Finance Wants Everything on Spot Instances. Including the Database.
Finance wants the whole fleet, primary database included, moved to spot instances for 70 percent savings. Defend what actually belongs on spot and what a reclaimed primary really costs.
Answer key reviewed by Branko Petric
Read first: AWS Databases: RDS and DynamoDB
Your decision
Finance has mandated a cloud cost cut and points at spot instances: 70 percent cheaper, and 'the cloud provider says interruptions are rare'. They want everything moved, including the primary Postgres server. Do you push back, and where do you draw the line? Defend your call.
The situation
After a rough quarter, finance mandates a 40 percent cut to the cloud bill and circulates a consultant's slide: "spot instances, up to 70 percent off, interruptions under 5 percent". The CTO forwards it with "let's do this everywhere". "Everywhere" includes the web tier, CI runners, async workers, and the primary Postgres instance with its two read replicas.
Context
The stack runs on AWS: about 40 on-demand EC2 instances, most of the bill in the web/worker fleets and the databases. Postgres replicates asynchronously to the replicas, typically 1-3 seconds behind, with nightly snapshots. The web tier sits behind an ALB but has no spot interruption handling today, and the worker jobs are not checkpointed. In the planning meeting the platform lead says: "an interruption is basically a reboot, and we survive reboots. I say we flip the whole account to spot and bank the savings."
Related challenges
- The NAT Gateway Bill Exploded. Add More Gateways?
NAT Gateway data processing is now your biggest line item. A teammate thinks it is just bandwidth. Defend where the cost really comes from.
- Your S3 Website Returns 403. Make the Bucket Public?
Objects in S3 return 403 and a teammate wants to turn off Block Public Access to fix it fast. Defend the right way to grant access.
- The t3 Instance Got Slow. Just Size It Up?
An app on a t3 instance crawls under steady load and a teammate wants a bigger instance. Defend what is actually throttling it.
- Access Keys Are Hardcoded on the EC2 Box. Just Rotate Them?
An app on EC2 authenticates to AWS with long-lived IAM user keys in a config file. One leaked. Defend the real fix.