Kubernetes: Trials and triumphs
summary
Three years of running CockroachDB on Kubernetes produces concrete outcomes: an in-house K8s operator (shipped end of 2021) that automates node decommissioning/recommissioning via Cockroach gRPC, multi-region multi-cloud deployments using Istio-based service discovery, and a smooth cloud migration with canary traffic. Disk resizing now uses a PVC autoscaler to increase sizes in minutes and shrink by copying (7 minutes–4 hours per node), vertical CPU autoscaling based on 7‑day p50/p90 reduces waste, and spot nodes save cost in staging but remain risky for production.
AI-written summary · about
chapters (20)
- 0:00What is City Storage Systems and how do they use CockroachDB?
- 1:08Why are they running CockroachDB?
- 2:19The benefits of Kubernetes
- 3:12Provisioning a CockroachDB cluster on Kubernetes
- 4:16Day 2 Operations - Scaling disks
- 6:35Day 2 Operations - Scale in horizontally
- 7:53Day 2 Operations - Scale out horizontally
- 9:18Lessons from the journey so far
- 10:44Kubernetes operator pattern
- 11:39Writing a CockroachDB operator
- 13:13Scaling all the things - Adding and removing nodes
- 15:22Scaling all the things - Disks Take 2
- 16:09Lessons learned from scaling
- 17:19Multi-region deployments and disaster recovery
- 19:28Cloud migration with CockroachDB and the Kubernetes Operator
- 21:47Lessons learned from building with a cloud-native mindset
- 23:40Vertical auto scaling - CPU recommendations
- 25:48Scaling Disks - Take 3 - Scale up
- 28:34CockroachDB on Spotnodes
- 30:59Key takeaways
related talks
33mWhat We’ve Learned Building a Multi-Region DBaaS on Kubernetes
31mFuture-proofing operational data at scale
29mHow City Storage Systems designed its Data Infrastructure to minimize toil
26mChallenges of Managing Stateful Applications in Kubernetes
22mRunning distributed stateful system on Kubernetes - Beyond Statefulsets
36mKubernetes Runs Anywhere, but Does your Data?
48mBuilding mission-critical applications with CockroachDB
45mLearning How to Pronounce Kubernetes to Production in 3 Months!
