CloudFormation部署的AWS Lambda中psycopg2模块缺失问题求解

Problem Description #

I've written an AWS Lambda function using a CloudFormation template to manage PostgreSQL roles/users. The code works locally, deploys successfully to Lambda, but throws an import error at runtime.

CloudFormation Lambda Snippet #

DbAdminLambda:
  Type: AWS::Lambda::Function
  Properties:
    Code:
      ZipFile: |
        import psycopg2
        import json
        def lambda_handler(event, context):
          print(json.dumps(event))
          # ... database logic ...
          cfnresponse.send(event, context, cfnresponse.SUCCESS, "UserCreatedOk", "CustomResourcePhysicalID")
    MemorySize: 128
    Description: create postgres roles / users
    Handler: index.lambda_handler

Runtime Error #

{ "errorMessage": "Unable to import module 'index': No module named 'psycopg2'", "errorType": "Runtime.ImportModuleError" }

I suspect this is because Lambda doesn't have the psycopg2 binary files. How can I get this code working? Do I have to compile the binaries and upload them to S3?

Solutions & Answers #

Great question—this is a common gotcha with Python packages that have binary dependencies like psycopg2 in Lambda. The core issue is that Lambda's default Python runtime doesn't include PostgreSQL client libraries or the pre-compiled psycopg2 binary tailored to its environment. Here are your most practical fixes:

1. Use a Pre-Built psycopg2 Lambda Layer (Easiest Option) #

Lambda Layers let you share dependencies across functions without bundling them into your code, which keeps your deployment package small and clean.

• Find a trusted pre-compiled psycopg2 layer that matches your Lambda's Python version and region. Community-maintained layers are widely available for common runtime versions.
• Update your CloudFormation template to attach the layer to your function:

  DbAdminLambda:
    Type: AWS::Lambda::Function
    Properties:
      Code:
        ZipFile: |
          # Your existing code here
      MemorySize: 128
      Description: create postgres roles / users
      Handler: index.lambda_handler
      Layers:
        - arn:aws:lambda:<your-region>:<layer-account-id>:layer:<psycopg2-layer-name>:<version>
  

  Replace the ARN placeholders with the actual values for your environment.

2. Bundle psycopg2-binary with Your Deployment Package #

If you prefer not to use layers, you can package a Lambda-compatible version of psycopg2 directly into your zip file:

• Use a Linux environment (or Docker to mimic Lambda's Amazon Linux 2 runtime) to install the correct binary wheel:

  mkdir -p package
  pip install psycopg2-binary -t package/ --platform manylinux2014_x86_64 --only-binary=:all:
  

• Zip the package contents along with your Lambda code:

  cd package
  zip -r ../lambda-deployment.zip .
  cd ..
  zip lambda-deployment.zip index.py
  

• Upload this zip to an S3 bucket, then update your CloudFormation template to reference the S3 location instead of inline ZipFile:

  DbAdminLambda:
    Type: AWS::Lambda::Function
    Properties:
      Code:
        S3Bucket: <your-s3-bucket-name>
        S3Key: lambda-deployment.zip
      # Keep other properties the same
  

3. Use a Lambda Container Image (Most Flexible) #

If you need full control over the runtime environment (e.g., custom PostgreSQL versions), build a container image that includes psycopg2:

• Create a Dockerfile using Lambda's official Python base image:

  FROM public.ecr.aws/lambda/python:3.10
  COPY requirements.txt .
  RUN pip install -r requirements.txt --target "${LAMBDA_TASK_ROOT}"
  COPY index.py ${LAMBDA_TASK_ROOT}
  CMD ["index.lambda_handler"]
  

  (Make sure requirements.txt includes psycopg2-binary or psycopg2)
• Build the image, push it to Amazon ECR, then update your CloudFormation template to use the image:

  DbAdminLambda:
    Type: AWS::Lambda::Function
    Properties:
      Code:
        ImageUri: <your-ecr-repo-uri>:<tag>
      # Keep other properties the same
  

Do I Have to Compile Binaries and Upload to S3? #

Not necessarily! Using a pre-built layer or psycopg2-binary (which provides pre-compiled wheels) avoids manual compilation. That said, if you need a custom build (e.g., for a specific PostgreSQL version or architecture), you can compile it on a Lambda-compatible environment (like an EC2 instance running Amazon Linux 2) and package it into a layer or zip file, then upload to S3 as part of your deployment.

内容的提问来源于stack exchange，提问作者mdarwin