Automated, Distributed Systems Testing for Kubernetes Controllers
Talkat KubeCon + CloudNativeCon North America 2021 · on Wednesday, 2021-10-13· at 16:30· Petree Hall D + Online·
summary
Sieve is an automated tester that runs controllers inside instrumented kind clusters, learns execution traces, and then replays 21 event sequences to inject timing faults; it needs no source changes and produces self-contained, replayable fault scripts. It actually finds concrete failures — e.g., a PVC resource leak in a Cassandra operator caused by a missed reconciler state (fixed by adding a finalizer) — and reports divergences in resource sets so developers can reproduce and fix safety-critical bugs.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 0:13Kubernetes Controllers
- 1:32Controllers
- 2:19Challenges
- 2:53Problems
- 3:30Solution
- 5:55correctness
- 10:14general purpose checks
- 12:05demo
- 15:48observability gaps
- 23:07other patterns
- 23:38internals
- 24:53analyzer
- 26:08conclusion
related talks
35mIntro: Testing SIG
38mFuzzing and Testing the Security of Kubernetes Admission Controllers
21mTesting Distributed Software on Kubernetes with PowerfulSeal
49mSite Reliability Engineering with Kubernetes
36mA Toolkit for Simulating Kubernetes Scheduling at Scale
32mTestable Kubernetes Operators
33mThe Power of a Unified Testing Approach in Kubernetes
29mDeep Dive: KWOK