What are “caps”? (And Why Won’t my Client Drop Them?)
summary
CephFS uses metadata caps (not CephX keys) that the MDS issues—examples: pin cap, auth cap, link cap, file cap—each with bits like shared/exclusive, read/write, cache/buffer; file caps include size and mtime. The MDS revokes caps and waits for affirmative client replies; buffered writes require flushing to OSDs, which can block others and trigger health alerts (default timeout ~60s). MDS cache must often exceed client caches. There is a libcephfs delegation API used by NFS Ganesha; no byte-range write caps today.
AI-written summary · about
chapters (11)
- 0:00Introduction to CephFS caps
- 1:21Coherency and metadata locks
- 4:21Types of capabilities
- 6:31Detailed capability breakdown
- 10:08MDS role and lock management
- 13:01Cap revocation and workflows
- 16:04Consequences of the cap system
- 18:32Health warnings and troubleshooting
- 21:59Programming with caps
- 25:30NFS delegation integration
- 28:03Q&A and concluding remarks
related talks
37mMaking CephFS (Much More) Great!
39mOptimizing CephFS with Combining MDS QoS Scheduling and Static-Dynamic Subtree Partitioning
27mThe Full Picture: Visualizing Service "Fullness" To Rethink Saturation Prevention
45mConcurrency Fount: a Pace-Regulated Supply of Newly Spawned Processes
34mDemystifying Kubernetes Resource Management
15mDesigning a Read-Intensive Architecture for Information Systems
32mKubernetes Namespaces Unleashed: Unlocking the Full Potential of You...
23mMCP = Mega Context Problem