Copyright (c) 2026 MindMesh Academy. All rights reserved. This content is proprietary and may not be reproduced or distributed without permission.
3.1.2.5. Identifying & Remediating Scaling Issues
3.1.2.5. Identifying & Remediating Scaling Issues
Your auto-scaling is configured, but the application still crashes under load. Scaling issues fall into predictable categories.
Common scaling failures and fixes:
| Symptom | Root Cause | Fix |
|---|---|---|
| New instances launch but can't serve traffic | Long bootstrap time (install dependencies at startup) | Bake AMI with pre-installed deps |
| Scale-out happens but too slowly | Scaling metric responds slowly to load | Use a leading indicator metric (queue depth, request count) |
| Instances scale out then immediately scale in | Thrashing — scale-in and scale-out thresholds too close | Add cooldown period (300s) or widen threshold gap |
| Hit max capacity but load continues growing | Max capacity too low | Increase ASG max, or redesign for horizontal scalability |
| Database becomes bottleneck at scale | Compute scales but data tier doesn't | Add read replicas, ElastiCache, or DynamoDB on-demand |
| API Gateway returns 429 | Throttling limit reached | Request limit increase, implement caching, use usage plans |
| ASG stays at its minimum even though the target is breached | Policy not attached/enabled, or launches failing | Confirm the policy on the ASG; read the scaling activity history (describe-scaling-activities) for errors |
| Sudden spikes arrive before new instances are healthy (minutes to boot) | Launch time longer than the spike's ramp | Warm pool of pre-initialized (Stopped) instances; predictive scaling for recurring peaks |
Scaling warmup and cooldown:
- Instance warmup: Time for a new instance to start serving traffic. ASG ignores the instance's metrics during warmup to prevent premature scale-in.
- Cooldown period: After a scaling action, wait before evaluating again. Prevents thrashing.
Launch template optimization: Use launch templates (not launch configurations — deprecated) with:
- Pre-baked AMIs (faster bootstrap)
- Instance type diversity (Spot + On-Demand mix)
- Placement groups for low-latency communication
Exam Trap: ASG target tracking scaling already handles cooldown intelligently — it prevents scale-in while scale-out is in progress. But step scaling uses a fixed cooldown timer. If the exam describes "instances scale out and immediately scale back in" with step scaling, increase the cooldown period. With target tracking, this problem rarely occurs.
Enroll to unlock the 3 practice questions written for this section, so you can test what you just read while it is fresh.
Enroll to unlock the 2 flashcards for this section and review them on a spaced-repetition schedule.
Written byAlvin Varughese
Founder•20 professional certifications