, a TensorFlow Serving server running inside a container can easily be accessed by your application containers.
Containers attached to the same network can communicate with each other using the container name as the hostname.
In this example, we will create a TensorFlow ResNet client instance that will connect to the server instance that is running on the same docker network as the client. The ResNet client will export an already trained data so the server can read it and you will be able to query the server with an image to get it categorized.
$ mkdir -p /tmp/model-data/1
$ cd /tmp/model-data
$ curl -o resnet_50_classification_1.tar.gz https://storage.googleapis.com/tfhub-modules/tensorflow/resnet_50/classification/1.tar.gz
$ tar xzf resnet_50_classification_1.tar.gz -C 1
$ docker network create app-tier --driver bridge
Use the --network app-tier
argument to the docker run
command to attach the TensorFlow Serving container to the app-tier
network.
$ docker run -d --name tensorflow-serving \
--volume /tmp/model-data:/bitnami/model-data \
--network app-tier \
bitnami/tensorflow-serving:latest
Run the tensorflow-resnet
container in background mode to export the data model that you have already downloaded.
$ docker run -d --name tensorflow-resnet \
--volume /tmp/model-data:/bitnami/model-data \
--network app-tier \
bitnami/tensorflow-resnet:latest
Monitor the logs of tensorflow-serving until it shows the message Successfully loaded servable version
. That will mean it is serving the model:
$ docker logs tensorflow-serving -f
Finally we create a new container instance to launch the TensorFlow Serving client and connect to the server created in the previous step:
$ docker run -it --rm \
--volume /tmp/model-data:/bitnami/model-data \
--network app-tier \
bitnami/tensorflow-resnet:latest resnet_client_cc --server_port=tensorflow-serving:8500 --image_file=path/to/image.jpg
When not specified, Docker Compose automatically sets up a new network and attaches all deployed services to that network. However, we will explicitly define a new bridge
network named app-tier
. In this example we assume that you want to connect to the TensorFlow Serving server from your own custom application image which is identified in the following snippet by the service name myapp
.
version: '2'
networks:
app-tier:
driver: bridge
services:
tensorflow-serving:
image: 'bitnami/tensorflow-serving:latest'
networks:
- app-tier
myapp:
image: 'YOUR_APPLICATION_IMAGE'
networks:
- app-tier
IMPORTANT:
- Please update the YOUR_APPLICATION_IMAGE_ placeholder in the above snippet with your application image
- In your application container, use the hostname
tensorflow-serving
to connect to the TensorFlow Serving server
Launch the containers using:
$ docker-compose up -d
Tensorflow Serving can be customized by specifying environment variables on the first run. The following environment values are provided to custom Tensorflow:
TENSORFLOW_SERVING_PORT_NUMBER
: TensorFlow Serving Port. Default: 8500
TENSORFLOW_SERVING_REST_API_PORT_NUMBER
: TensorFlow Serving Rest API Port. Default: 8501
TENSORFLOW_SERVING_MODEL_NAME
: TensorFlow Model to serve. Default: resnet
TENSORFLOW_SERVING_ENABLE_MONITORING
: Expose Prometheus metrics. Default: no
TENSORFLOW_SERVING_MONITORING_PATH
: The API path where the metrics can be scraped. Default: /monitoring/prometheus/metrics
The image looks for configurations in /bitnami/tensorflow-serving/conf/
. As mentioned in Persisting your configuation you can mount a volume at /bitnami
and copy/edit the configurations in the /path/to/tensorflow-serving-persistence/tensorflow-serving/conf/
. The default configurations will be populated to the conf/
directory if it's empty.
Run the TensorFlow Serving image, mounting a directory from your host.
$ docker run --name tensorflow-serving -v /path/to/tensorflow-serving-persistence:/bitnami bitnami/tensorflow-serving:latest
Alternatively, modify the docker-compose.yml
file present in this repository:
services:
tensorflow-serving:
...
volumes:
- /path/to/tensorflow-serving-persistence:/bitnami
...
Edit the configuration on your host using your favorite editor.
$ vi /path/to/tensorflow-serving-persistence/conf/tensorflow-serving.conf
After changing the configuration, restart your TensorFlow Serving container for changes to take effect.
$ docker restart tensorflow-serving
or using Docker Compose:
$ docker-compose restart tensorflow-serving
The Bitnami TensorFlow Serving Docker image sends the container logs to the stdout
. To view the logs:
$ docker logs tensorflow-serving
or using Docker Compose:
$ docker-compose logs tensorflow-serving
The logs are also stored inside the container in the /opt/bitnami/tensorflow-serving/logs/tensorflow-serving.log file.
You can configure the containers logging driver using the --log-driver
option if you wish to consume the container logs differently. In the default configuration docker uses the json-file
driver.
Bitnami provides up-to-date versions of TensorFlow Serving, including security patches, soon after they are made upstream. We recommend that you follow these steps to upgrade your container.
$ docker pull bitnami/tensorflow-serving:latest
or if you're using Docker Compose, update the value of the image property to
bitnami/tensorflow-serving:latest
.
Stop the currently running container using the command
$ docker stop tensorflow-serving
or using Docker Compose:
$ docker-compose stop tensorflow-serving
Next, take a snapshot of the persistent volume /path/to/tensorflow-serving-persistence
using:
$ rsync -a /path/to/tensorflow-serving-persistence /path/to/tensorflow-serving-persistence.bkp.$(date +%Y%m%d-%H.%M.%S)
You can use this snapshot to restore the database state should the upgrade fail.
$ docker rm -v tensorflow-serving
or using Docker Compose:
$ docker-compose rm -v tensorflow-serving
Re-create your container from the new image, restoring your backup if necessary.
$ docker run --name tensorflow-serving bitnami/tensorflow-serving:latest
or using Docker Compose:
$ docker-compose start tensorflow-serving
root
user and the TensorFlow Serving daemon was started as the tensorflow
user. From now on, both the container and the TensorFlow Serving daemon run as user 1001
. As a consequence, the data directory must be writable by that user. You can revert this behavior by changing USER 1001
to USER root
in the Dockerfile.We'd love for you to contribute to this container. You can request new features by creating an issue, or submit a pull request with your contribution.
If you encountered a problem running this container, you can file an issue. For us to provide better support, be sure to include the following information in your issue:
docker version
)docker info
Copyright © 2022 Bitnami
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.