Non-prod Auto Scaling Groups running at full capacity 24/7 waste 70% of costs. Here's how to add scheduled scaling with 5 lines of Terraform and reclaim those hours.
Reality check: Your development and QA environments run 24/7, but your team works 9-5, Monday-Friday.
The math:
Hours per week: 168 total
Developer hours: 40-50 (weekdays)
Wasted hours: 118-128 (nights + weekends)
Your dev environment is idle 70% of the time.
But you're paying 100% of the cost.
Typical dev environment:
- 3 EC2 instances (t3.large) = $225/month
- Needed hours: 40/week
- Actual usage: 70% waste
- You could pay: $70/month instead
Let me show you how to add scheduled scaling with literally 5 lines of Terraform and slash your non-prod costs by 65%.
💸 The Always-On Tax
Most dev/staging/QA environments look like this:
Auto Scaling Group config:
Min: 3 instances
Max: 6 instances
Desired: 3 instances
Schedule: None (runs 24/7)
Cost breakdown:
3 × t3.large instances
- $0.0832/hour × 3 × 730 hours = $182/month
- EBS volumes: $24/month
- Load balancer: $16/month
Total: $222/month
Annual: $2,664
Actual needed hours: 50 hours/week = 217 hours/month
Waste: 513 hours/month (70% waste!)
Wasted money: $155/month = $1,860/year
🎯 The Solution: Scheduled Actions
Auto Scaling Groups support scheduled actions - change capacity on a schedule.
Strategy:
Business hours (8 AM - 6 PM weekdays): Full capacity
Nights and weekends: Scale to zero (or minimal)
Implementation: 5 lines of Terraform per schedule.
🛠️ Terraform Implementation
Basic Weekday Schedule
# asg-scheduled-scaling.tf
# Scale UP for business hours (8 AM Monday-Friday)
resource "aws_autoscaling_schedule" "scale_up_weekday" {
scheduled_action_name = "scale-up-weekday"
min_size = 3
max_size = 6
desired_capacity = 3
recurrence = "0 8 * * 1-5" # 8 AM Mon-Fri (UTC)
autoscaling_group_name = aws_autoscaling_group.dev.name
}
# Scale DOWN for nights (6 PM Monday-Friday)
resource "aws_autoscaling_schedule" "scale_down_weekday" {
scheduled_action_name = "scale-down-weekday"
min_size = 1
max_size = 2
desired_capacity = 1
recurrence = "0 18 * * 1-5" # 6 PM Mon-Fri (UTC)
autoscaling_group_name = aws_autoscaling_group.dev.name
}
# Scale DOWN for weekends (Friday night)
resource "aws_autoscaling_schedule" "scale_down_weekend" {
scheduled_action_name = "scale-down-weekend"
min_size = 1
max_size = 2
desired_capacity = 1
recurrence = "0 18 * * 5" # 6 PM Friday (UTC)
autoscaling_group_name = aws_autoscaling_group.dev.name
}
# Scale UP Monday morning
resource "aws_autoscaling_schedule" "scale_up_monday" {
scheduled_action_name = "scale-up-monday"
min_size = 3
max_size = 6
desired_capacity = 3
recurrence = "0 8 * * 1" # 8 AM Monday (UTC)
autoscaling_group_name = aws_autoscaling_group.dev.name
}
That's it! 4 scheduled actions, 20 lines of code, done. ✅
Production-Ready Module
# modules/asg-scheduled-scaling/main.tf
variable "asg_name" {
description = "Auto Scaling Group name"
type = string
}
variable "business_hours" {
description = "Business hours configuration"
type = object({
min_size = number
max_size = number
desired_capacity = number
start_hour = number # 8 = 8 AM
end_hour = number # 18 = 6 PM
})
default = {
min_size = 3
max_size = 6
desired_capacity = 3
start_hour = 8
end_hour = 18
}
}
variable "off_hours" {
description = "Off-hours configuration"
type = object({
min_size = number
max_size = number
desired_capacity = number
})
default = {
min_size = 1
max_size = 2
desired_capacity = 1
}
}
variable "timezone_offset" {
description = "Hours offset from UTC (e.g., -5 for EST)"
type = number
default = 0
}
locals {
