Whilst "Dockerizing" a really simple PHP project to send feedback email, I had the unfortunate experience of tustling with the AWS PHP SDK in

Problem 1: Difficult to distinguish between SDK v2 and v3

The only way I've figured this out, is that examples with SesClient::factory are v2. Otherwise I would expect use Aws\Ses\SesClient; to perhaps indicate the version.

So this causes a lot of pain because I'm lazy and I'm looking for examples to get this working quickly. Unfortunately most of Google results are for v2 and don't work!

Problem 2: The API is really hard to use

Ignoring the fact that the SDK weighs in at 8.2MB, I guess AWS must have some automatic mapping to PHP and it really makes it HORRIBLE to use.

So what most people use in some API on top of the SDK or some other cut down third party library. Both options are not great.

It almost brings me to years how we have come a simple 1 line invocation to the mail() 40 lines of code of sesmail.php.

Sidenote: ssmtp is not an option since it's sadly synchronous and slow.

Problem 3: The AWS PHP SDK documentation sucks

It's not crystal clear what required and what's optional in the AWS SDK documentation for sending an email. Where is a minimalistic example? Where is a slightly more realistic sample?

Problem 4:

With $result = $SesClient->sendEmail([ you inline an object. How do you choose not to have ReplyToAddresses because the SDK moans if it's empty!



AWS ECS Workflow

Following up after ECS questions

docker-compose.yml example

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

ECS instance resources

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

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

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.
  • 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.


ecs-cli down will not work until you remove the ELB 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
docker push

Typically a 100mb upload

Then bring down the service and up the service like so:

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.


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'


  1. ecs-cli up --capability-iam --keypair spuul_sysadmin
  2. ecs-cli compose service up to create service
  3. Find VPC name
  4. Create ELB
  5. Create service from task and associate ELB
  6. Update Route 53 Failover record with ELB name in case things really go badly

Powered by Vanilla PHP feedback form