Skip to content

Reference

Scope

Look-up facts for Ceph: release and support matrix, feature-by-release table, verified configuration defaults, network ports, CephX capability syntax, security advisories, Rook compatibility, limits, and a hardening checklist. For the "why", see Explanation; for step-by-step tasks, see How-to Guides.

Release and Support Matrix

Ceph ships one named stable release roughly per year (version X.2.Z; X.1.Z are release candidates). The project keeps two stable series active at a time. Dates below come from the upstream doc/releases/releases.yml (checked 2026-09-25).

Release Series First GA Latest point release Upstream EOL Status (2026-09)
Umbrella v21 TBD 21.1.x release candidates TBD In development (RC stage)
Tentacle v20.2 2025-11-18 20.2.4 (2026-08-19) 2027-06-01 (target) Active, recommended for new clusters
Squid v19.2 2024-09-26 19.2.6 (2026-08-19) 2026-10-31 (target) Active, nearing EOL
Reef v18.2 2023-08-07 18.2.8 (2026-03-20) 2026-03-20 (actual) EOL
Quincy v17.2 2022 17.2.9 (2025-05-27) 2025-01-13 (actual) EOL
Pacific v16.2 2021 16.2.15 EOL EOL

Squid EOL date discrepancy

The upstream releases.yml and the Proxmox announcement both give 2026-10-31 as the estimated Squid EOL. At least one secondary source cites 2026-09-19 (unverified). Proxmox's own Ceph Squid repository timeline is tracked in Proxmox Reference. Plan Squid-to-Tentacle upgrades now either way.

Tentacle Point Releases

Version Date Highlights
20.2.0 2025-11-18 FastEC, ISA-L default EC plugin, SMB manager, SeaStore tech preview, mgmt-gateway, certmgr
20.2.1 2026-04-06 EC recovery fixes; FastEC denied for non-4K-aligned chunk sizes; NVMe-oF fast-failover overhaul; RBD EXCLUSIVE_TRANSIENT lock mode
20.2.2 2026-06-16 Rocky Linux 10 packages; MDS and PG log fixes
20.2.3 2026-08-05 New NVMe-oF mgr module (auto-creates .nvmeof pool on fresh installs); RGW ssl_ciphersuites, Kafka mTLS; hardware-monitoring dashboard
20.2.4 2026-08-19 Security release: CVE-2025-30156, CVE-2026-39944, CVE-2026-50152, CVE-2026-54330 (new aes256k CephX key type)

Feature Availability by Release

Feature Introduced Status (Tentacle 20.2) Notes
BlueStore default OSD backend Luminous (12.2) Default osd_objectstore = bluestore
FileStore — Removed Not supported since Reef (18.2)
PG autoscaler on by default Octopus (15.2) Default osd_pool_default_pg_autoscale_mode = on
cephadm orchestrator Octopus (15.2) Recommended installer Containers (Podman/Docker) + systemd
RocksDB LZ4 compression in BlueStore Squid (19.2) Default Part of bluestore_rocksdb_options
RGW User Accounts (IAM APIs) Squid (19.2) GA Tenant-level IAM deprecated in Tentacle
Crimson OSD Squid (19.2) tech preview Tech preview Not for production
SeaStore object store Tentacle (20.2) tech preview Tech preview NVMe-oriented; deployable with Crimson
FastEC (EC optimizations) Tentacle (20.2) GA, opt-in per pool allow_ec_optimizations; cannot be disabled once set
ISA-L default EC plugin Tentacle (20.2) Default for new clusters Upgraded clusters keep Jerasure default
SMB manager module Tentacle (20.2) GA cephadm-only; Samba + CTDB containers
NVMe-oF (NVMe/TCP) gateway cephadm nvmeof service in Reef 18.2.x; gateway groups in Squid Supported; much expanded in Tentacle Multiple namespaces, dashboard, new mgr module (20.2.3)
iSCSI gateway — Maintenance mode since 2022-11 Use NVMe-oF or native clients
mgmt-gateway + oauth2-proxy Tentacle (20.2) GA Single TLS entry point, SSO
certmgr Tentacle (20.2) GA cephadm-internal CA and rotation
Pool data availability score Tentacle (20.2) Tech preview, off by default enable_availability_tracking
mgr/restful, mgr/zabbix — Removed in Tentacle Use dashboard API / Prometheus
aes256k CephX key type 20.2.4 / 19.2.6 GA Fix for CVE-2025-30156

Configuration Defaults

Values checked in the tentacle branch of src/common/options/*.yaml.in (2026-09-25). Values can change between releases; confirm with ceph config help <option>.

