How the ECS Scheduler Places & Replaces Tasks

The ECS scheduler continuously compares a service's desired task count against its actual running task count and takes action to reconcile any difference. On the EC2 launch type, placement strategies (spread, binpack, random) determine which instance a new task lands on — 'spread' across Availability Zones is the default and most common choice for resilience, ensuring an AZ outage doesn't take down every task at once. On Fargate, AWS handles placement transparently since there's no visible instance layer to choose between. When a task fails a health check (either the ECS container health check or, more commonly, the load balancer target group's health check), ECS deregisters it from the target group, stops it, and launches a replacement — this entire loop typically completes in well under a minute, which is why ECS services can absorb individual task failures without visible downtime as long as desired count stays above the traffic-serving minimum.