A city clerk’s office runs a document OCR fleet on EC2 that backs u…

Question

A city clerk’s office runs a document OCR fleet on EC2 that backs up during quiet weeks but must expand when a mail-in form season fills an SQS queue. How should Auto Scaling be designed for reliability?

Accepted Answer

Combine target-tracking or step policies on CPU with scaling actions driven by SQS ApproximateNumberOfMessages so capacity follows both compute pressure and backlog — Quiet weeks shouldn’t pay for parade-day capacity, and a fat queue shouldn’t wait for a clock. Tie scaling to CPU and to how deep the OCR inbox is.

Answer

Keep a fixed fleet sized for peak OCR load year-round and disable scaling policies

Answer

Scale only on a nightly schedule regardless of queue depth or CPU

Answer

Terminate all instances when CPU is low even if the SQS backlog is still large

A city clerk’s office runs a document OCR fleet on EC2 that backs up during quiet weeks but must expand when a mail-in form season fills an SQS queue. How should Auto Scaling be designed for reliability?

Short Explanation

Full Explanation