Option Default (Tentacle) Meaning
osd_objectstore bluestore Backend for new OSDs
osd_pool_default_size 3 Replicas for new replicated pools
osd_pool_default_min_size 0 (= size - size/2, so 2 for size 3) Minimum replicas to accept I/O
osd_pool_default_pg_num 32 Initial PGs (autoscaler overrides)
osd_pool_default_pg_autoscale_mode on PG autoscaler mode for new pools
mon_target_pg_per_osd 100 (main/Umbrella: 200) Autoscaler target PG replicas per OSD
mon_max_pg_per_osd 250 (main/Umbrella: 500) Hard limit; pool creation refused above it
osd_pool_default_erasure_code_profile plugin=isa technique=reed_sol_van k=2 m=2 Default EC profile on fresh clusters
osd_pool_default_flag_ec_optimizations false Enable FastEC on new EC pools
osd_op_queue mclock_scheduler OSD op scheduler
osd_memory_target 4 GiB Per-OSD memory autotuning target
bluestore_min_alloc_size_hdd / _ssd 4 KiB / 4 KiB Allocation unit
bluestore_compression_mode none Inline data compression off by default
osd_scrub_min_interval 1 day Light scrub interval
osd_deep_scrub_interval 7 days Deep scrub interval
mon_osd_nearfull_ratio / backfillfull / full 0.85 / 0.90 / 0.95 Capacity thresholds
mon_allow_pool_delete false Guard against pool deletion
auth_service_ticket_ttl 1 hour CephX service ticket lifetime
ms_cluster_mode, ms_service_mode, ms_client_mode crc secure msgr2 modes, in preference order
ms_bind_port_min / ms_bind_port_max 6800 / 7568 OSD/MDS/MGR daemon port range

Network Ports

Port Protocol Used by
3300/tcp msgr2 Monitors (default)
6789/tcp msgr1 (legacy) Monitors
6800-7568/tcp msgr2/msgr1 OSD, MDS, MGR daemons (ms_bind_port_min/max)
8443/tcp HTTPS Ceph Dashboard (mgr module, SSL default)
9283/tcp HTTP mgr prometheus module exporter
80 or 443/tcp HTTP(S) RGW deployed by cephadm (default spec)
7480/tcp HTTP RGW Beast frontend default when run manually
4420/tcp NVMe/TCP NVMe-oF gateway listeners (NVMe/TCP IANA port)

Cluster Maps

Map Contents View Command
Monitor map Cluster fsid, monitor addresses, epochs ceph mon dump
OSD map Pools, replica sizes, PG numbers, OSD states ceph osd dump
PG map PG versions, up/acting sets, per-PG state, usage ceph pg dump
CRUSH map Devices, failure-domain hierarchy, rules ceph osd getcrushmap / ceph osd crush dump
MDS (FS) map MDS ranks, metadata pool, active/standby ceph fs dump

Erasure-Code Profiles

Overhead factor is (k+m)/k. Replication 3x has an overhead factor of 3.0 (200% extra).

Profile OSDs needed (min) Failures tolerated Overhead factor Extra raw capacity
2+2 (default profile) 4 2 2.0 100%
3+2 5 2 1.67 67%
4+2 6 2 1.5 50%
8+3 11 3 1.375 37.5%
Replicated size 3 3 2 3.0 200%

FastEC guidance from the upstream EC docs: use a stripe unit of at least 16 KiB with optimizations (default 4 KiB, set at pool creation), keep m <= 3 for block and file workloads, and use the jerasure or isa plugin with reed_sol_van.

CephX Capabilities

Keyword Meaning
allow Precedes access settings. Implies rw for MDS
r Read access
w Write access
x Execute (call class methods, auth operations)
class-read Subset of x. Call class read methods
class-write Subset of x. Call class write methods
* / all Full read, write, execute, and admin permissions
profile rbd Predefined profile for RBD access
profile osd OSD-to-monitor communication
profile mgr Manager daemon
profile mds MDS daemon

Daemon Keyrings

Daemon Keyring Location Required Capabilities
ceph-mon $mon_data/keyring mon 'allow *'
ceph-osd $osd_data/keyring mgr 'allow profile osd' mon 'allow profile osd' osd 'allow *'
ceph-mds $mds_data/keyring mds 'allow' mgr 'allow profile mds' mon 'allow profile mds' osd 'allow rwx'
ceph-mgr $mgr_data/keyring mon 'allow profile mgr' mds 'allow *' osd 'allow *'
radosgw $rgw_data/keyring mon 'allow rwx' osd 'allow rwx'

CephX Settings

Setting Scope Default
auth_cluster_required Daemon-to-daemon authentication cephx
auth_service_required Client-to-daemon authentication cephx
auth_client_required Daemon-to-client authentication cephx
auth_allowed_ciphers (mon map) Key types accepted aes,aes256k after upgrade to 20.2.4/19.2.6
auth_preferred_cipher (mon map) Key type for new keys aes256k on fresh installs; aes on upgraded clusters until changed
auth_service_cipher (mon map) Rotating service key cipher Set to aes256k during migration

