fastapi-response-streaming-lmi

FastAPI Response Streaming with Lambda Managed Instances

This example shows how to use Lambda Web Adapter to run a FastAPI application with response streaming on Lambda Managed Instances (LMI).

Lambda Managed Instances allows a single Lambda execution environment to handle multiple concurrent requests, improving throughput and reducing costs for I/O-bound workloads like streaming responses.

Prerequisites

Lambda Managed Instances requires a VPC with:

• At least one subnet (two subnets recommended)
• A security group that allows outbound traffic

If you don't have a VPC configured, you can use the default VPC or create one.

How does it work?

This example combines three Lambda features:

1. Lambda Web Adapter - Runs your FastAPI app on Lambda without code changes
2. Response Streaming - Streams responses back to clients as they're generated
3. Lambda Managed Instances - Handles multiple concurrent requests per execution environment

Key Configuration

LMICapacityProvider:
  Type: AWS::Serverless::CapacityProvider
  Properties:
    CapacityProviderName: !Sub "${AWS::StackName}-capacity-provider"
    VpcConfig:
      SubnetIds: !Ref SubnetIds
      SecurityGroupIds: !Ref SecurityGroupIds
    ScalingConfig:
      MaxVCpuCount: 20
      AverageCPUUtilization: 70.0

FastAPIFunction:
  Type: AWS::Serverless::Function
  Properties:
    CodeUri: app/
    Handler: run.sh
    Runtime: python3.13
    MemorySize: 2048
    Environment:
      Variables:
        AWS_LAMBDA_EXEC_WRAPPER: /opt/bootstrap
        AWS_LWA_INVOKE_MODE: response_stream
        PORT: 8000
    Layers:
      - !Sub arn:aws:lambda:${AWS::Region}:753240598075:layer:LambdaAdapterLayerX86:26
    CapacityProviderConfig:
      Arn: !GetAtt LMICapacityProvider.Arn
      PerExecutionEnvironmentMaxConcurrency: 64
    FunctionUrlConfig:
      AuthType: NONE
      InvokeMode: RESPONSE_STREAM

• AWS::Serverless::CapacityProvider - Creates the LMI capacity provider with VPC configuration
• CapacityProviderConfig.Arn - References the capacity provider
• CapacityProviderConfig.PerExecutionEnvironmentMaxConcurrency: 64 - Up to 64 concurrent requests per instance
• AWS_LWA_INVOKE_MODE: response_stream - Configures Lambda Web Adapter for streaming
• FunctionUrlConfig.InvokeMode: RESPONSE_STREAM - Enables streaming on the Function URL

Build and Deploy

First, get your VPC subnet and security group IDs:

# List subnets in your default VPC
aws ec2 describe-subnets --filters "Name=default-for-az,Values=true" \
  --query 'Subnets[*].[SubnetId,AvailabilityZone]' --output table

# List security groups
aws ec2 describe-security-groups --filters "Name=group-name,Values=default" \
  --query 'SecurityGroups[*].[GroupId,GroupName]' --output table

Build and deploy:

sam build --use-container
sam deploy --guided

During guided deployment, you'll be prompted for:

• SubnetIds - Comma-separated list of subnet IDs (e.g., subnet-abc123,subnet-def456)
• SecurityGroupIds - Comma-separated list of security group IDs (e.g., sg-abc123)

Verify it works

Open the Function URL in a browser. You should see a message stream back character by character, with a unique request ID prefix like [a1b2c3d4] This is streaming from Lambda Managed Instances!.

Test concurrent requests

To verify LMI is working, send multiple concurrent requests:

# Get your function URL
URL=$(aws cloudformation describe-stacks --stack-name fastapi-response-streaming-lmi \
  --query 'Stacks[0].Outputs[?OutputKey==`FastAPIFunctionUrl`].OutputValue' --output text)

# Send 10 concurrent requests
for i in {1..10}; do curl -s "$URL" & done; wait

Each response will have a different request ID, but they may share the same execution environment (visible in CloudWatch logs).

Considerations

When using LMI with streaming:

• VPC: LMI requires VPC configuration. Ensure your subnets have internet access (via NAT Gateway) if your function needs to call external services
• Shared state: FastAPI/Uvicorn handles concurrency natively, but avoid mutable global state
• Memory: With 64 concurrent requests, ensure sufficient memory (2048MB in this example)
• Timeouts: Streaming responses can run up to 15 minutes with Function URLs
• Scaling: MaxVCpuCount controls the maximum vCPUs the capacity provider can provision across all instances