Active/Active NFS Serving over CephFS

on YouTube ↗30m

" summary

Demonstrates an active/active NFS over CephFS setup that uses a single RADOS grace object and per-node per-epoch RADOS client DBs so multiple NFS-Ganesha heads coordinate recovery without a central controller. Adds a new CLI ganesha-rados-grace, stores node need/enforcing flags in the Ceph OMap, and runs a three-head Kubernetes/rook demo (nodes A/B/C) with fio IO; intended failover shows IO pause then resume, and grace periods auto-lift in seconds once clients reclaim.

AI-written summary · about

" chapters (9)

1. 0:00Introduction to Ceph NFS
2. 2:19Scaling challenges
3. 4:52Handling NFS recovery
4. 8:33Cluster wide grace period
5. 12:03Implementing shared state
6. 15:10Management and tools
7. 17:39Configuration and demonstration
8. 20:46Future directions and QA
9. 28:21Conclusion

" related talks

• 16mTips and Tricks for Managing and Administering Ceph ClustersMichael Hackett · Linux.conf.au 2021 Systems Administration Miniconf
• 21mKeynote: Ceph Developers AMA: Meet the Minds Behind Open Source CephMatt Benjamin, Daniel Gyovich · Cephalocon 2025
• 19mMicroCeph: Simplifying StoragePeter Sabaini · Open Source Summit Europe 2024
• 42m"cgroup2 Resource Control strategies w/ resctl-demo"Tejun Heo · Linux.conf.au 2021 Kernel Miniconf
• 32mBringing up FUSE mounts C/R supportAlexander Mikhalitsyn · Linux Plumbers Conference 2022
• 42mcgroup2 Resource Control strategies w/ resctl-demoTejun Heo · linux.conf.au 2021
• 20mDeep Dive into Ceph Block StorageMahati Chamarthy · Vault '19
• 31mUsing kAFS on Linux for Network Home DirectoriesJonathan Billings · Vault '20