CephX Migration Health Checks

Health check Cleared by
AUTH_INSECURE_SERVICE_KEY_TYPE Rotating mon/mgr/osd/mds keys to aes256k
AUTH_INSECURE_SERVICE_TICKETS ceph mon set auth_service_cipher aes256k
AUTH_INSECURE_ROTATING_SERVICE_KEY_TYPE Letting rotating keys expire (or wiping them)
AUTH_INSECURE_KEYS_CREATABLE mon_auth_allow_insecure_key = false or dropping aes from allowed ciphers
AUTH_INSECURE_CLIENT_KEY_TYPE Rotating every client.* key

Security Advisories

Published CVE Severity Summary Fixed in
2026-08-19 CVE-2025-30156 High (CVSS 8.9) CephX authentication bypass (unauthenticated AES-CBC, fixed IV) 20.2.4, 19.2.6
2026-08-19 CVE-2026-39944 High RGW STS session-token bit-flip privilege escalation 20.2.4, 19.2.6
2026-08-19 CVE-2026-50152 High Monitor config-key store unauthorized read 20.2.4, 19.2.6
2026-08-19 CVE-2026-54330 High RGW SigV4 header injection escalation 20.2.4, 19.2.6
2023-02-02 CVE-2023-46159 Medium RGW denial of service See upstream
2021-04-14 CVE-2021-20288 High Unauthorized global_id reuse in CephX See upstream

No fixes for EOL releases

The August 2026 fixes shipped only for Tentacle and Squid. Reef and older are EOL and remain vulnerable. Kernel RBD/CephFS clients need Linux 7.0+ (or a vendor backport, for example CentOS Stream 9/10) to use aes256k keys.

Rook Compatibility

Rook Status (2026-09) Supported Ceph Kubernetes Default Ceph image
v1.21 Beta (v1.21.0-beta.0) Squid, Tentacle v1.32 to v1.37 quay.io/ceph/ceph:v20.2.4
v1.20.7 Current stable (v1.20 released 2026-06) Squid, Tentacle v1.31 to v1.37 quay.io/ceph/ceph:v20.2.4

Rook v1.20 makes the Ceph-CSI operator the only supported way to configure the CSI drivers. Rook v1.21 will require msgr2 by default (kernel 5.11+; opt out with network.connections.requireMsgr2: false). Umbrella (v21) will need allowUnsupported: true on Rook v1.20.

NVMe-oF Gateway Limits

Item Limit (upstream docs)
Gateway groups per cluster 4
Gateways per group 8
HA model Active/standby per namespace, one load-balancing group per gateway
Minimum for HA 2 gateways plus listeners on every gateway
Recommended initiator reconnect window 1800 seconds (connect-all)

Scale and Performance Reference

Dimension Value Source
Largest published test cluster >10,000 OSDs (CERN "Big Bang III", Luminous RC) Ceph blog: New in Luminous, improved scalability
Earlier CERN test 30 PB on 7,200 OSDs (2015) Same
Total capacity Exabyte-scale deployments reported Vendor and community reports (not independently verified)
Objects per PG No hard limit Upstream docs

The following ranges are rough, unsourced estimates kept for orientation only. They have no documented hardware, version, or methodology. Run rados bench and fio on your own hardware for planning.

Operation 3 OSD (HDD) 3 OSD (SSD) 12 OSD (NVMe)
Seq write 300-500 MB/s 1-2 GB/s 5-10 GB/s
Seq read 400-600 MB/s 1.5-3 GB/s 8-15 GB/s
Random 4K write 500-1,000 IOPS 10k-30k IOPS 100k+ IOPS
Random 4K read 1,000-2,000 IOPS 20k-50k IOPS 200k+ IOPS
RBD block size Throughput IOPS Latency (P99)
4K random N/A 10k-50k 1-5 ms
64K sequential 500 MB/s-2 GB/s N/A 2-10 ms
1M sequential 1-5 GB/s N/A 5-20 ms

Hardening Checklist

  • Run Tentacle 20.2.4+ or Squid 19.2.6+ and complete the aes256k CephX key migration.
  • Keep auth_*_required = cephx; never set them to none.
  • Set ms_cluster_mode, ms_service_mode, ms_client_mode to secure where CPU budget allows.
  • Put the cluster (replication) network on an isolated VLAN or fabric.
  • Encrypt OSDs with dm-crypt (encrypted: true in the cephadm OSD spec).
  • Scope client caps to pools/namespaces (profile rbd pool=...); avoid distributing client.admin.
  • Terminate RGW TLS (Beast ssl_port or the cephadm certmgr/ingress) and enforce bucket policies and Block Public Access.
  • Front the Dashboard and monitoring stack with mgmt-gateway plus oauth2-proxy (Tentacle) or restrict access by network.
  • Keep mon_allow_pool_delete = false outside maintenance windows.
  • On Kubernetes, restrict traffic to Ceph ports with NetworkPolicies (Rook ships none by default).

Sources