Per-Subvolume IO Metrics in CephFS
summary
Per-subvolume IO metrics land inside the MDS using a client-side recorder that tags ops when MDS metadata marks a path as a subvolume, packs each op into a 64-bit packed op record (1 bit R/W, 32-bit latency μs, 31-bit payload), clients periodically send dumps to their MDS, and all servers forward to rank 0 aggregation where a 30-second sliding window keeps live IOPS, throughput and latency and prunes inactive subvolumes; metrics are exposed over the admin socket and exported as Prometheus JSON via admin socket.
AI-written summary · about
chapters (7)
- 0:00Introduction and motivation
- 2:31Requirements for metrics
- 7:31High-level architecture
- 10:03Implementation details
- 12:12Metric exposure and examples
- 15:23Future work and anomaly detection
- 20:46Q&A session
related talks
33mMonitoring Ceph with Prometheus
31mLong-Term Metrics with M3 and Prometheus
39mUsing EBPF to Develop a Powerful Tool for Diagnosing Ceph Performance Issues
34mDay 2 Operations : Make Friends with Your Ceph Cluster
31mMonitoring and Centralized Logging in Ceph
28mCommon Support Issues And How To Troubleshoot Them
37mM3 and Prometheus, Monitoring at Planet Scale for Everyone
44mCeph Crash Telemetry - Observability in Action