docs/en/Intel-CCAI-Development-Manual/Integrate-new-service.md

3.7 KiB
Raw Permalink Blame History

title description published date tags editor dateCreated
08. Integrate New AI Services How to integrate new AI Services with CCAI Framework true 2022-05-17T07:17:20.084Z ccai markdown 2022-03-11T03:18:18.401Z

Once you have new services, to make it work to be able to accept requests from outside of the CCAI container and give back the result of one specific AI task, you will have to deploy those services in the CCAI container.

Where to put those services file to

Please extract the CCAI release tar file, saying ccaisf_release_xx-xxx.tar.gz and copy your files and directories organized in a runtime hierarchy to the folder docker/app_rootfs.

A fast CGI service example as:

ccaisf_release_xx-xxx/docker/app_rootfs
├── etc
│   ├── lighttpd
│   │   ├── conf-available
│   │   │   ├── 16-classification.conf
│   │   ├── conf-enabled
│   │   │   ├── 16-classification.conf -> ../conf-available/16-classification.conf
├── opt
│   ├── fcgi
│   │   └── cgi-bin
│   │       ├── fcgi_classfication

A gRPC service example as:

ccaisf_release_xx-xxx/docker/app_rootfs
├── etc
│   └── sv
│       ├── grpc_inference_service_speech_1
│       │   ├── run
│       │   └── supervise -> /tmp/grpc_inference_service_speech_1/supervise
│   ├── runit
│   │   └── runsvdir
│   │       └── default
│   │           ├── grpc_inference_service_speech_1 -> /etc/sv/grpc_inference_service_speech_1
└── usr
    └── sbin
        ├── grpc_inference_service

Where to put related Neural Network Models file

Models will be installed to the folder, /opt/intel/service_runtime/models on host, CCAI will map the folder to containers /opt/fcgi/cgi-bin/models, so you can write your debian package configuration files and build your deb package to install your models, otherwise you can also put your models to CCAI release folder to use CCAIs helper script to build your deb package:

  1. Put models to ccaisf_release_xx-xxx/package/models.
  2. Modify ccaisf_release_xx-xxx/package/models/debian/control to add your package.
  3. Add a service-runtime-models-xxx.install file to ccaisf_release_xx-xxx/package/models/debian to install your models.

How to enable services via API gateway

As we had described in chapter 6.3, for fast CGI services, you need to add/change conf files and put those conf files under specific folders so that API gateway will recognize your services and launch them according to the configuration file description.

For gRPC services, if it is a brand new services, you need to do the following steps to enable the service:

  1. Create a folder under docker/app_rootfs/etc/sv/, example:
$> mkdir docker/app_rootfs/etc/sv/your_grpc_service
  1. Write a script named run to launch your service, and put the script to docker/app_rootfs/etc/sv/your_grpc_service, example:
$>cat > docker/app_rootfs/etc/sv/your_grpc_service/run < EOF
!/bin/bash
exec /usr/sbin/your_service
EOF
  1. Create a link to manage the service at runtime.
$>ln -sf /tmp/your_grpc_service/supervise docker/app_rootfs/etc/sv/your_grpc_service/
  1. Create a link to enable your gRPC service
$>ln -sf /etc/sv/your_grpc_service docker/app_rootfs/etc/runit/runsvdir/default/

How to generate new container image

Please refer to Chapter 5, execute release_build.sh to generate a new container image or deb package for models.

$> cd ccaisf_release_xx-xxx
$> ./release_build.sh image
$> ./release_build.sh models_package