This is a sample application to demonstrate Datadog serverless observability features. It is based out of the Image Processing AWS Serverless Workshop but with the following differences:

• Python, instead of Node.js
• It uses the Serverless Framework for deployment.
• It uses HTTP calls and SNS messages for coordination, instead of Step Functions

• Serverless Framework version 3.3 or newer. Follow their instructions to install it.
• A Rekognition Collection:

The application requires an AWS Rekognition Collection. You can create one with the aws-cli tool:

aws rekognition create-collection --collection-id <your-rekognition-collection-id>

IMPORTANT: The only thing that you need to manually create in your AWS account to deploy the application is the Rekognition Collection, the rest will be handled by the CloudFormation template in the serverless.yml file.

Once you have the Serverless framework installed, you can deploy the application following these commands:

git clone https://github.com/arapulido/face-recognition-serverless-app
cd face-recognition-serverless-app
npm install
serverless deploy --region=<your_aws_region> --param="ddenv=facerecognition"  --param="ddApiKey=<datadog_api_key>" --param="owner=<your_name>" --param="rekognition-collection-id=<your_rekognition_collection_id>"

Once it has been deployed, from the CLI output, note the HTTP endpoint for the face-search function and redeploy the application with it as a new parameter:

serverless deploy --param="ddenv=facerecognition"  --param="ddApiKey=<datadog_api_key>" --param="owner=<your_name>" --param="rekognition-collection-id=<your_rekognition_collection_id>" --param="face-search-endpoint=<face_search_function_endpoint>"

Once deployed, visit the "CloudFormation" section in the AWS Console to check the application and the resources that were created.

There are two ways to use this application, directly using the lambda functions or through a web application.

1. Upload test images with or without faces to your newly created S3 bucket (you can find the correct S3 bucket in the CloudFormation Resources section for your application)
2. Call the detect-faces endpoint with one of the images as parameter:

curl -X POST <detect-faces-endpoint> --data '{"srcBucket": "<s3-bucket-id>", "name":"<image-filename>", "userId":"<random-user-id>"}'

The workflow of functions will do the following:

• face-detection will check if the picture contains 1 face. If it has more than 1 face or no faces, will return an error.
• face-search will check if the same face is already part of the Rekognition collection, and will err if that's the case (duplicated face).
• face-index will index the face into the Rekognition collection (to find duplicates later).
• persist-metadata will store some metadata into a DynamoDB table.

There is a sample web application that will upload a picture to S3 and will called the first function of the pipeline directly, for a more end-to-end user experience. Follow the instructions to deploy and use that application in its README.md file.