Glossary
Short definitions of the terms used throughout the planner.
- backfill
- The data movement Ceph performs to bring an OSD in line with where CRUSH says data should live, for example after recovery or a topology change.
- backfillfull
- The 90 % OSD fill ratio (mon_osd_backfillfull_ratio) above which Ceph refuses to backfill more data onto that OSD.
- CephFS
- Ceph’s POSIX-compatible distributed filesystem, built on top of RADOS pools.
- CRUSH
- The algorithm Ceph uses to compute where every placement group is stored, without a central lookup table.
- device class
- The drive tier Ceph assigns an OSD to (hdd, ssd or nvme), used to steer placement and estimate recovery speed.
- erasure coding
- A redundancy scheme that splits an object into k data chunks plus m coding chunks instead of keeping full copies.
- Learn more
- failure domain
- The level of the cluster topology — host by default, or OSD — within which CRUSH avoids placing more than one copy of the same object.
- Learn more
- full ratio
- The 95 % OSD fill ratio (mon_osd_full_ratio) above which Ceph blocks writes entirely.
- m
- The number of coding chunks computed alongside the data chunks of an erasure coded pool.
- Learn more
- MGR
- A Ceph manager: provides cluster statistics, the dashboard and management modules, usually colocated with a monitor.
- min_size
- The minimum number of reachable copies or chunks a pool needs before Ceph accepts writes.
- Learn more
- MON
- A Ceph monitor: keeps the authoritative map of the cluster; Proxmox recommends three for high availability.
- nearfull
- The 85 % OSD fill ratio (mon_osd_nearfull_ratio) at which Ceph starts raising a health warning.
- Learn more
- OSD
- An Object Storage Daemon — the process, usually backed by one drive, that actually stores data on a Ceph node.
- pg_autoscale_mode
- The per-pool setting that lets the PG autoscaler adjust pg_num automatically instead of leaving it fixed.
- Learn more
- RBD
- RADOS Block Device — the Ceph pool application used for Proxmox VM and container disk images.
- recovery
- Ceph rebuilding missing copies or chunks onto surviving OSDs after a failure.
- target_size_ratio
- The expected share of cluster capacity a pool will use, fed to the PG autoscaler alongside pg_autoscale_mode.
- Learn more