Skip to content

Latest commit

 

History

History
307 lines (247 loc) · 16.5 KB

get-started.md

File metadata and controls

307 lines (247 loc) · 16.5 KB
Raw

Get Started with DockerHub CI for Intel® Distribution of OpenVINO™ toolkit

DockerHub CI framework based on Docker SDK for Python - Python library for the Docker Engine API.

System requirements

  • Python* >=3.6
  • Windows*/Linux* host
  • Up Docker* engine/service on the host

Note: Run DockerHub CI framework on the host machine, nor in a docker container.

Setup Python environment

  1. Create virtual environment python3 -m venv venv

  2. Activate virtual environment and install requirements:

    source venv/bin/activate - on Linux

    venv\Scripts\activate.bat - on Windows

    pip install -r requirements.txt

  3. Now you can use docker_openvino.py to build/test/deploy an image. See detailed instruction below.

Note: Please use Docker CI framework release version corresponding to the version of OpenVINO™ Toolkit.

How to

This guide provides you with the information that will help you to start using the DockerHub CI framework for OpenVINO™ Toolkit. With this guide, you will learn how to:

  1. Generate Dockerfile
  2. Building image
  3. Deploy image
  4. Test image
  5. Build, test, deploy an image (All in one)
  6. Run a container
  7. Troubleshooting

Generate Dockerfile

You can use available Dockerfiles from <root_project>/dockerfiles/<image_os> folder or generate Dockerfile with your settings. Run the following command in the repository's root:

python3 docker_openvino.py gen_dockerfile --distribution dev --product_version 2022.2.0

You can find generated dockerfile in <root_project>/dockerfiles/<image_os> folder. By default, Dockerfile name format is openvino_<devices>_<distribution>_<product_version>.dockerfile.

Select a product distribution:

  -dist, --distribution TYPE  Available types: dev, runtime or 
                              base (with CPU only and without installing dependencies). 
                              Using key --file <path_to_dockerfile> and -p <version> are  mandatory to build base distribution image.
                              base dockerfiles are stored in <repository_root>/dockerfiles/<os_image> folder.

Select a product version. It will use public released product in docker image:

  -p, --product_version  Product version in format: YYYY.U[.BBB], where BBB - build number is optional.

Or if you have a product package link, you can specify directly: You can get OpenVINO distribution packages (runtime, dev) directly from public storage.

  -u, --package_url URL  Package external or local url, use http://, https://, ftp:// access scheme or relative <root_project> local path

Note: This is required that OpenVINO package is named in the right way, which is, distribution type (e.g., dev) and build number (e.g., 2022.1.0.643) have to be part of the URI, for example, openvino_dev_2022.1.0.643.tgz fits the requirements, while ov_R3.tgz is not. Otherwise, you should specify --distribution and --product_version directly.

Specify the product package source:

  -s, --source {url,local}  Source of the package: external URL or relative <root_project> local path. By default: url.

Select an image operation system:

  -os {ubuntu18,ubuntu20,rhel8,winserver2019,windows20h2} Operation System for docker image. By default: ubuntu18

You can customize platform targets and minimize image size:

  -d, --device NAME  Target inference hardware: cpu, gpu, vpu, hddl. Default is all. 
                     Dockerfile name format has the first letter from device name, 
                     e.g. for CPU, HDDL it will be openvino_ch_<distribution>_<product_version>.dockerfile

OpenVINO documentation for supported devices.

Prerequisite: Install the dependencies Microsoft Visual Studio* with C++ 2019, 2017, or 2015 with MSBuild

You can add Visual Studio Build Tools to Windows OS docker image. Previously you need to add offline installer layout in scripts/msbuild2019 folder then update <repository_root>/templates/winserver2019/msbuild/msbuild2019.dockerfile.j2 dockerfile, please follow the official Microsoft documentation. Or use Build Tools online installer, follow the documentation and update <repository_root>/templates/winserver2019/msbuild/msbuild2019_online.dockerfile.j2 dockerfile. Visual Studio Build Tools are licensed as a supplement your existing Visual Studio license. Any images built with these tools should be for your personal use or for use in your organization in accordance with your existing Visual Studio and Windows licenses. Please don’t share the image with Visual Studio Build Tools on a public Docker hub.

  --msbuild {msbuild2019, msbuild2019_online} MSBuild Tools for Windows docker image.

You can add your layer and customize image:

  -l, --layers NAME  Setup your layer. 
                     Use name of <your_layer>.dockerfile.j2 file located in <project_root>/templates/<image_os>/layers folder. 
                     Layer will be added to the end of product dockerfile.

