MSR(Mutli-Step Retry): An Generalization of CRUSH Allowing Multiple OSDs Per Failure Domain- S. Just @Cephstorage
MSR(Mutli-Step Retry): An Generalization of CRUSH Allowing Multiple OSDs Per Failure Domain- S. Just  @Cephstorage
Uploaded November 2025 | Updated September 2026, 2 weeks ago
MSR(Mutli-Step Retry): An Generalization of CRUSH Allowing Multiple OSDs Per Failure Domain - Sam Just, IBM

There are use cases where one might, for example, want to spread an 8+6 erasure coded pool such that no host (or rack) has more than 4 shards. Existing CRUSH rules struggle with this because CHOOSELEAF is the only way to allow an out OSD to be mapped to another failure domain, but CHOOSELEAF does not allow the placement of more than one OSD per failure domain. MSR rules generalize the CRUSH algorithm to allow retrying the full sequence of selections while still respecting placement limitations. This talk will describe the algorithm, implementation, and use cases.
MSR(Mutli-Step Retry): An Generalization of CRUSH Allowing Multiple OSDs Per Failure Domain- S. JustEnterprise-Ready Ceph: Simplifying Storage Management, Operations, and Deployment- Steven UmbehockerCuttle Out the Middleman Direct Shard Access in Erasure Coded Poolsmp4Crimson/SeaStore for Ceph Developer Summit - VampireState of the Cephalopod - India 2026Ceph User + Dev Monthly - August 2026Keynote: IBM’s Commitment, Investment, and Direction for Ceph in 2026 - Gilad Sade & Neha OjhaDelivering Enterprise-Grade Object Storage: Bloomberg’s IAM Journey in a Multi-Tenant Arch... S. GaoReclaim CephFS Capacity Through Pool Tiering + MigrationCeph RGW Refactoring Meeting 2025-12-10HPC TriggersNVMe oFgateway for Ceph Developer Summit - Vampire
Ceph |

MSR(Mutli-Step Retry): An Generalization of CRUSH Allowing Multiple OSDs Per Failure Domain- S. Just

SHARE TO X SHARE TO REDDIT SHARE TO FACEBOOK WALLPAPER