Will your PostgreSQL operator crack under chaos?
summary
Chaos tests use Litmus to kill pods, restart primaries and replicas, force PVC deletion, and run a client app that inserts incremental numbers to measure select/insert rates and detect lost rows. Two operators are compared: CloudNativePG and Patroni-based operator (Crunchy/PGO). Both handle simple failovers with minimal client downtime, but aggressive PVC deletions cause CloudNativePG to experience multi-experiment downtime and observed switchovers lose about 2–3 inserts unless synchronous_commit is enabled. Litmus reliability sometimes requires manual scripting.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 4:41PostgreSQL operator choice
- 8:58How PostgreSQL operators work
- 13:40Litmus for chaos testing
- 15:02Chaos testing scenarios
- 22:50PostgreSQL operator experiments
- 27:29Lessons with Litmus
- 29:57Future plans
- 30:18Q&A
related talks
18mChaos testing of a Postgres cluster on Kubernetes
28mLightning Talk: Cloud Native PostgreSQL - Running PostgreSQL on Kubernetes
30mRunning PostgreSQL on Kubernetes in Production: Day 2 Operations with CloudNativePG
50mDemystifying Kubernetes for Postgres DBAs; A Guide to Operators
34mRun PostgreSQL:The Kubernetes way
05mFive Minutes by CloudNativePG River
27mFrom Zero to Hero: Scaling Postgres in Kubernetes Using the Power of...
43mA Kubernetes Operator for PostgreSQL - Architecture and Design