Whatever Can Go Wrong, Will Go Wrong – Rook/Ceph and Storage Failures
Talkat KubeCon + CloudNativeCon North America 2020 Virtual · on Thursday, 2020-11-19· at 16:50· Intrado Virtual Event Platform,·
summary
Claims that Rook/Ceph heals fast are tested with a 3-node AWS cluster with 6 OSDs and a load generator called Sherlock driving 12 MySQL pods with 100GB PVCs. Deleting an OSD pod triggers immediate re-creation, a brief IOPS/latency blip, then Ceph marks OSD down, peers, and returns placement groups to active+clean. Rebooting a node takes a third of storage offline, causes peering and data redistribution, and likewise recovers with short visible impact; RBD mirroring and RGW notes appear (GA timelines discussed).
AI-written summary · about
chapters (14)
- 0:00Intro
- 0:25Agenda
- 0:59Storage Overview
- 2:41Storage Terms
- 4:49Process Terms
- 6:59MTTR
- 9:15MTBF
- 10:41RTO
- 12:07RookCeph
- 12:46Ceph Components
- 14:47Rook Components
- 16:44RookCeph Components
- 20:47First Demo
- 29:33Summary
related talks
35mBest Practices of Production-Grade Rook/Ceph Cluster
31mRook: Intro and Deep Dive With Ceph
31mRook: Intro and Deep Dive With Ceph Storage
35mDisaster Recovery of Stateful Applications in a Multi-Cluster Environment
27mRook: Intro and Ceph Deep Dive
01h 21mHands On with Rook: Ceph & Kubernetes
19mUsing Rook in Production on National Research Platform
35mCritical Troubleshooting Tools for Rook Clusters