# Adjust hours for timezone
start_hour_utc = (var.business_hours.start_hour - var.timezone_offset) % 24
end_hour_utc = (var.business_hours.end_hour - var.timezone_offset) % 24
}
# Scale up for business hours (weekdays)
resource "aws_autoscaling_schedule" "scale_up" {
scheduled_action_name = "${var.asg_name}-scale-up-weekday"
min_size = var.business_hours.min_size
max_size = var.business_hours.max_size
desired_capacity = var.business_hours.desired_capacity
recurrence = "${local.start_hour_utc} * * 1-5"
autoscaling_group_name = var.asg_name
}
# Scale down for nights/weekends
resource "aws_autoscaling_schedule" "scale_down" {
scheduled_action_name = "${var.asg_name}-scale-down-weekday"
min_size = var.off_hours.min_size
max_size = var.off_hours.max_size
desired_capacity = var.off_hours.desired_capacity
recurrence = "${local.end_hour_utc} * * 1-5"
autoscaling_group_name = var.asg_name
}
output "schedules_created" {
value = {
scale_up = aws_autoscaling_schedule.scale_up.scheduled_action_name
scale_down = aws_autoscaling_schedule.scale_down.scheduled_action_name
}
}
Usage
# main.tf
# Dev environment - keep 1 instance off-hours
module "dev_scaling" {
source = "./modules/asg-scheduled-scaling"
asg_name = aws_autoscaling_group.dev.name
timezone_offset = -5 # EST
business_hours = {
min_size = 2
max_size = 5
desired_capacity = 2
start_hour = 8 # 8 AM EST
end_hour = 18 # 6 PM EST
}
off_hours = {
min_size = 1
max_size = 2
desired_capacity = 1 # Keep 1 instance for CI/CD
}
}
# QA environment - keep 1 instance off-hours
module "qa_scaling" {
source = "./modules/asg-scheduled-scaling"
asg_name = aws_autoscaling_group.qa.name
timezone_offset = -5
business_hours = {
min_size = 3
max_size = 6
desired_capacity = 3
start_hour = 7 # 7 AM EST (earlier for pre-work testing)
end_hour = 20 # 8 PM EST (later for evening testing)
}
off_hours = {
min_size = 1 # Keep 1 for overnight tests
max_size = 2
desired_capacity = 1
}
}
# Staging - keep 1 instance on weekends
module "staging_scaling" {
source = "./modules/asg-scheduled-scaling"
asg_name = aws_autoscaling_group.staging.name
timezone_offset = -5
business_hours = {
min_size = 2
max_size = 4
desired_capacity = 2
start_hour = 6 # 6 AM Monday
end_hour = 22 # 10 PM Friday
}
off_hours = {
min_size = 1
max_size = 2
desired_capacity = 1 # Keep 1 on weekends
}
}
📊 Cost Savings Breakdown
Dev Environment (Keep 1 Instance Off-Hours)
Before:
3 × t3.large running 24/7
- 3 × $0.0832/hour × 730 hours = $182/month
- Annual: $2,184
After (3 instances business hours, 1 instance off-hours):
Business hours (50 hrs/week = 217 hrs/month): 3 instances
Off-hours (118 hrs/week = 513 hrs/month): 1 instance
3 × $0.0832 × 217 + 1 × $0.0832 × 513
= $54 + $43 = $97/month
Annual: $1,164
Savings: $85/month = $1,020/year (47% reduction!) 🎉
QA Environment (1 Instance Off-Hours)
Before:
4 × t3.xlarge running 24/7 = $485/month
After:
Business hours (60 hrs/week): 4 instances
Off-hours (108 hrs/week): 1 instance
4 × $0.1664/hour × 260 hrs + 1 × $0.1664/hour × 470 hrs
= $173 + $78 = $251/month
Savings: $234/month = $2,808/year (48% reduction!)
Typical Organization (5 Non-Prod Environments)
2 Dev + 2 QA + 1 Staging = 5 environments
Average cost before: $300/month each = $1,500/month
After scheduled scaling:
- Dev (2): $97/month each = $194
- QA (2): $150/month each = $300
- Staging: $120/month
Total after: $614/month
Savings: $886/month = $10,632/year 💰
💡 Pro Tips
1. Account for Timezone
Scheduled actions use UTC time. Convert from your timezone:
# EST (UTC-5)
start_hour = 8 # 8 AM EST
utc_hour = 8 + 5 = 13 # 1 PM UTC
recurrence = "0 13 * * 1-5" # Use UTC hour in cron
Or use the module which handles conversion automatically!
