The healthy-host-count alarm fired because a Terraform-managed rolling deployment inadvertently kept HealthyHostCount at 1 for 5 consecutive minutes (13:55–13:59 UTC), tripping the 4-minute sustained threshold designed to distinguish real outages from normal rolling-deploy churn. CloudTrail shows Terraform deregistered warrn-api-1 at 13:55:12, re-registered it at 13:58:00, then deregistered warrn-api-2 only 21 seconds later — before api-1 had passed its ALB health checks — so the count never recovered to 2 between the two operations. CloudWatch fired ALARM at 14:01:22 and auto-resolved to OK at 14:03:22. Both EC2 instances remained running with passing status checks throughout; no crash, no stop-instance event, and no 5xx errors were observed. The service was degraded to a single host for ~5 minutes but was never completely down.