Rook Deployed Scalable NFS Clusters Exporting CephFS
summary
Rook and Kubernetes deploy NFS-Ganesha pods that export CephFS; Nautilus adds a RADOS-backed coordination store so RADOS epoch coordination keeps atomic epoch records and per-server flags, enabling active‑active clusters instead of passive failover. libs cephfs implements session-kill/reclaim interfaces to avoid MDS timeout races (MDS default 60s, NFS grace ~90–120s). The system supports per-subvolume scaling, IP migration via Kubernetes, and containerized failover without pacemaker; a working Nautilus prototype exists with minor cleanup remaining.
AI-written summary · about
chapters (13)
- 0:00Intro
- 3:16Why Would You Need a Ceph/NFS Gateway?
- 5:24Active/Passive Deployments
- 6:46Goal: Active/Active Deployment
- 7:25Goals and Requirements
- 8:41Ganesha NFS Server
- 10:36NFS Protocol
- 12:30NFS Client Recovery
- 15:10NFS Server Reboot Epochs
- 16:16New in Ganesha: Coordinating Grace Periods
- 18:28Challenge: Layering NFS over a Cluster FS
- 21:10New In Nautilus: Client Reclaim of Sessions
- 26:19Tying it all together with Rook and Kubernetes
related talks
37mRook Deployed Scalable NFS Clusters Exporting CephFS
37mRook Deployed Scalable NFS Clusters Exporting CephFS
29mSimplifying Storage in Kubernetes with Rook-Ceph
35mDistributed storage with Rook
35mBest Practices of Production-Grade Rook/Ceph Cluster
31mRook: Intro and Deep Dive With Ceph Storage
27mRook: Intro and Ceph Deep Dive
12mRook: Why Would You Ever Deploy Ceph Inside Kubernetes?