Schrodinger's Elephant
summary
Argues that containerizing Postgres buys portability and CI/CD convenience but exposes real limits: large databases hit the Linux OOM killer, default schedulers misplace heavyweight instances, and spinning replicas for a 6 TB cluster is slow and costly. Recommends a push-based liveness model using a distributed consensus store and tools like Patroni to update status, exposes metrics over REST (not just logs), runs a Kubernetes statefulset plus sidecar powa for per-pod performance, and reports prototypes and gaps (log hooks and centralized search still problematic).
AI-written summary · about
chapters (12)
- 0:00Start
- 1:14Quantum Superposition
- 2:05Why Would You Put Anything in a Box
- 3:53Continuous Integration Continuous Delivery
- 4:57Orchestration
- 5:21Container Orchestration
- 8:42Out of Memory Killer
- 12:20Observability
- 14:47Health Track
- 29:09Replication Status
- 36:17Performance Analysis
- 40:50Why You Should Use Stateful Sets
related talks
54mContainerized Clustered PostgreSQL
30mThe database trends that are transforming your database infrastructure forever
48mWeb scale for the rest of us
18mPower Hour, Managing Databases at Scale – Peter Zaitsev
58mSuper Jumbo Delux
04h 15mTopical Edition on Databases - Track 2
32mContaining an Elephant
34mPanel Discussion: Databases