Ondřej Plátek Blog
PhD candidate@UFAL, Prague. LLM & TTS evaluation. Engineer. Researcher. Speaker. Father.

What to setup first on AWS?

What to setup first?

is the most important question. AWS is useful, well documented but overwhelming.

Let me introduce my strategy.

Billing alerts

There is always danger of accidentally spending money. Set up alerts may prevent

  • overstepping a planned budget,
  • ignoring increased usage of your service,
  • keeping useless resources deployed. If not setup one is able to spend a fortune on unfitting infrastructure.

Setup logging

If you never experienced random, rare, hard to replicate bugs, you are lucky. For the rest of us is logging the crucial tool examine behaviour over long period of time and across many deployments and devices.

CloudWatch is very easy to integrate and archive, but it lacks more advanced features. Luckily, there is a plenty of tools which can analyze, filter and monitor logs and provide much prettier and efficient interface. Often they easily integrate with Cloudwatch which or can be used directly. I would like to mention two of them:

  • Rapid7 InsightIps for easy to use filters and setting up event notifications. Plus I like their UI.
  • Loggly excels with their query language for the logs where you can query across different source of log streams with common value e.g. time stamps.

Infrastructure as a code (IaC)

There has been written a lot about IaC, and rightly so. I recommend following posts and videos to check:

However, I would like to make another point. IaC is so awesome also because it is an alternative view on the deployment.

One type of documentation already exists. This is the default AWS documentation. The form is well known; Click on resource XY (or use this http request). Select this parameter. Wait few minutes and continue by deploying resource YZ. The typical procedural view.

IaC brings completely different and more concise documentation. Terraform documentation feels very different: Declare resource XY. You have to provide this parameters, and there another three completely optional parameters with sensible defaults. Such documentation offers more high-level overview. I find it much easier to read.

Typically, I read first Terraform documentation to learn the high-level concepts. Later, to understand the documentation more deeply I switch to the AWS documentation. It is very refreshing. IaC gives you benefit of many examples to learn from because there is a lot of open-source development in IaC community.

I really enjoy combining the tools for debugging: Terraform plan & deploy commands are the commands where I discover the bugs. The AWS cli often helps me analyze the problems, but returning to gold old fashion web browser and AWS console gives me alternative view where it is often on the first sight obvious what is wrong.

List of services

  • EC2
    • machines on demand - IaC can be used to build the machine image and also to manage the life cycle of the machine.
    • autoscaling groups - which can automatically scale number of identical machines based on traffic which the machines serve.
  • ALB or ELB - load balancers for balancing the traffic among different machines.
  • ECR - hosted docker container registry.
  • ECS/Fargate - hosted containers on demand. I have only a little experience with Fargate, but it works like magic, no machines needed to be managed.
  • CloudWatch - basic tool to collect, display and monitor logs.
  • S3 - popular object storage.
  • Route 53 - managed DNS service. Very simple & cool together with Terraform since it allows to change domain names for services.