Spot instances on EKS: discount and disruption in one package

On Amazon EC2, you can pay for servers the predictable way — reserved instances or savings plans mean committing ahead of time, typically for a year, with part or all of the cost prepaid. Spot instances trade that commitment for a much deeper discount, and according to Blake Stoddard, that trade works for most workloads once you pair it with auto-scaling groups:

Spot instances give us the ability to get the compute we need at an even deeper discount than a 1-year RI or savings plan rate would, without the same commitment. Combine the price with seamless integration with auto-scaling groups and it makes them a no-brainer for most of our workloads.

The big catch with spot instances? AWS can take them back from you with a two-minute notice.

That two-minute reclamation window explains the price. You request capacity and get it, but the instance can disappear at any moment, so cheap compute comes with the assumption that it will eventually be revoked.

Not everything belongs on spot

Because of that reclaim risk, spot capacity is not a fit for every component. Workloads that require high availability need to stay off spot, and exactly which ones depends on what you are running. That makes spot adoption a classification exercise before it is a pricing exercise: determine what can tolerate interruption and what cannot.

Building for resilience rather than around it

The design constraint is also the benefit. If an application continues serving reliably while the servers underneath it can be taken away at any time, it rests on a genuinely resilient foundation. That property — fault-tolerant architecture as the default — is arguably worth more than the cost savings, because it forces you to assume failure instead of hoping to avoid it.