Ceph Squid reaches end of life today, just as Umbrella 21.2.0 ships
On 31 October 2026, Ceph Squid (19.x) reaches its estimated end of life, two days after Umbrella 21.2.0 — the 21st stable release — shipped. The migration path is not direct: Ceph allows only one major upgrade at a time, and Umbrella moves to a Rocky Linux 10 base that requires x86-64-v3 CPUs.
31 October 2026. Ceph Squid (19.x) reaches its estimated end of life. 28 October 2026. Version 21.2.0, codenamed Umbrella, ships — the project’s 21st stable release. Between the two sits a constraint that shapes every migration: Ceph allows only one major upgrade at a time, and Umbrella changes the underlying system base.
Squid stops today, Tentacle remains the way out
The active-releases table in the official documentation is unambiguous. Squid (19.x), released 26 September 2024 and pushed to 19.2.7 on 12 October 2026, has its estimated end of life set for 31 October 2026 — today. From here on, no more security fixes or backports for the branch.
Tentacle (20.x), released 18 November 2025 and brought to 20.2.4 on 19 August 2026, remains the reference stable branch, with end of life on 1 June 2027. Umbrella (21.x), freshly shipped, runs until 28 October 2028. The mechanics are therefore clear: a cluster on Squid must pass through Tentacle first, and only then — if it wants to — through Umbrella. There is no Squid → Umbrella shortcut: Ceph enforces one major version at a time, and that is a design rule, not a team preference.
That rule has an immediate operational consequence. Waiting too long on Squid means accumulating two sequential version jumps, each with its own health checks, scrubs and daemon failovers. The release of Umbrella does not shorten that path; it makes waiting more expensive.
Umbrella 21.2.0: what changes under the hood
Umbrella is not a cosmetic release. The release notes list changes that touch the core of storage.
First, optimized erasure coding (“Fast EC”), a new I/O path that includes EC direct reads, CLS support for EC reads, OMAP support and accelerated CRC checking for data and first parity. For clusters that encode large volumes of data with EC, it is a potential performance win — and one more reason to test before rolling out.
Second, CephX authentication hardening. Umbrella introduces the AES256KRB5 cipher as the new default, replacing weaker legacy key types, plus online key rotation for mon, OSD and MDS daemons during upgrades. The release also fixes CVE-2025-30156, CVE-2026-39944, CVE-2026-50152 and CVE-2026-54330 — including an AES-CBC misuse enabling an authentication bypass. For an exposed cluster, that is the strongest security argument in the batch.
Finally, broader changes: the PG autoscaler raises its default ceilings (mon_target_pg_per_osd from 100 to 200, mon_max_pg_per_osd from 250 to 500), which can lead existing pools to split into more PGs after upgrade; Crimson/Seastore keeps maturing with EC support; the RGW gains object deduplication (tech preview) and AES-256-GCM encryption; and the dashboard migrates to the IBM Carbon design system.
Two operational details deserve a warning. First, because the PG autoscaler defaults rise, existing pools may split into more PGs after upgrade — harmless for a healthy cluster, but worth watching on clusters with many small OSDs, where the guidance is to pin the old values before upgrading. Second, Umbrella continues the long march toward Crimson/Seastore, the next-generation OSD, now with erasure-coding support. None of it is ready to bet production on; it is the direction of travel, and it is why staying current matters more than the headline features.
The mechanics of a Ceph major upgrade, in practice
Concretely, a major upgrade on Ceph is driven through cephadm: you update the monitor and manager daemons first, then the OSDs, then the MDS and RGW, letting the cluster settle between each step. Ceph orchestrates this rolling upgrade without interrupting service — data keeps being served during the migration — but every step assumes a healthy cluster to begin with. That is why the documentation always gives the same instruction: check ceph -s, resolve warnings, then upgrade one major version at a time, never skipping a rung and never attempting to go back down. A Squid cluster aiming for Umbrella must therefore pass through Tentacle, let the cluster settle, then continue. It is a two-leg marathon, not a sprint.
The hidden cliff: Rocky Linux 10 and x86-64-v3
The detail most operators will discover at upgrade time is not a feature — it is the base. Umbrella container images are based on Rocky Linux 10 (el10), which requires a CPU supporting the x86-64-v3 microarchitecture level. Hosts with older processors will not be able to run the containerized daemons of Umbrella.
In plain terms: a Squid cluster running on older hardware can move to Tentacle without trouble — Tentacle stays on an older base — but will hit the x86-64-v3 wall the day it tries to step up to Umbrella. That is a check to do before planning, not during the maintenance window.
The second requirement is softer: Python 3.9 or newer is required on all hosts for cephadm. Here too, a host pinned to an old distribution can block the upgrade for a reason that has nothing to do with Ceph itself.
# 1. Check the current cluster version
ceph versions
# 2. Check x86-64-v3 support (CPU microarchitecture level)
lscpu | grep -E 'Model name|Flags' | head -2
# x86-64-v3 requires AVX2; look for avx2 in the flags:
lscpu | grep -o 'avx2' || echo "avx2 missing — containerized Umbrella is blocked"
# 3. Check the Python version for cephadm
python3 --version
# 4. Before upgrading: check cluster health
ceph -sA release cadence that is accelerating
The version timeline tells a story between the lines. Squid lasted just over two years (September 2024 → October 2026). Tentacle runs on a similar horizon (November 2025 → June 2027). Umbrella pushes the deadline to two and a half years. But Umbrella’s release lands only three days before Squid’s end of life — a short overlap that leaves little time for a Squid operator to validate Tentacle before their branch becomes orphaned.
That is not a criticism of the Ceph community; it is a fact of life for teams that run distributed storage without a full-time owner. Storage gets upgraded when it has to be, not when it is convenient — and the window between major versions is narrowing. None of this is an argument for skipping upgrades; the opposite is true. The tighter the cadence, the more it pays to stay one version behind the bleeding edge and one ahead of end-of-life. Operators who treat Squid’s deadline as a forcing function — upgrade to Tentacle now, evaluate Umbrella on a test cluster over the winter, migrate in spring — will never face a two-major-version emergency. Operators who wait will, eventually, do exactly that, and it will cost more than a routine window.
Verdict
If you run a Ceph Squid cluster, start the move to Tentacle now: your branch is end-of-life today, and the only way out is one major version at a time. Do not jump straight to Umbrella: Ceph’s sequential rule forbids it anyway, and Umbrella changes the system base — first confirm your hosts support x86-64-v3 and Python 3.9+ before even planning. And if you are already on Tentacle, stay there for now: it is the stable branch, maintained until 1 June 2027, and Umbrella is three days old — give it time to prove itself in production before adopting it. Ceph Squid does not vanish overnight: it simply becomes a cluster with no safety net, and in distributed storage the only thing that matters more than capacity is the version underneath it. Run ceph -s before you plan anything — a healthy cluster makes the whole migration a routine window, and an unhealthy one makes it an incident.