2. Keep Minimal Capacity for Tests
Don't scale to absolute zero if you have:
- Overnight CI/CD jobs
- Automated testing suites
- Health checks that need to pass
off_hours = {
min_size = 1 # Keep 1 instance
desired_capacity = 1
}
3. Stagger Scale-Up Times
Avoid resource contention by staggering start times:
# Dev starts at 8 AM
recurrence = "0 8 * * 1-5"
# QA starts at 8:15 AM
recurrence = "15 8 * * 1-5"
# Staging starts at 8:30 AM
recurrence = "30 8 * * 1-5"
4. Monitor Scale Events
Set up CloudWatch alarms:
resource "aws_cloudwatch_metric_alarm" "scale_up_success" {
alarm_name = "dev-scale-up-verification"
comparison_operator = "LessThanThreshold"
evaluation_periods = 1
metric_name = "GroupDesiredCapacity"
namespace = "AWS/AutoScaling"
period = 300
statistic = "Average"
threshold = 3
alarm_description = "Dev ASG failed to scale up at 8 AM"
dimensions = {
AutoScalingGroupName = aws_autoscaling_group.dev.name
}
}
5. Add Holidays
Extend weekend shutdown for holidays:
# Thanksgiving week shutdown
resource "aws_autoscaling_schedule" "thanksgiving" {
scheduled_action_name = "thanksgiving-shutdown"
min_size = 0
max_size = 1
desired_capacity = 0
start_time = "2024-11-27T00:00:00Z"
end_time = "2024-12-02T08:00:00Z"
autoscaling_group_name = aws_autoscaling_group.dev.name
}
⚡ Quick Start
# 1. Add scheduled actions to existing ASG
terraform apply
# 2. Verify schedules created
aws autoscaling describe-scheduled-actions \
--auto-scaling-group-name dev-asg
# 3. Test scale-down manually (optional)
aws autoscaling set-desired-capacity \
--auto-scaling-group-name dev-asg \
--desired-capacity 0
# 4. Wait and verify instances terminate
aws ec2 describe-instances \
--filters "Name=tag:aws:autoscaling:groupName,Values=dev-asg" \
--query 'Reservations[].Instances[].State.Name'
# 5. Watch your bill drop next month 🎉
⚠️ Common Gotchas
1. Cron Syntax is UTC, Not Local Time
# WRONG - This is NOT 8 AM EST
recurrence = "0 8 * * 1-5" # This is 8 AM UTC = 3 AM EST
# RIGHT - 8 AM EST = 1 PM UTC
recurrence = "0 13 * * 1-5"
2. Min/Max/Desired Must Be Consistent
# WRONG - Desired > Max
min_size = 0
max_size = 2
desired_capacity = 3 # Error!
# RIGHT
min_size = 0
max_size = 3
desired_capacity = 3
3. Conflicting Schedules
# WRONG - Both run at 8 AM Monday
recurrence = "0 8 * * 1-5" # Weekday scale up
recurrence = "0 8 * * 1" # Monday scale up (conflicts!)
# RIGHT - Monday schedule is redundant, remove it
recurrence = "0 8 * * 1-5" # Covers Monday too
4. Stateful Applications
If your instances store state (databases, sessions):
- Don't scale to 0
- Use lifecycle hooks to drain connections
- Or better: make instances stateless!
🎯 Decision Matrix
| Environment | Recommended Strategy | Expected Savings |
|---|---|---|
| Dev | Keep 1 instance off-hours | 40-50% |
| QA | Keep 1 instance off-hours | 50-60% |
| Staging | Keep 1 instance on weekends | 30-40% |
| Demo | Scale to 1 except during demos | 60-70% |
| Training | Keep 1 instance, scale up for sessions | 70-80% |
📈 Real-World Example
Company: SaaS startup with 3 non-prod environments
Before scheduled scaling:
Dev: 3 × t3.large × 24/7 = $182/month
QA: 4 × t3.xlarge × 24/7 = $485/month
Staging: 2 × t3.large × 24/7 = $121/month
Total: $788/month
Annual: $9,456
After scheduled scaling:
Dev: 3 inst. business hours, 1 off-hours = $97/month
QA: 60 hrs/week, 1 inst off-hours = $251/month
Staging: Weekdays 2 inst., weekends 1 inst. = $120/month
Total: $468/month
Annual: $5,616
Savings: $320/month = $3,840/year (41% reduction!)
Implementation time: 1 hour
Lines of Terraform: ~40
Ongoing maintenance: Zero
🎯 Summary
The Problem:
- Non-prod environments run 24/7
- Developers work 40-50 hours/week
- 70% of capacity is wasted
- Costs accumulate unnecessarily
The Solution:
- Auto Scaling scheduled actions
- 5 lines of Terraform per schedule
- Scale down nights and weekends
- Zero application changes needed
The Result:
- Typical savings: 40-60% on non-prod
- One-time setup, runs forever
- No operational overhead
- Works with existing ASGs
- Keeps minimal capacity for CI/CD
Stop running dev environments at full capacity 24/7 when your team works 9-5. Add scheduled scaling and reclaim those wasted hours. ⏰
Implemented scheduled scaling? How much did you save? Share in the comments! 💬
Follow for more AWS cost optimization with Terraform! ⚡
SOCIAL SHARE CARD GENERATOR