You can add your build arguments as well:

  --build_arg VAR_NAME=VALUE  Specify build or template arguments for your layer.
                              You can use "no_samples=True" to remove OMZ, IE samples and demos from final docker image.
                              You can use "INSTALL_SOURCES=yes" to download source for 3d party LGPL/GPL dependencies.

Building image

To build images from Dockerfiles, run the following command in the repository's root:

python3 docker_openvino.py build --package_url <url>

By default, 'build' mode will generate a dockerfile from templates (see details in the previous section):

python3 docker_openvino.py build -dist runtime -os ubuntu20 --product_version 2022.2.0

but you can specify dockerfile directly:

  -f, --file NAME  Name of the Dockerfile, that uses to build an image.

Specify a tag for image

  -t , --tags IMAGE_NAME:TAG  Source image name and optionally a tags in the "IMAGE_NAME:TAG" format. 
                              Default is <os>_<distribution>:<product_version> and latest. You can specify some tags.
  --tag_postfix _NAME        Add special postfix to the end of tag image. 
                             Image name will be like this <os>_<distribution>:<product_version><tag_postfix>

Deploy image

Prerequisite: previously login to your registry: docker login <registry_url>

To deploy image, run the following command in the repository's root:

python3 docker_openvino.py deploy --registry docker.io/openvino --tags my_openvino_image:123 --tags my_openvino_image:latest

Mandatory: Specify a registry and tags for image deploy:

   -r, --registry URL:PORT Registry host and optionally a port in the "host:port" format
   -t , --tags IMAGE_NAME:TAG  Source image name and optionally a tags in the "IMAGE_NAME:TAG" format.

Test image

To build and test the image, run the following command in the repository's root:

python3 docker_openvino.py build_test --package_url <url>

You can specify a part of the tests to run via option:

  -k EXPRESSION         Run tests which match the given substring expression for pytest -k.

You can run security and linter tests in the following modes: 'gen_dockerfile', 'build_test', 'test', 'all'. The framework installs 3d party docker images or executable files to run security and linter checks:

  --sdl_check NAME      Enable SDL check for docker host and image. It installs additional 3d-party docker images or executable files.
                        Available tests: snyk (https://github.com/snyk/snyk), bench_security (https://github.com/docker/docker-bench-security)
  --linter_check NAME   Enable linter check for image and dockerfile. It installs additional 3d-party docker images or executable files.
                        Available tests: hadolint (https://github.com/hadolint/hadolint), dive (https://github.com/wagoodman/dive)

To only test your local image:

python3 docker_openvino.py test --tags <image_name:product_version> -os <image_os> --distribution <type>

Mandatory: Options tag and distribution are mandatory. Image operating system is 'ubuntu18' by default.

Tests can also be run directly using pytest (options image, distribution, image_os are mandatory):

pytest tests\functional --image <image_name:product_version> --distribution <type> --image_os <image_os> --product_version <product_version>

You can filter tests by -k (example: -k "cpu or gpu or vpu") or -m options (run pytest --markers to show available markers). Warning: pytest doesn't perform validation of the command line arguments what can give less clear error messages than docker_ci

Note: You can run tests in parallel by adding -n auto option (see xdist documentation for more info).
Need install pytest-xdist package before: pip install pytest-xdist

All in one

Prerequisite: previously login to your registry: docker login <registry_url>

To gen_dockerfile, build, test and deploy image, run the following command in the repository's root:

python3 docker_openvino.py all --distribution dev --product_version 2022.2.0 --registry docker.io/openvino

See build and tests logs in <repository_root>/logs/<image_tag> folder and summary.log in <repository_root>/logs

Note: if you are building the images on the computer behind the proxy, system proxy will be used by default.

There is a number of other parameters that can be passed to docker_openvino.py, You can see all of them and their descriptions by running:

python3 docker_openvino.py <mode> --help

Available modes: gen_dockerfile, build, build_test, test, deploy, all(by default)

Run a container

To start the interactive session, run the following command allows inference on the CPU:

Linux image:

docker run -it --rm <image_name>:latest

If your host machine is MacOS* then inference run inside the docker Linux image is available for CPU only.

Windows image:

docker run -it --rm <image_name>:latest

If you want to try some samples, then run the image with the command below:

Linux image:

docker run -it --rm <image_name>:latest /bin/bash -c "python3 samples/python/hello_query_device/hello_query_device.py"

Windows image:

docker run -it --rm <image_name>:latest cmd /S /C "python samples\python\hello_query_device\hello_query_device.py"

Please follow the appropriate steps from the Install Guide for Windows to get GPU access from Windows container.

