AWS Lambda: The Architecture Nobody Explains Properly
Every Lambda invocation you've ever made runs on an EC2 Spot instance, inside a Firecracker microVM, managed by a fleet of services most people never see. In this deep dive I show you exactly how — plus the two 2025 features (Managed Instances + Durable Functions) that change the game.
⏱️ CHAPTERS:
0:00 - The myth of serverless
0:16 - The real architecture (Worker Hosts + Firecracker)
1:36 - How your code actually runs (Execution Environments)
2:28 - The 3 invocation types overview
3:24 - Sync invocation — full flow deep dive
4:53 - Async invocation + DLQ deep dive
6:00 - Event Source Mapping (Kinesis/SQS) deep dive
7:26 - Inside the MicroVM — the Runtime API
9:20 - Lambda Managed Instances (re:Invent 2025)
10:21 - Durable Functions (re:Invent 2025)
10:56 - Which Lambda flavor for which workload
11:40 - Wrap-up
🧠 WHAT YOU'LL LEARN:
• How Firecracker (Rust-based VMM) boots a microVM in ~100ms
• The difference between a Worker Host, an Execution Environment, and a microVM
• Why the 3 invocation types (sync, async, ESM) have completely different retry semantics
• The real story on cold starts — and why most advice you've read is wrong
• Why VPC Lambda cold starts are no longer a problem (Hyperplane ENIs, 2019)
• Lambda Managed Instances: EC2 pricing model + connection pooling built in
• Lambda Durable Functions: steps + waits model, how it compares to Step Functions
• The concurrency limit trap that's caused real production outages
🏗️ WHO THIS IS FOR:
Junior-to-senior backend engineers and architects who use Lambda in production and want to understand what's actually happening — not just "upload code, it runs."
📄 READ THE FULL BREAKDOWN:
/ aws-lambda-architecture-deep-dive
🔗 CONNECT:
LinkedIn: / joud-awad
Medium: / joudwawad
X (Twitter): https://x.com/TheJoud97
GitHub: https://github.com/JoudAwad97