AWS Builder Center
Understanding the Shared Responsibility Model Through a Multi-AZ Deployment

Understanding the Shared Responsibility Model Through a Multi-AZ Deployment

The Shared Responsibility Model is one of those concepts that's easy to recite and surprisingly easy to misunderstand in practice. "AWS secures the cloud, you secure what's in the cloud" sounds clean in a slide deck. It didn't actually mean anything to me until I built a multi-AZ VPC and had to figure out, piece by piece, which failures were mine to fix and which weren't.

I am a computer science student specializing in cloud computing primarily in AWS

The version I memorized vs. the version I needed

Before building anything real, my understanding of the Shared Responsibility Model was basically: "AWS handles the hardware, I handle my code." True, but useless in the moment something actually breaks — it doesn't tell you which side of the line a specific problem sits on.
Building a VPC with an Application Load Balancer and an Auto Scaling group across two Availability Zones forced me to get specific about it, because when something didn't work, I had to know where to even start looking.

What was never going to be my problem

AWS guarantees the existence and independence of the Availability Zones themselves — the physical facilities, power, cooling, and the network backbone connecting them. I didn't provision any of that, can't see it, and importantly, can't misconfigure it. If an entire AZ has a genuine outage, that's inside AWS's side of the line, and no setting in my account would have prevented or caused it.
This mattered practically: early on, when a health check started failing, I briefly wondered if it was an AWS-side networking issue. It wasn't, and it almost never is. Ruling that out quickly meant looking at my own configuration first — which turned out to be exactly right.

What was absolutely my problem

Everything that actually went wrong while building this lived entirely on my side of the line:
  • Whether I was actually spanning two Availability Zones, or only thought I was. Multi-AZ isn't automatic — it depends on which subnets I attached to the Auto Scaling group. Getting this wrong doesn't throw an error; it just quietly defeats the entire point of the architecture.
  • Security group rules controlling what could reach my EC2 instances. AWS provides the mechanism; every rule inside it was a decision I had to make and could get wrong.
  • The IAM role attached to my Auto Scaling launch template. If instances launch without the right permissions, that's not a platform failure — that's a role I configured incorrectly.
  • Whether my route tables correctly classified subnets as public or private. This is pure configuration, entirely on my side, and it's exactly the kind of mistake that looks like a "networking problem" but is actually just a "my problem" problem.

The realization that made it click

The Shared Responsibility Model isn't really about AWS vs. you as two teams working on separate parts of the same system. It's closer to: AWS guarantees the primitives are real and available; you guarantee you're using them correctly. AWS guarantees two independent Availability Zones exist. It does not guarantee that your Auto Scaling group is actually configured to use both of them — that's a decision, and decisions are always your responsibility.
Once I started treating every debugging session with that question — "is this a primitive that AWS guarantees, or a decision I made?" — I stopped wasting time looking in the wrong place. Almost everything that goes wrong in an early AWS project is a decision, not a platform failure.

What I'd tell someone learning this for the first time

Don't try to memorize the line — build something that forces you to find it. A single-AZ static site won't teach you this, because there's barely a line to find. A multi-AZ deployment will, because half your bugs will be "which side is this even on" before they're "how do I fix this."

Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article