Auto Scaling Is Adding Instances. Response Times Are Still Climbing.

title: "Auto Scaling Is Adding Instances. Response Times Are Still Climbing." description: "Scaling kicks in, new instances launch, but response times keep rising and you can't understand why. The gap between InService and actually ready is where this lives." date: "2026-04-06" series: "aws-daily-with-divine" tags: ["aws", "auto-scaling", "lifecycle-hooks", "performance"] linkedinUrl: "https://www.linkedin.com/posts/divine-chukwu-63bb04145_aws-autoscaling-devops-activity-7444292480021225472-z3av"

If you've used Auto Scaling, you've probably experienced this. The scaling kicks in, new instances launch, but response times keep climbing and you can't understand why.

Here's what is actually happening.

The InService lie

When Auto Scaling launches a new instance, by default it immediately marks it as InService and the load balancer starts sending traffic to it.

But the instance might not actually be ready yet. The OS is still booting. Your application is still starting up. Dependencies are still loading. Requests hit it before it can handle them properly, which causes errors and slow responses — even though Auto Scaling is doing exactly what you configured it to do.

Auto Scaling working correctly and your application being ready are two different things.

The fix: lifecycle hooks

Lifecycle hooks let you pause the instance in a Pending state before it goes InService. During that pause you can:

• Run scripts
• Install software
• Pull configs
• Warm up the application cache
• Run health checks

Only when your script signals ready does Auto Scaling move the instance to InService and the load balancer starts sending traffic.

Concrete numbers

Say your application takes 3 minutes to fully initialize — loading config, connecting to the database, warming the cache.

Setting one up

Adding a launch lifecycle hook is one CLI command:

aws autoscaling put-lifecycle-hook \
  --auto-scaling-group-name my-asg \
  --lifecycle-hook-name on-instance-launch \
  --lifecycle-transition autoscaling:EC2_INSTANCE_LAUNCHING \
  --heartbeat-timeout 600 \
  --default-result ABANDON

Then in your bootstrap script, when initialization is genuinely complete:

aws autoscaling complete-lifecycle-action \
  --lifecycle-hook-name on-instance-launch \
  --auto-scaling-group-name my-asg \
  --lifecycle-action-result CONTINUE \
  --instance-id "$INSTANCE_ID"

ABANDON as the default result is a deliberate choice — if your script never signals ready before the timeout, the instance is removed instead of taking traffic. Fail closed, not fail open.

Have you experienced this before? How did you fix it?