When you need to move a legacy system to AWS or set up new infrastructure, searching for “AWS best practices” shows a lot of impressive architectures.
Things like EKS, multi-region setups, full Terraform modules, and service mesh.
Of course, those setups make total sense for big tech companies with huge traffic and dedicated infrastructure teams.
But in SI projects or smaller teams where I have mostly worked, the reality is very different.
There is rarely a dedicated DevOps team, and once the project ends, a small group of developers has to maintain everything themselves.
If you ignore that reality and just copy a textbook setup, it ends up becoming something neither the ops team nor I can easily handle later.
So instead of chasing fancy setups, I tried to set practical standards for how to use AWS effectively within real-world limits.
Move uploaded files off the server disk to S3
In older legacy projects, it was very common to save user-uploaded images or attachments directly on the server VM’s local disk (like an /upload folder).
When everything runs quietly on a single server, it seems fine. But once you move to AWS and start using Docker containers or replacing servers, those local files get in the way.
If a container restarts, files can disappear, or if you add a second server, the files are out of sync and images break.
So when moving to AWS, moving file storage to S3 was one of the most helpful changes.
Even without setting up CloudFront as a full CDN in front, just uploading to a private S3 bucket and giving out short-lived Presigned URLs for downloads is more than enough.
Once files are moved out like that, the server itself becomes easy to tear down and redeploy anytime without worrying about losing data.
Keeping deployment simple with Docker and Jenkins
In older legacy projects, it was still common to build a .war file locally, upload it via SFTP, and restart Tomcat manually.
Because it relies on manual steps, people eventually forget a config file or panic when they need to roll back to the previous version.
On the other hand, bringing in Kubernetes felt too heavy for a small team to learn and maintain.
A practical middle ground was EC2 + Docker + Jenkins.
Just having Jenkins build and deploy a Docker image on push, with an easy way to roll back to the previous image tag if something breaks, cut deployment stress in half.
Also, moving away from SSH-ing into servers to run tail -f and simply collecting app logs in CloudWatch Logs made finding issues much faster.
Adding basic alarms for CPU, disk, and 5xx errors was enough to avoid most surprises.
Moving sessions to Redis to make servers stateless
After moving files to S3, the next problem was user sessions.
When sessions lived inside Tomcat memory, restarting the server logged everyone out, and adding a second server broke login state across instances.
So I spun up Redis on ElastiCache and moved sessions out there.
With files in S3 and sessions in Redis, the application servers finally became stateless.
That made zero-downtime deploys and scaling out an extra server much less stressful.
Separating configs and secrets with Parameter Store
Before, it was common to put DB passwords or API keys directly in config files like application.yml, or manage separate config files on each server.
It was always easy to accidentally commit secrets to Git or overwrite the wrong config on production.
Moving those values into AWS Parameter Store or Secrets Manager cleanly separated code from configuration.
The server fetches the right values for its environment at startup, so no secrets stay in the repo and we don’t have to babysit config files during deploys.
Offloading heavy follow-up work to SQS
When handling an order or request, running things like sending emails or calling external APIs synchronously meant that if an external service slowed down, our main response slowed down with it.
For tasks that didn’t need an immediate result on the user’s screen, I pushed messages to an SQS queue and let a background worker handle them.
Cutting that direct link kept temporary external issues from taking down the main service.
In the end, for a small team, taking only what the current team can actually handle makes more sense than stacking the newest AWS services.
That is what the company wanted at the time, and honestly, if I were the owner, I would have wanted the exact same thing.