ALB Target Group Slow Start Latency Under Concurrency

Direct Answer

ALB target group slow start does not apply a fixed latency to new targets. It withholds traffic until targets pass health checks and paces new connections at roughly 60 per second per target during ramp-up. Latency under concurrency is therefore a function of your health-check interval, healthy threshold, and request arrival rate, not a constant ALB delay. The slow start mechanism is independent of EC2 T-instance CPU credit balance; credit depletion is a target-side capacity issue.

Confirmed Facts

• Target registration time equals the health-check interval multiplied by the healthy threshold count.
• ALB slow start limits new connections per target during the warm-up window.
• Slow start duration is configured per target group using SlowStartDurationSeconds.
• T-instance CPU credits regulate instance burst capacity and do not control ALB connection pacing.

Likely Explanation for Concurrent Latency Spikes

When concurrent requests arrive while targets are warming up, the ALB may queue or delay requests because new connections are being introduced gradually. If targets have not satisfied the healthy threshold, they receive no traffic at all, which can concentrate load on already healthy targets and raise response times. CPU credit depletion on T-instances can amplify latency once traffic reaches the instance, but it does not create the ALB slow start delay itself.

Steps for This Case

1. Identify the affected target group ARN.
2. Read the health-check interval and healthy threshold:

aws elbv2 describe-target-groups --target-group-arns "$TARGET_GROUP_ARN" --query 'TargetGroups[0].{Interval:HealthCheckInterval, HealthyThreshold:HealthyThresholdCount}' --output table

1. Check whether SlowStartDurationSeconds is set on the target group in the console or API.
2. Compare CloudWatch HealthyHostCount and TargetResponseTime during the spike.
3. On T-instances, review CPUCreditBalance and CPUUtilization during the same window to test whether credit depletion correlates with the latency.

Missing Diagnostic Detail

What are the configured health-check interval and healthy threshold count for this target group? These values determine the registration latency and decide whether the fix belongs in load balancer configuration, target sizing, or both.

Assumptions and Uncertainty

• This applies to Application Load Balancers, not Network Load Balancers.
• The roughly 60 connections per second ramp figure is a documented default reference; actual enforcement can vary and should be confirmed in your environment.
• CPU credit correlation is unverified for your instances and remains a hypothesis until CloudWatch metrics are reviewed.
• This is a conservative draft pending verification of your target group health-check and slow start settings.