An Auto Scaling group processes SQS messages. CPU utilization remains low because jobs wait on external services, but queue latency rises when demand increases. Average processing time is stable. Which metric best supports target tracking for this worker fleet?

1. Approximate visible queue backlog divided by the number of in-service worker instances.
2. The unnormalized visible message count with no consideration of worker count.
3. Total messages received since the queue was created.
4. Use ASGAverageCPUUtilization as the only target-tracking metric despite persistently low CPU.