Historical record: 3.2 staging observations
Documentation home · Current support scope
This record summarizes the 3.2 staging pre-release observations from 25–26 October 2026. It is not a claim that newly resolved 3.3.0 dependencies have the same runtime results. Implemented contracts, historical measurements and current build status are separate kinds of information.
Recorded environment and sources
The earlier four-node environment used Linux amd64, Alpine, containerd, kernel FUSE and bidirectional kubelet mount propagation, with ordinary local-CSI PVCs. A literal single-node lab observation later used published 3.2.0 images. Exact versions, component sources and image identities are preserved in the reports; they are historical observations, not current dependency or platform pins.
The first-usable final workers used source f697c11; their manager/node binaries
matched the earlier 27ef0e8 round. Later candidates and observations have their
own component/source boundaries. A result from one tuple does not establish a
new tuple's behavior just because its version number is similar.
Observed scope
| Area | Recorded boundary |
|---|---|
| File semantics | Basic ownership/timestamps, sparse/truncate and append/rename behavior, hardlink/symlink publication and durable handle lifetime. |
| Application use | Content publication and restart/readback with the recorded Webtrees and VikingCMS workloads. File-backed databases and general POSIX behavior remained excluded. |
| Existing mounts | Scoped serving/coordinator and frontend recovery observations; application Pod restart alone did not prove a fresh kernel stage. |
| Credentials | Same-CA rotation/overlap/cutoff observations on the recorded runtime, not emergency CA revocation. |
| Membership | Scoped minority refusal, controlled learner replacement and original-member fencing. Permanent-node disposal was not established. |
| Delete/Retain | Exact owned API/backing reclamation and preserved same-authority retained reinstall within the recorded identity boundary. |
| Single node | Published 3.2.0 RF1 I/O, component restart, normal unmount/remount and reclamation. No redundant-storage failure tolerance is implied. |
Historical timings
The final first-usable soak acknowledged 362 and 379 records from the two writers, with all four readbacks passing. Normal Delete removed 65 captured API identities and nine backing path pairs within a conservative 137.472-second observation. Retained explicit disposal removed ten API identities and its backing pair within 157.822 seconds. Observed content publication took roughly 88 seconds; earlier short-deadline HTTP failures remain failures. These are observed samples, not production latency or availability targets.
The later eb1 no-fault run failed with RPC deadlines. The separately recorded d421 20-minute baseline passed with 68 records per writer and four readbacks. That later source-bound outcome did not identify the older stall's cause or qualify the final floating-input build. RWO implementation and local/same-host observations do not establish multi-host takeover for a different release.
Original records
- First-usable 3.2 report
- Application observations
- Single-node observation
- Later d421 deployment/soak record
Old test harnesses and backlog items have been removed by operator decision. Original reports remain archives and can refer to procedures available only in Git history. Current operation follows the guides linked from the documentation home, not old milestone acceptance instructions.