Files
felhom.eu/REPORT-release-2026-10-26.md
T

107 lines
8.7 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
# REPORT — the Monday release, run on Sunday 2026-10-25 (attended)
**Plain words first.**
1. **Fix A and fix B are proven.** A restore whose app does not start says so and keeps both copies (D49 = A); the old
copy goes only after 3 minutes of stable running. A `.fab` import that cannot download its version refuses and
touches nothing.
2. **Hub 0.170.0 is live.** Tester 1's recovery package went to the hub in one transaction; no alarm mail after.
3. **Agent 0.165.0 is vouched** (18:38:49 CET). It runs on all four hosts; on a BYO host our SSH file is removed — one
bundle run late (R-1285).
4. **Controller 0.343.0 is on every box:** 4 of 4 hosts (Tester 1, demo-hp, demo-felhom, the BYO test host) and the
scratch guest 9202. None stayed behind.
5. **The catalog is pushed:** n8n sends no usage data, OpenGist has 256 MiB, Homepage is behind the family gate.
**The D47 upgrade count:** Tester 1 3 of 3 apps got an install id, demo-felhom 5 of 5, demo-hp 11 of 12 (kimai is held
stopped since 2026-10-14 and the start-up step skips a held app — R-1286). **0 units set aside on any box.**
**The dated checks:** R-232 — DONE (last night's DooPlex database copy reached the restic sweep, snapshot `3973450f`,
folder `20261024-220001`, no alarm mail since 2026-10-23 22:20Z). R-1183 — partial today (one hub snapshot file per
slot, `07:00Z` and `13:00Z`; the hub restarted at 15:50Z for 0.170.0, so the morning log is gone); the full read stays
due 2026-10-26.
## What ran, in order
- **Part 1, fix A (R-1270 follow-up):** one helper, „started" = running on every sample of a 45 s window, no
`RestartCount` growth, every healthcheck `healthy`; the undo copy released after 3 more minutes (10 s samples, 30 min
watch; only a restore begun after the watch ends it). Found live and fixed on the way: docker's `-f` template failed on
every app without a healthcheck (now the JSON is read). Live on 9202 rc.6: an image that exits after 20 s → „még nem
indult el", copy kept, one operator line; the original unit back → released 3 min later.
- **Part 2, fix B (R-1279):** the D48 helper before step 1 of the import. Live: a bundle copy with a missing tag →
refused with the D48 sentence, the app running, its volumes untouched.
- **Part 3, hub 0.170.0 (R-1254):** asked → yes → manifest bump, hard refresh, sync of `Deployment/hub` alone,
Synced/Healthy, digest `sha256:6478498d…`. Tester 1 ran the escrow ceremony through the page's endpoints (the new code
stored by the operator as a CC secret); the hub logged „the new package is on the hub". No alarm mail after the
deploy (the mailbox as the second channel; a bind mail to Tester4 arrived, the control).
- **Part 4, agent 0.165.0 (R-1257, R-1250):** `release-agent.sh` from a clean clone; signed jobs to the three boxes
(76/76, the appliance drop-in in force by `sshd -T`). **No BYO host existed, so one was built** on demo-hp: VM 9371, a
plain Proxmox VE 9.2 from an answer-file ISO, Felhom installed with `--mode byo` as Tester4 (the operator deleted the
old host and stored the retrieval passphrase as a CC secret). Our 0.164.0 drop-in planted first; after 0.165.0 still
there; a second config bundle removed it, wrote nothing, `sshd -T` read the owner's own setting. Vouched by the
operator.
- **Part 5, controller 0.343.0:** rc.6/rc.7 on 9202; the set-aside alarm added and red-proved (Part 5.2); `0.343.0`
pushed after CI 2330; per-box floors → the three boxes at 17:17Z; box checks: 0 ERROR lines, all containers up,
https-only green; the D47 count above; on Tester 1 a unit restore and a set-aside restore — the item back, the copy
released after 3m0s each. Asked → yes → global floor 17:53:42Z (MinAgent 0.158.1, read back in the hub log; the BYO
host moved 18 s later); per-box floors cleared; a 30-minute watch: the hub sent 0 alarms (`global-floor-watch.txt`) and no alarm mail arrived (the mailbox as the second channel).
- **Part 6, catalog:** Homepage's family gate tried on the drill on 9202 (stranger 302/401, household 200); asked →
yes → `ebbf1f4` pushed; demo-hp read it 8 minutes later.
- **Part 7, D44/D45 = all picks:** confirm lines tagged, text-follows-code changes made (old words struck through), the
four code picks became R-1280–R-1283; R-1265 and R-1266 closed.
- **R-1250:** released, but its live proof cannot run before the agent's next due window (Tuesday 10:00–15:59 local; all
boxes measured on 2026-10-24) — a dated check on 2026-10-27 with the operator's yes for DooPlex's qBittorrent as peer.
