One step ECS deployment

Why does Docker suck?

• Difficult to inspect images (layers can break you)
• Trivial things like port changes can wreak havoc
• So many issues I have forgotten them as a self defence mechanism against insanity
• Allows developers to bring in crazy ass dependencies easily

Why might you need Docker?

• Allows developers to bring in crazy ass dependencies easily
• Legacy app that doesn't expose http interface cleanly
• Legacy app that doesn't have a restful interface
• Legacy app that takes forever to start up
• Legacy app that has too many moving parts aka state
• Keep sockets open use case

So how do you deploy Docker images on AWS?

• DIY on EC2 (not managed by AWS)
• Amazon Elastic Container Service (Amazon ECS)
• BETA Amazon Elastic Container Service for Kubernetes (Amazon EKS)
• BETA Amazon Fargate

AWS ECS

• cluster links to your EC2 instances
• service links to your load balancer
• task links to your `docker run` parameters

Building this out

• Orchestrating this stuff with Cloudformation is a total total pain
• AWS console is ok for the cluster and service part
• Task stuff is best done with ecs-cli

ecs-cli

version: '2'
services:
    bugzilla:
      image: uneet/bugzilla-customisation
      restart: on-failure
      mem_reservation: 1g
      mem_limit: 3g # assuming a t2.medium host with 4GB of RAM
      ports:
        - 80:80
      environment:
            SES_SMTP_USERNAME: ${SES_SMTP_USERNAME}
            SES_SMTP_PASSWORD: ${SES_SMTP_PASSWORD}
            SES_VERIFIED_SENDER: ${SES_VERIFIED_SENDER}
            MYSQL_HOST: ${MYSQL_HOST}
            MYSQL_PORT: ${MYSQL_PORT}
            MYSQL_DATABASE: ${MYSQL_DATABASE}
            MYSQL_USER: ${MYSQL_USER}
            MYSQL_PASSWORD: ${MYSQL_PASSWORD}
            PARAMS_URL: ${PARAMS_URL}
            COMMIT: ${COMMIT}
      logging:
            driver: awslogs
            options:
              awslogs-region: ap-southeast-1
              awslogs-group: bugzilla

my own one step deploy.sh

deploy.sh

• maintains a notion of staging, demo, production environments
• checks your credentials are setup
• substitutes the environment variables
• creates a new task using ecs-cli

Travis CI

• One step build/deployment
• Build logs

Notes

• Rolling back to versioned image names
• CloudWatch logs are glorious
• ecs-checks for IAC
• Read my blog for details
• Canary metrics are HTTPCode ELB 5xx Count AND TargetResponseTime

Thank you