New Read Balancer in Ceph
summary
A new read balancer for Reef adds a rebalance score shown by ceph osd pool ls detail and an offline CLI path using the OSD map tool that outputs pg primary changes (pg_upmap_primary) into out.txt. It targets replicated pools only, runs metadata-only swaps with no data movement, and on a demo pool reduces score from 1.63 to 1.00 after nine proposed changes. It’s greedy, per-pool, assumes homogeneous OSDs, and can apply commands or run dry‑runs.
AI-written summary · about
chapters (22)
- 0:00Introduction to new read balancer
- 1:25Capacity vs read balancing differences
- 4:18Real-life balancing example
- 8:12Read balancing before Reef
- 9:36Rebalancer use cases in Reef
- 10:18New metrics and commands
- 13:50Primary reassignment process
- 16:44High-level rebalancer design
- 17:06Demo setup and initial checks
- 18:10Dry run with OSD map tool
- 21:20Applying changes to cluster
- 22:23Future plans and improvements
- 23:04Performance testing progress
- 24:09Q&A on EC pools and commands
- 25:54Single command vs incremental changes
- 27:22Homogeneous vs heterogeneous clusters
- 28:49Interaction with auto scaler
- 30:13Pool-specific rebalancing scope
- 31:38Comparison with OSD map tool
- 33:46Network latency considerations
- 35:33Integration into manager module
- 36:15Call for early adopters
related talks
27mFrom Chaos To Clarity: Systematic Ceph Cluster Analysis for Rapid Troublesho...
35mMaking Ceph Fast in the Face of Failure
33mBalancing data in Apache Ozone
25mSupporting 3 Availability Zones Stretch Cluster
38mEfficient Ceph Performance Troubleshooting in Production Using eBPF
25mExplicitly Supporting Stretch Clusters in Ceph
49mIntro to Ceph, the Distributed Storage System
20mRemote Replication in MicroCeph: RBD and Beyond