Black Friday sale: an extra 40% off every course until Monday, November 30.New Year sale: an extra 30% off every course until Friday, January 1.Memorial Day sale: an extra 25% off every course until Monday, May 31.SysAdmin Day sale: an extra 25% off every course until Sunday, August 1.Back to School sale: an extra 25% off every course until Sunday, August 22.Labor Day sale: an extra 25% off every course until Monday, September 6.Black Friday sale: an extra 40% off every course until Monday, November 29.New Year sale: an extra 30% off every course until Saturday, January 1.Choose your certification →

Copyright (c) 2026 MindMesh Academy. All rights reserved. This content is proprietary and may not be reproduced or distributed without permission.

2.1.1.9. Auto Scaling Patterns for Serverless and Container Workloads (Lambda Concurrency, HPA, Karpenter, Application Auto Scaling)

2.1.1.9. Auto Scaling Patterns for Serverless and Container Workloads (Lambda Concurrency, HPA, Karpenter, Application Auto Scaling)

💡 First Principle: Scale the layer that is actually short of capacity: invocations, tasks or pods, or the nodes underneath them. Each layer has its own controller, and scaling one does not scale the others.

AWS Lambda:
  • Concurrency = requests per second x average duration. Accounts start with 1,000 concurrent executions per Region; each function can add up to 1,000 execution environments every 10 seconds.
  • Reserved concurrency both guarantees and caps a function's concurrency (requests above it are throttled), which protects a downstream database or API from bursts. It does not pre-warm anything.
  • Provisioned concurrency keeps pre-initialized environments ready, removing cold starts for latency-sensitive paths, and is billed while configured. Automate it with Application Auto Scaling: scheduled actions for known peaks and target tracking on provisioned concurrency utilization.
  • SQS event sources: set maximum concurrency (2 to 1,000) on the event source mapping to cap how many concurrent invocations one queue drives. Throttling a queue consumer with reserved concurrency instead makes batches fail and return to the queue, and each failed receive counts toward the redrive policy's maxReceiveCount, so healthy messages can land in the dead-letter queue. Keep reserved concurrency at or above the total maximum concurrency of the function's SQS sources.

Amazon ECS: Service auto scaling uses Application Auto Scaling with target tracking (average CPU or memory, or ALBRequestCountPerTarget), step scaling or scheduled actions to change the task count. On Fargate that is all you scale. On EC2 capacity, a capacity provider with managed scaling grows and shrinks the Auto Scaling group to fit the tasks.

Amazon EKS:
  • Horizontal Pod Autoscaler (HPA) changes the number of pod replicas from CPU, memory or custom metrics (KEDA adds event-driven sources such as queue depth). It does not add nodes: when the cluster is full, new pods stay Pending.
  • Cluster Autoscaler adds or removes nodes in existing node groups (Auto Scaling groups) when pods are unschedulable, so it is limited to the instance types each group was built with.
  • Karpenter watches unschedulable pods and launches right-sized instances directly from a broad set of instance types and purchase options (including Spot), then consolidates underused nodes. EKS Auto Mode runs this node management for you, and Fargate profiles remove node scaling for the selected pods.

Application Auto Scaling is the common engine for scalable targets beyond EC2: ECS services, Lambda provisioned concurrency, DynamoDB capacity, Aurora replicas, SageMaker AI endpoint variants and more.

SymptomLayer to scale
Lambda cold starts at a known 8 a.m. peakProvisioned concurrency with scheduled Application Auto Scaling
Lambda floods a database during a queue backlogSQS event source maximum concurrency (plus RDS Proxy)
ECS tasks saturated as requests growTarget tracking on ALBRequestCountPerTarget or CPU
EKS pods Pending after HPA scales outKarpenter (or Cluster Autoscaler) for nodes

⚠️ Exam Trap: Answering "configure the Horizontal Pod Autoscaler" when the symptom is pods stuck in Pending. HPA already created the pods; the cluster needs node capacity from Karpenter, Cluster Autoscaler, EKS Auto Mode or Fargate.

See how it connects
Alvin Varughese
Written byAlvin Varughese
Founder•26 professional certifications earned