Geographical Redundancy with rbd-mirror: Best...
summary
rbd-mirror (introduced in Jewel) replicates RBD by consuming a client-side rbd journal and replaying it to a remote image; it supports one-way and two-way mirroring with explicit promote/demote operations and per-pool or per-image enablement (rbd feature enable, rbd mirror pool enable, rbd mirror image enable). Nautilus makes daemons stateless and centralizes key/config handling; automation exists via ceph-ansible, a new Juju charm, and partial support in Rook. Benchmarks show reads barely change but journaling dramatically hurts small writes — roughly a 10x overhead for 4k writes — while mirroring itself adds little extra cost.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 1:31Replication in a set cluster
- 6:01RBD journaling
- 7:28Enable journaling
- 9:36RBDmirror
- 11:24Oneway mirroring
- 14:29Setting up rbdmirror
- 17:23How to start rbdmirror
- 19:35Connecting peers
- 22:32Deployment automation facilities
- 25:11Performance implications
- 29:55Benchmarks
- 31:53Integration with Cloud
- 34:33Checklist
related talks
12mUsing rbd Migration Tooling in Our Ceph Operations
19mElegantly Dumb, Powerfully Backed Mirroring: Community over Complexity
32mSchroedingers Backup
36mRook: Intro and Deep Dive With Ceph Storage
31mHow to Use Mirroring and Caching to Optimize your Container Registry
37mCache Me If You Can: P2P Image Sharing in Kubernetes with Spegel
25mData protection design for business continuity
29mReliable and Cost-Effective: Open Storage Strategies for Kubernetes