Containers on Spot: Is the 90% Discount Worth the Risk?

You're running containers in the cloud, and someone tells you that switching to spot instances can cut your compute bill by up to 90%. But then they mutter something about 'evictions' and 'two-minute warnings.' Is it actually worth it? The short answer: yes, for the right workloads, but only if you design for interruption from the start. Here's how to think about it.

What's the Real Discount?

Let's get the numbers straight. AWS EC2 Spot Instances offer up to a 90% discount compared to On-Demand pricing (AWS EC2 Spot). That's not a typo. For a containerized workload that can tolerate interruptions, the savings are enormous. But that discount comes with a catch: AWS can reclaim the instance with only a two-minute warning (AWS Documentation). Azure and Google have similar programs—Azure Spot VMs give a significant saving but no SLA, with eviction notices as short as 30 seconds (Azure Spot VMs). Google Cloud Spot VMs are up to 91% cheaper than standard on-demand pricing, and they guarantee at least a 60% discount, with up to 30 seconds' notice before preemption (Google Cloud Spot VMs).

Why Containers Are a Natural Fit for Spot

Containers are stateless by design. You can scale them out, kill them, and restart them elsewhere without missing a beat. That's exactly the kind of workload spot instances are built for. AWS specifically lists containerized workloads as a good fit for Spot (AWS EC2 Spot). The same goes for CI/CD, big data, and web servers—all of which are commonly containerized. So if you're already running Kubernetes on EKS, or ECS, or even plain Docker on EC2, you can shift a portion of your cluster to spot nodes and start saving immediately.

The Eviction Problem

But let's not sugarcoat it. Evictions happen. When AWS needs the capacity back, it gives you a two-minute heads-up. Azure gives you 30 seconds. Google gives you 30 seconds. In that time, your containers need to shut down gracefully, save any state, and move elsewhere. If you haven't built for that, you'll lose work and frustrate users. The key is to design your containerized apps to be resumable. That means using external storage for any persistent data, making sure your tasks can be retried, and having a controller that automatically replaces evicted pods.

A Real-World Example

Imagine you're running a batch processing pipeline that reads from a queue, crunches data, and writes results to S3. Each job takes about 10 minutes. You could run this on a c5.xlarge On-Demand instance for $0.17/hour (AWS EC2 Price List API). But if you switch to Spot, you might pay only $0.017/hour—a 90% savings. Over a month of continuous processing, that's a drop from $122.40 to $12.24. Even if you have to retry a few jobs due to evictions, you're still way ahead. But if you're running a customer-facing API with strict latency requirements, spot is a no-go. The risk of a two-minute shutdown is too high.

A Hybrid Approach: The Best of Both Worlds

You don't have to choose all or nothing. A smart strategy is to run your critical, stateful workloads on On-Demand or Reserved Instances, and use Spot for the burstable, fault-tolerant parts. For example, in a Kubernetes cluster, you can have a node group of On-Demand instances for your main services, and a separate node group of Spot instances for your batch jobs, CI runners, or any workload that can handle interruption. AWS even has a feature called EC2 Auto Scaling that can combine purchase options and instance types to optimize performance and cost (AWS EC2 Auto Scaling). You can also use Spot for dev/test environments, where a little downtime is acceptable.

The Verdict

So, is the 90% discount worth the risk? For containerized workloads that are stateless and fault-tolerant, absolutely. The savings are too big to ignore, and the operational overhead of handling evictions is manageable with modern orchestration tools. But for anything that requires high availability, don't even think about it. The most important thing to remember: spot instances are a tool, not a silver bullet. Design your containers to be interruptible, and you can cut your compute bill dramatically. Ignore the risk, and you'll learn the hard way.

Sources

• AWS EC2 Spot - https://aws.amazon.com/ec2/spot/
• AWS Documentation - https://docs.aws.amazon.com/ec2/
• Azure Spot VMs - https://learn.microsoft.com/en-us/azure/virtual-machines/spot-vms
• Google Cloud Spot VMs - https://cloud.google.com/compute/docs/instances/spot
• AWS EC2 Price List API - https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonEC2/current/us-east-1/index.json
• AWS EC2 Auto Scaling - https://aws.amazon.com/ec2/autoscaling/