Linux image:

To enable GPU access, make sure you've built the image with support for GPU and run:

docker run -itu root:root --rm --device /dev/dri:/dev/dri <image_name>:latest

If your host system is Ubuntu 20, follow the Configuration Guide for the Intel® Graphics Compute Runtime for OpenCL™ on Ubuntu* 20.04.

If your host system is Windows with WSL2, use the following command to start the container:

docker run -it --rm --device /dev/dxg --volume /usr/lib/wsl:/usr/lib/wsl <image_name>:latest

Note: You need to meet the following prerequisites to use a GPU device inside WSL2:

  • Windows 10 with 21H2 update or Windows 11
  • Intel iGPU drivers should be installed on the host (version 30.0.100.9684 or newer)

Read more:

To run inference on the VPU, make sure you've built the image with support for VPU and run:

docker run -itu root:root --rm --device-cgroup-rule='c 189:* rmw' -v /dev/bus/usb:/dev/bus/usb <image_name>:latest

To run inference on the HDDL, make sure you've built the image with support for HDDL and setup HDDL driver on host machine, follow the configuration guide for HDDL device:

docker run -itu root:root --rm --device=/dev/ion:/dev/ion -v /var/tmp:/var/tmp <image_name>:latest

Note: If the application runs inference of a network with a big size(>4MB) of input/output, the HDDL plugin will use shared memory.
You should mount /dev/shm as volume in this case:

docker run -itu root:root --rm --device=/dev/ion:/dev/ion -v /var/tmp:/var/tmp -v /dev/shm:/dev/shm <image_name>:latest

Also, in some cases you can encounter a permission error for files in /dev/shm (see hddldaemon.log). The possible cause is uid and gid of the container user are different from uid and gid of the user which had created hddldaemon service on the host.
You can try one of these solutions:

  • create the user in the docker container with uid and gid which are the same with the HDDL daemon's user
  • set container user umask to 000: umask 000
  • start HDDL daemon on the host as root and start container as root docker run -u root (NOT RECOMMENDED)

And to run inference on all hardware targets supported, make sure you've built the image correctly and run:

docker run -itu root:root --rm --device=/dev/ion:/dev/ion -v /var/tmp:/var/tmp --device /dev/dri:/dev/dri --device-cgroup-rule='c 189:* rmw' -v /dev/bus/usb:/dev/bus/usb <image_name>:latest

If you want to try some samples, then run the following commands:

docker run -itu root:root --rm --device=/dev/ion:/dev/ion -v /var/tmp:/var/tmp --device /dev/dri:/dev/dri --device-cgroup-rule='c 189:* rmw' -v /dev/bus/usb:/dev/bus/usb <image_name>:latest
/bin/bash -c "omz_downloader --name googlenet-v1 --precisions FP16 && omz_converter --name googlenet-v1 --precision FP16 && curl -O https://storage.openvinotoolkit.org/data/test_data/images/car_1.bmp && python3 samples/python/hello_classification/hello_classification.py public/googlenet-v1/FP16/googlenet-v1.xml car_1.bmp CPU"
/bin/bash -c "omz_downloader --name googlenet-v1 --precisions FP16 && omz_converter --name googlenet-v1 --precision FP16 && curl -O https://storage.openvinotoolkit.org/data/test_data/images/car_1.bmp && python3 samples/python/hello_classification/hello_classification.py public/googlenet-v1/FP16/googlenet-v1.xml car_1.bmp GPU"
/bin/bash -c "omz_downloader --name googlenet-v1 --precisions FP16 && omz_converter --name googlenet-v1 --precision FP16 && curl -O https://storage.openvinotoolkit.org/data/test_data/images/car_1.bmp && python3 samples/python/hello_classification/hello_classification.py public/googlenet-v1/FP16/googlenet-v1.xml car_1.bmp MYRIAD"
/bin/bash -c "omz_downloader --name googlenet-v1 --precisions FP16 && omz_converter --name googlenet-v1 --precision FP16 && curl -O https://storage.openvinotoolkit.org/data/test_data/images/car_1.bmp && python3 samples/python/hello_classification/hello_classification.py public/googlenet-v1/FP16/googlenet-v1.xml car_1.bmp HDDL"

Troubleshooting

If you see a missing apt package that needs for OpenVINO product in Linux docker image, please install a missing component directly via apt install command and create issue on GitHub* Issues. We will check a missing package to meet Intel(R) security policy. Please see SECURITY for details to follow security guideline.

If you got a proxy issues, please setup proxy settings for Docker Engine. See the Proxy section in Install the DL Workbench from Docker Hub* topic. DockerHub CI uses a system proxy to generate Dockerfile and build a docker image by default.

* Other names and brands may be claimed as the property of others.