Making CephFS (Much More) Great!
summary
Session reports that multi‑MDS recovery remains fragile but offers concrete fixes: an option to bring whole FS down for safe upgrades, a proposed CLI wrapper to gate dangerous commands, and a request tracker for limiting the CephFS cloning queue (CSI clones currently cap at four). Pinning modes include manual, distributed, ephemeral and random pinning at 0.1% as a fallback. Performance work includes journaling improvements, kernel client read‑ahead patches, and a new cap acquisition throttle (suggested defaults: 50k threshold, decay ~10s).
AI-written summary · about
chapters (15)
- 0:00Introduction and setup
- 1:05Multi-cluster recovery challenges
- 3:33Improving MDS failure recovery
- 7:46SMB gateway and CephFS integration
- 10:35Multi-active MDS upgrades
- 15:55Pinning and subvolume group defaults
- 17:44CephFS snapshot and clone limits
- 19:07Ephemeral pinning strategies
- 23:16Kernel client issues and debugging
- 29:26Kernel client read-ahead bug
- 30:50Snapshot deletion scalability
- 31:57Client-side metrics and observability
- 34:53CephFS-top and future dashboards
- 36:01Extended attributes for backups
- 37:06Closing remarks and thanks
related talks
34mWhat are “caps”? (And Why Won’t my Client Drop Them?)
19mKVM/ARM at Scale: Improvements to the MMU in the Face of Hardware Growing Pains
44mUserspace Assisted Scheduling via Sc...
28mExploring an Architecture-neutral KVM MMU
45mConcurrency Fount: a Pace-Regulated Supply of Newly Spawned Processes
22mBeyond Boolean: Solving Hard Problems in Feature Flag Architecture at Scale
15mDesigning a Read-Intensive Architecture for Information Systems
27mInsights on the Re-Architecting to a Single Codebase, Running on Different Hardware...