Platform For AI:Deploy model services using custom images

Step 1: Prepare the environment

Before you begin, ensure you have:

• Virtual Private Cloud (VPC): Your EAS service, ECS instance, and ACR registry must use the same VPC for stable, secure image transfers.

• Container Registry (ACR): An ACR Enterprise Edition instance with a namespace and image repository. See Use an Enterprise Edition instance to push and pull images.

• Development environment: An ECS instance with this configuration:

  • Instance Type: ecs.u1-c1m2.large

  • Image: Alibaba Cloud Linux 3.2104 LTS 64-bit

  • Extension: Docker Community Edition

  • Network: Add your VPC to the ACR Enterprise Edition instance access control list. See Configure access control for an ACR instance over a VPC.

  Note

  You can also use a local machine or DSW environment.

  • Local development: Install Docker to build images locally.

  • DSW development: In the Actions column, click Create Image to build and save the image to ACR. During deployment, select Custom Image from the dropdown. See Create a DSW instance image.

Step 2: Prepare application files

Create a project folder named my-app with these three files:

1. requirements.txt (Application dependencies)

   flask
   gunicorn

2. app.py (Web application code)

   from flask import Flask
   
   app = Flask(__name__)
   
   @app.route('/hello/model')
   def hello_world():
       # Integrate your model inference or other business logic here.
       return 'Hello World from Gunicorn!'
   
   # Note: app.run() is not needed. Gunicorn will start the application.

3. Dockerfile (Image build instructions)

   # 1. Use an official lightweight Python image as the base
   FROM python:3.9-slim
   
   # 2. Set the working directory
   WORKDIR /app
   
   # 3. Copy the requirements file and install dependencies, leveraging Docker's cache to speed up subsequent builds
   COPY requirements.txt .
   RUN pip install --no-cache-dir -r requirements.txt -i https://mirrors.aliyun.com/pypi/simple/
   
   # 4. Copy the application code
   COPY app.py .
   
   # 5. Expose the service port
   EXPOSE 8000
   
   # 6. Define the default container start command (which you can override with the "Command" in the EAS console)
   CMD ["gunicorn", "-w", "4", "-b", "0.0.0.0:8000", "app:app"]

Step 3: Build and push the image

1. Navigate to your project folder (my-app).

2. Log in to ACR. Replace placeholders with your information.

   # Example: docker login --username=your_user my-registry.cn-hangzhou.aliyuncs.com
   docker login --username=<username> <ACR_repository_domain_name>

3. Build and tag the image.

   # Example: docker build -t my-registry.cn-hangzhou.aliyuncs.com/my-namespace/flask-app:v1 .
   docker build -t <ACR_repository_domain_name>/<namespace>/<image_repository_name>:<version_number> .

4. Push the image to ACR.

   # Example: docker push registry.cn-hangzhou.aliyuncs.com/my-namespace/flask-app:v1
   docker push <ACR_repository_domain_name>/<namespace>/<image_repository_name>:<version_number>

Step 4: Deploy the service

1. Log on to the PAI console. Select a region on the top of the page. Then, select the desired workspace and click Elastic Algorithm Service (EAS).

2. Click Deploy Service and select Custom Model Deployment > Custom Deployment.

3. Configure the following key parameters:

   • Deployment Method: Select Image-based Deployment.

   • Image Configuration: Select Image Address and enter the full image address.

     This example uses ACR in the same Alibaba Cloud account, so no authentication is needed.

   • Command to Run: gunicorn -w 4 -b 0.0.0.0:8000 app:app

     Note: Commands in the EAS console override the CMD in your Dockerfile. Specify the command here for easier debugging.

   • Port Number 8000.

   • Deployment: Select a CPU resource like ecs.c6.large from Public Resources.

   • VPC: Select a VPC, vSwitch, and security group.

     Ensure your VPC is in the ACR access control list. Missing VPC configuration causes ImagePullBackOff errors.

4. Click Deploy. Deployment succeeds when the service status changes to Running.

Q: Why does my service fail with an ImagePullBackOff error?

The ImagePullBackOff error means EAS cannot pull your container image. This typically indicates one of these issues:

1. Network connectivity issues

   Your EAS service's VPC must have network access to the image registry.

   • Action: Verify your EAS service's VPC is in your ACR instance's access control list.

   • For cross-region deployments: Connect VPCs using Cloud Enterprise Network (CEN). See Access an ACR Enterprise Edition instance across regions or from an IDC.

   • For public internet pulls: Ensure your VPC has a NAT Gateway and public network access is enabled for your ACR repository.

2. Authentication failure

   Private registries require authentication credentials (ACR instances in the same account don't require credentials).

   • Action: Verify the username and password in Image Configuration are correct with pull permissions.

3. Image not found

   The image path or tag is incorrect, or the image doesn't exist in the registry.

   • Action: Verify the image address, namespace, and tag match what you pushed to the registry.

4. Insufficient system disk space

   The node's local storage is full and cannot store the image.

   • Action: Expand the system disk for your EAS service resource group.