Autoscaling can scale a failure too
Automatic scaling responds to load without asking why the load exists.
An attack, a retry storm or a runaway loop all look like demand, so the platform obligingly provides more capacity and a larger bill. It can also scale a faulty version faster than anybody notices. Limits and alerting on the scaling itself are what stop a resilience feature amplifying the problem.
More on Cloud resilience
- Multi-region is not the same as multi-providerTwo towns, one mains
- Managed services move responsibility, not all riskThe van takes the machines
- Private networking does not automatically mean trustedNobody asks twice
- Egress is part of the cloud security boundaryWatch the loading bay
- Snapshots can copy sensitive data silentlyThe copy keeps nothing
- Serverless still has an attack surfaceThe box went, the openings stayed