From Fleeticide To Recovery: When Your Auto-Scaling Policy Becomes Your Worst Enemy
Conference (INTERMEDIATE level)
Horizontal scaling sounds simple in a README, but the reality of a Spring Boot fleet on ECS can quickly turn into "Fleeticide." We've all seen it: connection pools saturate, health checks flap, and suddenly your auto-scaling policy becomes an auto-failing policy.
In this session, we go behind the scenes of a real-world scaling disaster. We'll dissect our initial naive assumptions and dive into the technical shifts that saved our production environment. We will move beyond "CPU/Memory 101" to look at:
The Observability Gap: Tracking the "hidden" metrics (Thread pools, DB connections) that actually dictate survival.
AWS Patterns for the Real World: Moving beyond basic CloudWatch alarms to custom metrics and predictive scaling that respects Spring Boot's startup overhead.
Architecting for Elasticity: How to bake scaling-awareness into the developer workflow to prevent connection storms and cascading failures.
You'll leave with a battle-tested checklist of what to monitor, architectural patterns for graceful scale-out, and the hard-won wisdom of why you must control your application's failure points before they control you.
In this session, we go behind the scenes of a real-world scaling disaster. We'll dissect our initial naive assumptions and dive into the technical shifts that saved our production environment. We will move beyond "CPU/Memory 101" to look at:
The Observability Gap: Tracking the "hidden" metrics (Thread pools, DB connections) that actually dictate survival.
AWS Patterns for the Real World: Moving beyond basic CloudWatch alarms to custom metrics and predictive scaling that respects Spring Boot's startup overhead.
Architecting for Elasticity: How to bake scaling-awareness into the developer workflow to prevent connection storms and cascading failures.
You'll leave with a battle-tested checklist of what to monitor, architectural patterns for graceful scale-out, and the hard-won wisdom of why you must control your application's failure points before they control you.