0%
November 6, 2024

Lambda Function Running Python Docker Image

aws

docker

lambda

python

Contents

Repository

Summary

  • Dependencies are installed in docker image, size is not an issue any more.

  • We can build a Flask app as usual, and then create a wsgi_handler.py that wrap the Flask's app object as a lambda handler, after that, we are done!

  • I have intensionally installed huge packages which results in a docker image of size 725MB, and such an image have cold-start for 3s only!

serverless.yml

service: llm-trial

provider:
  name: aws
  region: ap-southeast-2
  stage: dev
  timeout: 900
  iam:
    role:
      name: ${self:service}-${self:provider.stage}-role
  environment:
    AZURE_OPENAI_API_KEY:
    AZURE_OPENAI_ENDPOINT:
    PROXY_CURL:
  ecr:
    images:
      llm-trial:
        path: ./

functions:
  api:
    image:
      name: llm-trial
    timeout: 900
    events:
      - http: ANY /
      - http: ANY /{proxy+}

custom:
  pythonRequirements:
    dockerizePip: true
    slim: true
    noDeploy:
      - pip
      - autopep8
      - debugpy

package:
  exclude:
    - node_modules/**
    - .dockerignore
    - Dockerfile
    - docker-compose*

wsgi_handler.py

from aws_lambda_wsgi import response
from src.main import app  # Import your Flask app

def lambda_handler(event, context):
    return response(app, event, context)

Dockerfile

FROM public.ecr.aws/lambda/python:3.10
WORKDIR /var/task
COPY . .
ENV PYTHONPATH=/var/task

RUN yum install -y \
    gcc \
    gcc-c++ 
RUN pip install --no-cache-dir -r requirements.txt
RUN yum remove -y gcc gcc-c++ && \
    yum clean all

CMD ["wsgi_handler.lambda_handler"]