## Versions, digests, CI by commit
- **Hub 0.170.0** `sha256:6478498d…`.
- **Agent 0.165.0** binary `e9d04a47ee85222ce6c472633ecdcebe1baa88d6836848f9a7da6435a9adb7d3`, bundle
`8dea1f34d77ba40aa44c5461c3728c2ca94bab46854a6bd8b86191f520dfe85a`; agent CI by commit: `c8d3c17` 2306 ✓, `4a1a849` 2325 ✓, `4697f8a` 2326 ✓, `6b6ba0c` 2334 ✗ (R-1290 — the registry, not the code).
- **Controller 0.343.0** `sha256:bb3da72a054c45751e3baf6e4bad43d1561a952717c5f426d65a5aeed0413a3b` from `e2a9761`.
CI: `7720439` 2319 ✓, `2149a13` 2321 ✓, `35232de` 2322 ✓, `1ee0353` 2324 ✓, `3a1f09d` 2327 ✓, `935e624` 2328 ✓,
`e2a9761` 2330 ✓, `fe98377` 2333 ✓.
- **Catalog** `ebbf1f4` 2332 ✓, `cd0cf48` (REPORT).
- **felhom.eu** `ade357a1` 2329 ✓, `36ef8dcd` 2331 ✓, this commit (CI by hash in the final message).
- **Installer 1.36.1** on main, NOT published (R-1284).
## Register
Before **219**, after **219**; opened **11** (R-1280–R-1290), closed **11** (R-1251, R-1252, R-1253, R-1269, R-1270,
R-1271, R-1279, R-1257, R-1272, R-1265, R-1266). R-1254: the hub half closed in its row, the ceremony half open. R-1250:
VERIFY with a dated check. `unproven.py --summary`: no number moved (35 of 55 not walked).
## Observations
- FILED: R-1284 — the installer's preflight printed every refusal code twice („401401"), so its 401/404/409 sentences
never showed (R-1167's 409 sentence included); fixed on main (1.36.1, red-proved), the publish is the operator's.
- FILED: R-1285 — BYO: our drop-in is removed one bundle run late (the old wrapper plans the run that installs the new one).
- FILED: R-1286 — demo-hp's kimai held since 2026-10-14, no row named it.
- FILED: R-1287 — a set-aside mails „1 of 1 apps failed to back up" (red); nothing failed.
- FILED: R-1288 — after a set-aside restore, the next backup sets the reinstall's own unit aside too (a second copy and
alarm).
- FILED: R-1289 — the hub's „fleet's newest controller" can be a dev build no box runs (`0.343.0-dev.r1251a`).
- FILED: R-1280–R-1283 — the four D44 code picks.
- FILED: R-1290 — the agent's CI is RED since ~18:00Z: four old agent packages (0.159.0–0.160.0) were removed from the
registry by something outside the repos, and the `published` gate still expects the newest 10 (`6b6ba0c`, job 2334).
Not the release's code (job 2326 on the release commit was green); the keep rule is the operator's.
- NOT-A-FINDING: the BYO host's first bundle read 71/76 (`pve:store-grant:felhom-pbs`) — the PBS storage was being added;
76/76 two seconds later and on every run after.
- NOT-A-FINDING: the `.fab` import status reads `"ok": true` beside its `error` — `ok` is the HTTP envelope, the job's
result is `error` and the steps.
- NOT-A-FINDING: 9202's alarms read „DROPPED … notifier disabled" — 9202 has no hub; the line is the positive observable
that the event was raised (one per case).
## Teardown (three layers)
- **Machine:** every throwaway app removed through the product (9202: Vikunja, Tandoor, Homepage; Tester 1: Tandoor —
no container, volume, `app.yaml` or set-aside left). 9202 back on the released 0.343.0, its drill mirror and my saved
`controller.yaml` copy removed, my `0.343.0-rc.3…7` images removed (older rc images of earlier sessions left as found).
Tester 1 keeps its new escrow package on purpose (the old recovery code no longer works; the new one is a CC secret).
- **Host:** the BYO VM uninstalled through the product's `--uninstall` (rc 0), then VM 9371 destroyed with its disk; its
ISO removed from demo-hp's `local`; demo-hp lists only VM 341 and its old ISO, as found. DooPlex: only the hub deploy;
the BYO ISO work folder deleted; the session's password, session and job files shredded.
- **Hub:** the three per-box floors cleared; the global floor 0.343.0 and the vouched manifest stay (the release). The
host record `Tester4-f967d4` stays until the operator deletes it (session only — in STATUS); customer Tester4 is kept.
**Secrets:** used only through `sec`; the evidence scanned for every stored value used (0 hits) and the BYO root
password (0 hits); one CSRF token blanked in a saved page snippet. **System poster:** no fact changed. **Instruction
files:** none edited.