Building Blocks of Amazon ECS

So, what’s Amazon Elastic Container Service (ECS)? ECS is a managed service for running containers on AWS, designed to make it easy to run applications in the cloud without worrying about configuring the environment for your code to run in. Using ECS, you can easily deploy containers to host a simple website or run complex distributed microservices using thousands of containers.

Getting started with ECS isn’t too difficult. To fully understand how it works and how you can use it, it helps to understand the basic building blocks of ECS and how they fit together!

Amazon EC2 building blocks

We currently provide two ways to run containers: EC2 and Fargate. With Fargate, the Amazon EC2 instances are abstracted away and managed for you. Instead of worrying about ECS container instances, you can just worry about tasks. In this post, the infrastructure components used by ECS that are handled by Fargate are marked with a *.

Instance*

EC2 instances are good ol’ virtual machines (VMs). And yes, don’t worry, you can connect to them (via SSH). Because customers have varying needs in memory, storage, and computing power, many different instance types are offered. Just want to run a small application or try a free trial? Try t2.micro. Want to run memory-optimized workloads? R3 and X1 instances are a couple options. There are many more instance types as well, which cater to various use cases.

AMI*

Sorry if you wanted to immediately march forward, but before you create your instance, you need to choose an AMI. An AMI stands for Amazon Machine Image. What does that mean? Basically, an AMI provides the information required to launch an instance: root volume, launch permissions, and volume-attachment specifications. You can find and choose a Linux or Windows AMI provided by AWS, the user community, the AWS Marketplace (for example, the Amazon ECS-Optimized AMI), or you can create your own.

Region

AWS is divided into regions that are geographic areas around the world (for now it’s just Earth, but maybe someday…). These regions have semi-evocative names such as us-east-1 (N. Virginia), us-west-2 (Oregon), eu-central-1 (Frankfurt), ap-northeast-1 (Tokyo), etc.

Each region is designed to be completely isolated from the others, and consists of multiple, distinct data centers. This creates a “blast radius” for failure so that even if an entire region goes down, the others aren’t affected. Like many AWS services, to start using ECS, you first need to decide the region in which to operate. Typically, this is the region nearest to you or your users.

Availability Zone

AWS regions are subdivided into Availability Zones. A region has at minimum two zones, and up to a handful. Zones are physically isolated from each other, spanning one or more different data centers, but are connected through low-latency, fiber-optic networking, and share some common facilities. EC2 is designed so that the most common failures only affect a single zone to prevent region-wide outages. This means you can achieve high availability in a region by spanning your services across multiple zones and distributing across hosts.

Amazon ECS building blocks

Container

Well, without containers, ECS wouldn’t exist!

Are containers virtual machines?
Nope! Virtual machines virtualize the hardware (benefits), while containers virtualize the operating system (even more benefits!). If you look inside a container, you would see that it is made by processes running on the host, and tied together by kernel constructs like namespaces, cgroups, etc. But you don’t need to bother about that level of detail, at least not in this post!

Why containers?
Containers give you the ability to build, ship, and run your code anywhere!

Before the cloud, you needed to self-host and therefore had to buy machines in addition to setting up and configuring the operating system (OS), and running your code. In the cloud, with virtualization, you can just skip to setting up the OS and running your code. Containers make the process even easier—you can just run your code.

Additionally, all of the dependencies travel in a package with the code, which is called an image. This allows containers to be deployed on any host machine. From the outside, it looks like a host is just holding a bunch of containers. They all look the same, in the sense that they are generic enough to be deployed on any host.

With ECS, you can easily run your containerized code and applications across a managed cluster of EC2 instances.

Are containers a fairly new technology?
The concept of containerization is not new. Its origins date back to 1979 with the creation of chroot. However, it wasn’t until the early 2000s that containers became a major technology. The most significant milestone to date was the release of Docker in 2013, which led to the popularization and widespread adoption of containers.

What does ECS use?
While other container technologies exist (LXC, rkt, etc.), because of its massive adoption and use by our customers, ECS was designed first to work natively with Docker containers.

Container instance*

Yep, you are back to instances. An instance is just slightly more complex in the ECS realm though. Here, it is an ECS container instance that is an EC2 instance running the agent, has a specifically defined IAM policy and role, and has been registered into your cluster.

And as you probably guessed, in these instances, you are running containers.

AMI*

These container instances can use any AMI as long as it has the following specifications: a modern Linux distribution with the agent and the Docker Daemon with any Docker runtime dependencies running on it.

Want it more simplified? Well, AWS created the Amazon ECS-Optimized AMI for just that. Not only does that AMI come preconfigured with all of the previously mentioned specifications, it’s tested and includes the recommended ecs-init upstart process to run and monitor the agent.

Cluster

An ECS cluster is a grouping of (container) instances* (or tasks in Fargate) that lie within a single region, but can span multiple Availability Zones – it’s even a good idea for redundancy. When launching an instance (or tasks in Fargate), unless specified, it registers with the cluster named “default”. If “default” doesn’t exist, it is created. You can also scale and delete your clusters.

Agent*

The Amazon ECS container agent is a Go program that runs in its own container within each EC2 instance that you use with ECS. (It’s also available open source on GitHub!) The agent is the intermediary component that takes care of the communication between the scheduler and your instances. Want to register your instance into a cluster? (Why wouldn’t you? A cluster is both a logical boundary and provider of pool of resources!) Then you need to run the agent on it.

Task

When you want to start a container, it has to be part of a task. Therefore, you have to create a task first. Succinctly, tasks are a logical grouping of 1 to N containers that run together on the same instance, with N defined by you, up to 10. Let’s say you want to run a custom blog engine. You could put together a web server, an application server, and an in-memory cache, each in their own container. Together, they form a basic frontend unit.

Task definition

Ah, but you cannot create a task directly. You have to create a task definition that tells ECS that “task definition X is composed of this container (and maybe that other container and that other container too!).” It’s kind of like an architectural plan for a city. Some other details it can include are how the containers interact, container CPU and memory constraints, and task permissions using IAM roles.

Then you can tell ECS, “start one task using task definition X.” It might sound like unnecessary planning at first. As soon as you start to deal with multiple tasks, scaling, upgrades, and other “real life” scenarios, you’ll be glad that you have task definitions to keep track of things!

Scheduler*

So, the scheduler schedules… sorry, this should be more helpful, huh? The scheduler is part of the “hosted orchestration layer” provided by ECS. Wait a minute, what do I mean by “hosted orchestration”? Simply put, hosted means that it’s operated by ECS on your behalf, without you having to care about it. Your applications are deployed in containers running on your instances, but the managing of tasks is taken care of by ECS. One less thing to worry about!

Also, the scheduler is the component that decides what (which containers) gets to run where (on which instances), according to a number of constraints. Say that you have a custom blog engine to scale for high availability. You could create a service, which by default, spreads tasks across all zones in the chosen region. And if you want each task to be on a different instance, you can use the distinctInstance task placement constraint. ECS makes sure that not only this happens, but if a task fails, it starts again.

Service

To ensure that you always have your task running without managing it yourself, you can create a service based on the task that you defined and ECS ensures that it stays running. A service is a special construct that says, “at any given time, I want to make sure that N tasks using task definition X1 are running.” If N=1, it just means “make sure that this task is running, and restart it if needed!” And with N>1, you’re basically scaling your application until you hit N, while also ensuring each task is running.