Launch instance on service deployment - terraform

Question

I am using Terraform to deploy an ECS EC2 cluster on AWS.
My pipeline is creating a new task definition from docker-compose, then updates the service to use this task definition.
Desired count is 1, deployment_minimum_healthy_percent is 100 and deployment_maximum_percent is 200.
Expected behavior is that the autoscaling group will launch a new EC2 instance to deploy the new task, then kill the old one.
What happens is that I get an error message : "service X was unable to place a task because no container instance met all of its requirements. The closest matching container-instance has insufficient memory available."
No instance is created and the deployment is rolled back. How can I make sure that an extra instance is created when deploying my service ?
Here is my Terraform code :
resource "aws_ecs_cluster" "main_cluster" {
name = var.application_name
tags = {
Environment = var.environment_name
}
}
data "aws_ecs_task_definition" "main_td" {
task_definition = var.application_name
}
resource "aws_ecs_service" "main_service" {
name = var.application_name
cluster = aws_ecs_cluster.main_cluster.id
launch_type = "EC2"
scheduling_strategy = "REPLICA"
task_definition = "${data.aws_ecs_task_definition.main_td.family}:${data.aws_ecs_task_definition.main_td.revision}"
desired_count = var.target_capacity
deployment_minimum_healthy_percent = 100
deployment_maximum_percent = 200
health_check_grace_period_seconds = 10
wait_for_steady_state = false
force_new_deployment = true
load_balancer {
target_group_arn = aws_lb_target_group.main_tg.arn
container_name = var.container_name
container_port = var.container_port
}
ordered_placement_strategy {
type = "binpack"
field = "memory"
}
deployment_circuit_breaker {
enable = true
rollback = true
}
lifecycle {
ignore_changes = [desired_count]
}
tags = {
Environment = var.environment_name
}
}
Auto-scaling group :
data "template_file" "user_data" {
template = "${file("${path.module}/user_data.sh")}"
vars = {
ecs_cluster = "${aws_ecs_cluster.main_cluster.name}"
}
}
resource "aws_launch_configuration" "main_lc" {
name = var.application_name
image_id = var.ami_id
instance_type = var.instance_type
associate_public_ip_address = true
iam_instance_profile = "arn:aws:iam::812844034365:instance-profile/ecsInstanceRole"
security_groups = ["${aws_security_group.main_sg.id}"]
root_block_device {
volume_size = "30"
volume_type = "gp3"
}
user_data = "${data.template_file.user_data.rendered}"
}
resource "aws_autoscaling_policy" "main_asg_policy" {
name = "${var.application_name}-cpu-scale-policy"
policy_type = "TargetTrackingScaling"
autoscaling_group_name = aws_autoscaling_group.main_asg.name
estimated_instance_warmup = 10
target_tracking_configuration {
predefined_metric_specification {
predefined_metric_type = "ASGAverageCPUUtilization"
}
target_value = 40.0
}
}
resource "aws_autoscaling_group" "main_asg" {
name = var.application_name
launch_configuration = aws_launch_configuration.main_lc.name
min_size = var.target_capacity
max_size = var.target_capacity * 2
health_check_type = "EC2"
health_check_grace_period = 10
default_cooldown = 30
desired_capacity = var.target_capacity
vpc_zone_identifier = data.aws_subnet_ids.subnets.ids
wait_for_capacity_timeout = "3m"
instance_refresh {
strategy = "Rolling"
preferences {
min_healthy_percentage = 100
}
}
}
Module is published here https://registry.terraform.io/modules/hboisgibault/ecs-cluster/aws/latest