docker-compose.yml example

version: '2'
services:
  web:
	image: kaihendry/count
	ports:
	 - "80:8080"
	mem_limit: 1043333120

Mem limit was figured by taking Available Memory, t2.micro is 995 and getting bytes like so:

~$ echo $((995 << 20))
1043333120

Assuming you have the docker-compose.yml ready and ecs-cli configured, start up the cluster of a EC2 instance running the service like so:

ecs-cli up --capability-iam --keypair example_sysadmin
ecs-cli compose service ps

Setting up the load balancer is manual

Create the service by attaching the load balancer to the task

export CLUSTER_NAME=default

aws --region ap-southeast-1 ecs create-service --service-name "ecscompose-service-count" \
	--cluster "$CLUSTER_NAME" \
	--task-definition "ecscompose-count" \
	--load-balancers "loadBalancerName=ecs-count,containerName=web,containerPort=9000" \
	--desired-count 1 --deployment-configuration "maximumPercent=100,minimumHealthyPercent=50" --role ecsServiceRole

Make it’s in the VPC of the created ECS (look at tags)
Make sure the security groups are permissive
Make sure the health check is /, not /index.html
Setup Route 53 to point to ELB (e.g. ecs.example.com)
Website requires SSL
You do not need to setup ELB in the service, in fact it’s a lot easier if you don’t since it gets confused by port re-mappings.

Caveats

ecs-cli down will not work until you remove the ELB https://console.aws.amazon.com/support/home?region=us-west-2#/case/?displayId=1820472381&language=en. So you need to terminate the instance manually if you want to save some money.

ecs-cli up by default will create a security group that you can’t ssh into.

Developer workflow

aws ecr get-login --profile example --region us-west-2
docker build -t website .
docker tag website:latest 111111111.dkr.ecr.us-west-2.amazonaws.com/website:latest
docker push 111111111.dkr.ecr.us-west-2.amazonaws.com/website:latest

Typically a 100mb upload

Then bring down the service and up the service like so: http://s.natalian.org/2016-07-27/ecs.txt

Bringing down a service

ecs-cli compose service down

Will bring down the particular service. Note that it’s conceivable that many services run on an instance.

You can kill the entire cluster like ecs-cli down --force, but I don’t recommend that because the VPC will change when you re-establish it, so your previously manually set up load balancer will not work.

Current workaround is to manually terminal the EC2 instance from the Web console if I am not requiring it.

DNS failover

Notice that “Evaluate Target Health” is enough for the Failover rule to know the load balancer is out of action. You do not need a health check!

We fail over to Cloudfront since we require SSL to work.

Upgrade flow

Upgrading the container image

Running two instances with the container running on each and then doing a ecs-cli compose service up should upgrade the service I’m told, but I doubt it works. Some careful orchestration with Health limits.

Upgrading the EC2 instance image

Upgrading an instance basically means killing it and starting it again. I guess that would entail a scale event to launch a new instance. ssh to it to ensure it’s uptodate (security group permitting). And then decommission the old one by removing it from the load balancer.

Issues

No way to terminate last EC2 Instance in cluster:

$ ecs-cli scale --size 0 --capability-iam
INFO[0002] Waiting for your cluster resources to be updated
INFO[0003] Cloudformation stack status                   stackStatus=UPDATE_IN_PROGRESS
ERRO[0035] Failure event                                 reason= resourceType=AWS::CloudFormation::Stack
ERRO[0035] Error executing 'scale': Cloudformation failure waiting for 'UPDATE_COMPLETE'. State is 'UPDATE_ROLLBACK_COMPLETE'

Steps

ecs-cli up --capability-iam --keypair spuul_sysadmin
ecs-cli compose service up to create service
Find VPC name http://s.natalian.org/2016-08-02/1470124829_2558x1404.png

Create ELB http://s.natalian.org/2016-08-02/1470125741_2558x1404.png

Create service from task and associate ELB
Update Route 53 Failover record with ELB name in case things really go badly