Why it matters
A working-hours schedule covers about a third of the week, so an always-on test instance can cost close to three times what it needs to.How nOps finds it
nOps reads the last 14 complete days of on-demand EC2 billing. An instance is flagged when the environment nOps assigns to it is development, staging, QA, test, sandbox, UAT or demo, it was billed on all 14 days, it ran on average at least 20 hours a day, and it ran at least 20 hours on at least 3 of the 4 weekend days. Instances in Auto Scaling groups, EKS or Karpenter, and instances mostly covered by Reserved Instances or Savings Plans, are left out. The saving is the last 7 days of on-demand cost, scaled to a month, times the share of the week outside 60 hours (about 64%). Savings under $5 a month are not shown.Before you start
Agreement from the instance owners on working hours, and permission to start and stop instances.How to implement
- Agree a schedule with the team (for example 07:00 to 19:00 on weekdays in their time zone).
- Set it up with Instance Scheduler on AWS, or with two EventBridge Scheduler rules that call
ec2:StopInstancesandec2:StartInstances. - Tag the instances so the scheduler picks them up, and make sure the instances can stop cleanly (no instance store data to keep).
- Check after a week that they stop and start as expected.
Risks and considerations
- EBS volumes are still billed while instances are stopped.
- Data on instance store volumes is lost on stop.
- Public IPv4 addresses that are not Elastic IPs change on restart.
- To roll back, remove the schedule tag or disable the scheduler.
FAQs
Why are Auto Scaling and Kubernetes nodes left out?
Why are Auto Scaling and Kubernetes nodes left out?
Their capacity is managed by the group or cluster. Stopping a node by hand only makes it launch another one. Schedule the group or cluster instead.