AWS offers: the following container systems:
Docker images are stored in:
. docker hub https://hub.docker.com
is a public repository for basic images
Amazon ECR – Elastic Container Registry
– you can also keep your private images here
you can also use the public repository
the basic docker file contains:
this builds the docker image, you then push/pull to/from the repository you are using.
you run the image and it creates a live docker container – important to know the difference!
ECS: AWS Elastic Container Service – AWSs own container platform
EKS: Elastic kubernetes Service – AWSs managed kubernetes container service
AWS Fargate: a serverless container platform service, works with ECS and EKS
ECR: a repository for storing container images – Elastic Container Repository
important for exam! with ECS
ECS – EC2 Launch Type
when you launch a container you are launching an ECS task on an ECS cluster
you must provision and maintain the infra on the EC2 instances
but each EC2 instance must run the ECS Agent to register and operate as an ECS Cluster
AWS then takes care of precisely on which instance your containers are launched! you don’t specify it.
Fargate Launch Type for ECS
we do not provision any infra ie no ec2 instances needed
it is serverless
you just create your container task definitions
AWS runs the ECS Tasks for you based on your CPU/RAM requirement
to scale in Fargate you simply increase the number of tasks, not instances
way easier to manage than ec2 launch type
ECS iam roles for ECS
this is where you are using ec2 instances
you have your ec2 instance profile for this, it is used by the ecs agent,
makes the api calls to ecs service
sends container logs to cloudwatch logs
pull docker image from ecr
the ecs task roles:
this applies to ecs and fargate launch types
you create a role for each task
you use different roles for different ecs services that you run
these you define in your task definition for your ecs service
ECS Load Balancer Integrations
Do NOT use Elastic Load Balancer for ECS – you should not use the old ELB as this has only minimal container features, and it has NO Fargate support
ALB – supported, ok for most use cases
NLB – recommended only if you need high throughput/performance or to pair with an AWS Private Link
ECS Data Volumes (EFS) – these allow for data persistence
you mount an EFS mounted onto the ECS tasks
this works for both EC2 and Fargate launch types
this means tasks running in any AZ can share the same data
Fargate + EFS = serverless
use case: when you need persistent multi-az shared storage for your containers
important – exam!
remember: S3 CANNOT be mounted as a file system!
only EFS can do this.
ECS Service Auto Scaling
there are a number of possibilities for this.
uses AWS Application Auto Scaling to automatically increase or decrease the desired number of ECS tasks
it measures following metrics:
ECS Service Average CPU Utilization
ECS Service Average Memory Utilization
ALB Request Count per Target – from ALB
you can also use
Target Tracking – a scale based on target value for a specified CloudWatch metric
Step Scaling – based on a specified CloudWatch Alarm
Scheduled Scaling which is based on a specified date and time – this is predictable
remember that scaling at the ecs service auto scaling task level is NOT the same as scaling at the ec2 instance auto scaling level
Fargate Auto Scaling – this is much easier to set up as it is serverless
So, for Auto Scaling for the EC2 Instance launch type:
this works by adding underlying EC2 instance according to demand
We can use
Auto Scaling Group scaling
this scales asg acc to cpu util
then it adds ec2 instances over time
or use the new more advanced system called
ECS Cluster Capacity Provider
– this is a much better option by far
this is used to auto-provision and scale the infra for your ecs tasks
it is paired with an ASG
adds EC2 instances when needed
ECS Cluster Capacity Provider Auto-Scaling is the much better option.
ECS Scaling – Service CPU usage example
cloudwatch metric monitors cpu usage
and triggers a cloudwatch alarm in turn, this then scales via the auto scaling group for the cluster adding ec2 instances as required.
ECS Rolling Updates
when updating from v1 to v2, we can speficy who many tasks can be started and stopped, and which order
we can set a min % of tasks healthy
and a maximum %
default 100% min and 200% max
the max tells you how many more you can create…
the system is allowed to terminate up to the min %.
eg min 50% max 100%
we start with 4 tasks
this means we can terminate half the tasks ie 2 in this case at one time…
we can then perform the update on those instances in turn
ECS Tasks can also be invoked by linking to Event Bridge or to SQS message queuing
Amazon ECR Elastic Container Registry
stores and manages your docker images on AWS
there is a private and public repo – the public is ECCR public gallery at gallery.ecr.aws
fully integrated with ECS
access controlled by iam policy
it supports image vulnerability scanning, versioning, image tags and image lifestyle.
need to be aware of this repo for the exam!
Elastic Kubernetes Service
enables use of Kubernetes on AWS as alternative to ECS
open source, whereas ECS is AWS proprietary
similar to ECS but different API
EKS supports EC2 and Fargate
use case: if already using kubernetes
or wants to migrate to it, can be actually used on any cloud or with no cloud, not just aws
EKS uses “EKS Pods” in place of “ECS Tasks” – otherwise the same thing – remember this for the exam.
AWS App Runner
priced per cpu and gbyte.
This is a fully managed container application service which is serverless and managed AWS service for easy deployment of web apps and APIs
start with source code or container image
automatically builds and deploys the app for you
autoscaling, HA, Lb, encryption and connects to your vpc
also connects to DBs and message queues