Serverless with Python Look ma, no servers! Alejandro Guirao @lekum github.com/lekum lekum.org 

What is Serverless? ● No servers? ● FaaS (functions as a service) ● Elasticity 

Hosted solutions 

The pioneer: AWS Lambda S3 bucket DynamoDB table Kinesis Streams SNS Topic SES e-mail Cognito CloudFormation stack CloudWatch logs & events CodeCommit Config Alexa skill Lex API Gateway AWS Lambda Event sources 

AWS Lambda limits 

AWS Lambda: Manual deploy def my_handler(event, context): message = 'Hello {} {}!'.format(event['first_name'], event['last_name']) return { 'message' : message } 

Context object get_remaining_time_in_millis() function_name function_version invoked_function_arn memory_limit_in_mb aws_request_id log_group_name log_stream_name identity client_context 

AWS Lambda: Manual deploy aws lambda create-function \ --region eu-west-1 \ --function-name HelloPython \ --zip-file fileb://deployment-package.zip \ --role role-arn \ --handler hello_python.my_handler \ --runtime python3.6 \ --timeout 15 \ --memory-size 512 

AWS Lambda: Manual deploy $ aws lambda invoke \ --function-name \ --payload '{"first_name":"Margaret", "last_name":"Hamilton"}' out.json $ cat out.json {"message": "Hello Margaret Hamilton!"} 

Testing AWS Lambda ● Decouple business logic ● Test with sample events ● Use lambci/lambda docker run -v $PWD:/var/task lambci/lambda:python3.6 my_module.my_handler 

AWS Lambda: chalice ● Microframework for API Gateway + Lambda functions ● Flask-like ● A CLI that creates and configures the AWS resources needed (Lambda, API Gateway, IAM…) and makes and uploads the deployment package ● Officially supported by AWS 

AWS Lambda: chalice $ pip install chalice $ chalice new-project helloworld && cd helloworld $ cat app.py from chalice import Chalice app = Chalice(app_name='helloworld') @app.route('/', methods=['POST']) def greeting(): body = app.current_request.json_body message = 'Hello {} {}!'.format(body['first_name'], body['last_name']) return { "message" : message } $ chalice deploy --dev 

AWS Lambda: zappa ● Zappa is a tool for deploying your Python WSGI apps to AWS ● Manages the lifecycle of the AWS resources and your application versioning (via S3) ● Advanced features as remote invocations or scheduling configuration and async task execution $ zappa init $ zappa deploy 

Use case: CARTO + Matplotlib + Amazon AWS 

Other hosted solutions ● Azure Functions ● IBM BlueMix Cloud Functions (based on OpenWhisk) ● Iron.io (based on Docker containers) ● Google cloud functions does NOT yet support Python 

Do-It-Yourself 

DIY: Iron functions ● Based on Docker containers: stdin and stdout as communication mechanism ● Pulls images from a Docker registry (Dockerhub) ● Configuration based on a func.yml file ● Creation of a function: $ fn build && fn push $ fn apps create helloapp && fn routes create helloapp /hello 

DIY: Iron functions import os import sys import json if not os.isatty(sys.stdin.fileno()): obj = json.loads(sys.stdin.read()) first_name = obj.get("first_name") last_name = obj.get("last_name") message = 'Hello {} {}!'.format(first_name, last_name) print(json.dumps({"message": message})) 

DIY: Apache OpenWhisk Complex Architecture: ● Event-based programming with triggers, rules and actions $ wsk action create hello hello.py --kind python:3 $ wsk action invoke hello --param first_name Margaret --param last_name Hamilton --result 

DIY: OpenFaas ● Works on top of Docker Swarm or Kubernetes ● Autoscales to meet the concurrency needed ● Configuration via .yaml or UI provider: name: faas gateway: http://127.0.0.1:8080 functions: hello: lang: Dockerfile handler: ./handler image: alexellis2/faas-hello-python3 

DIY: fission ● Runs on Kubernetes (locally on Minikube) ● 100 msec cold-start latencies $ fission function create --name hello --env python3 --code hello.py $ fission route add --function hello --url /hello import json from flask import request def main(): data = json.loads(request.get_data().decode()) message = 'Hello {} {}!'.format(data['first_name'], data['last_name']) return json.dumps({"message": message} 

Happy hacking! Alejandro Guirao @lekum lekum.org lekum/serverless-with-python