DSN

2024

Mutiny! How does Kubernetes fail, and what can we do about it?

Fault-Error-Failure chain:

• Fault: a static defect in software
• Error: an incorrect internal state
• Failure: external, incorrect behavior

What kind of failures?

• misconfigurations
• communication errors
• other underlying errors like OS

Etcd alterations can recreate a majority (54/81) of real-world failures analyzed.

The core idea is that given a injected fault, why the fault is propagated to the whole system.