gates / gates (push) Successful in 2m8s
Reads only on the N100 and demo-hp; no box changed, nothing released. Claude-Session: https://claude.ai/code/session_0159rPz1ZhFKsS53msqPYxtS
59 lines
8.1 KiB
Markdown
59 lines
8.1 KiB
Markdown
# The N100 — the lived-in box (diary)
|
||
|
||
**What it is.** `demo-felhom` (N100 mini-PC, guest 9201, `felhom.demo-felhom.eu`). From 2026-10-15 it is used the way a
|
||
household uses a box, and it changes only through the product. **The rule:** `architecture/09-update-architecture.md`
|
||
§3 decision 254 — the dashboard, the hub, the normal release train and the OS update rings only; no deliberate failure
|
||
without the operator's word for that one act; reads and observation are free. Tiers: `runbooks/target-selection.md`.
|
||
|
||
**What it runs (2026-10-15):** controller 0.314.0, agent 0.156.1. Apps: opengist (standing), qBittorrent, Prowlarr,
|
||
Sonarr, Radarr (installed 2026-10-15 through the dashboard). Plex: the operator installs it. Drives: the system SSD and the
|
||
USB drive `hdd_1` (ADATA HD710 PRO / TOSHIBA MQ04ABF100, 931.5 GB), which is also the whole-guest backup target.
|
||
|
||
**How to add a line.** Anyone — the operator or a session — who notices something on the N100 adds a dated line below:
|
||
what happened, who noticed, and the register row it became. Prose here is not a record; the row is.
|
||
|
||
## Diary
|
||
|
||
- **2026-10-15 — the rule starts.** Written as decision 254. (CC, n100-lived-in task)
|
||
- **2026-10-15 — the USB drive stays one drive.** The operator asked for a split in two. Read first: it holds no family
|
||
files — 2.7 GB of the box's nightly whole-guest backups (`dump/`), 3.1 GB of app backups and an off-site restore copy,
|
||
422 MB of leftover app data from removed apps. It is the whole-guest backup target, so the product refuses to retire it,
|
||
and the box has no other drive for that role. Nothing was changed. → R-1062 (the operator decides), R-1063 (the
|
||
dashboard's half-state on that refusal, read from the code). (CC)
|
||
- **2026-10-15 — media apps installed through the dashboard.** qBittorrent, Prowlarr, Sonarr, Radarr — each healthy in
|
||
under 35 s; the box chose `hdd_1` for their own config (the wizard asks no drive). The box wired them by itself:
|
||
Sonarr `…/media/tv` and Radarr `…/media/movies`, tag `hdd-1`, qBittorrent as their client with categories
|
||
`tv-hdd-1` / `movies-hdd-1` saving under `…/media/torrents/` (the same drive, so imports can be hardlinks); Prowlarr
|
||
linked to both. A stranger meets the gate on Prowlarr/Sonarr/Radarr and qBittorrent's own login. (CC)
|
||
- **2026-10-15 — films never go off-site, proven before any download.** The controller's own classifier over the
|
||
deployed apps: off-site takes no media path; the cross-drive tier would take only qBittorrent's downloads and is not
|
||
on for it (with one data drive it falls back to the SSD with state only); the whole-guest backup excludes the drives
|
||
(its log: „excluding bind mount point mp8"). (CC)
|
||
- **2026-10-15 — demo-hp, not the N100:** the leftover `/mnt/sizing-bench` was removed through the dashboard; it held two
|
||
apps' cross-drive copies and the dashboard said nothing about them → R-1064. The hub showed demo-hp OK at its next report (14:56 UTC). (CC)
|
||
- **2026-10-15 — read before the operator's Plex install:** a freshly claimed Plex gets no libraries until the next
|
||
wiring run (the claim does not start one) → R-1065; the sheet asks for „Bekötés most" only if no library appears.
|
||
(CC, from the code)
|
||
- **2026-10-16 — the file manager's password, set by the box at the operator's yes.** Controller 0.320.0 arrived by the
|
||
normal update path. The N100 did not know FileBrowser's password (set by hand in September, R-513 `operator`), so the
|
||
household could not open the file manager. The dashboard's „Új jelszó beállítása" (pressed through its endpoint, after
|
||
the operator's yes in chat): 200 in 3.5 s; the page now shows the password behind „Megjelenítés"; inside the box the new
|
||
password logs in (200), admin/admin is refused (401), the state is `generated`. Evidence: `audits/logins-folders-2026-10-16/live/n100-fb-reset-*.txt`. (CC)
|
||
- **2026-10-17 — the file manager unreachable from outside for ~21 minutes, after controller 0.327.0.** 0.327.0 arrived by
|
||
the normal update path (11:43:38 UTC) and moved FileBrowser off Traefik's network until its default password is
|
||
checked (R-1126); the box checked and joined it (11:43:44), but Traefik kept the address it had picked at the
|
||
container's start, so `files.demo-felhom.eu` timed out (seen by the release's box walk). Controller 0.327.1 (a file
|
||
route by name) reached the N100 by the normal update path at 12:04:45; `files.demo-felhom.eu` 200, the default login
|
||
refused (401). No data touched. Register: fixed within the release, recorded in R-1126's close. Evidence:
|
||
`audits/filebrowser-memory-ci-2026-10-17/release/box-walk.txt`. (CC)
|
||
- **2026-10-26 — cloudflared 2026.10.3 (Cloudflare's eMudhra notice), and the host clock found 14 s slow.** Controller
|
||
0.345.1 arrived by the normal update path (per-box floor 20:06:04 UTC); the controller recreated cloudflared on
|
||
2026.10.3 by itself, the tunnel was down for about 4 s (a no-app name answered 530, then 404 again). While measuring
|
||
the gap the Proxmox host's clock read 14 s behind DooPlex and demo-hp: `chrony` has failed at every boot since at least
|
||
2026-10-16 („Could not send notification to $NOTIFY_SOCKET"); demo-hp, same package and options, runs it. Register:
|
||
R-1300 (the release), R-1301 (the clock). Evidence: `audits/cloudflared-2026.10.3-2026-10-26/` (`obs-n100-clock.txt`). (CC)
|
||
- **2026-11-05..07 — the nightly Proxmox update step refused three nights in a row.** `proxmox-grub` 2.12-9+pmx4 needs `grub-efi-amd64` of the same version; the update wrapper put `proxmox-grub` in the Proxmox group (its boot-package rule matched only names that START with `grub`), so the whole step was refused (R6) and 34 Proxmox updates waited; an `os_update_failed` mail each morning. Apps not touched. Agent 0.172.1 puts `proxmox-grub` with the boot packages; it reaches the box by the operator's vouch and two signed jobs. Register: fixed in the session (size rule), recorded in the agent's CHANGELOG. Evidence: `audits/demo-check-2026-11-07/n100-pve-step-R6.txt`. (CC)
|
||
- **2026-11-10 — every place that holds data was measured (read-only; the space list).** Host system disk 38.09 of 100.86 GB (38 %), 4.6 GiB more than on 10-15: the box holds 8 kernels (8.15 GB) and none was ever removed → R-1368. The agent's self-update folder holds 48 old binaries (724 MB) and there are 11 old agent copies beside the binary (155 MB) → R-1368. The host journal sits at its cap (4.28 GB, 74 files, oldest entry 10-04) → R-1367. The three whole-box copies of 2026-07-28 still lie on the system disk (17.59 GB) → R-1087; four template archives (2.56 GB) → R-1368. On the USB drive, `backups/offsite-restore` holds 1.65 GB of two apps that are not installed any more → R-1372. Nothing was deleted or changed. Table: `audits/space-list-2026-11-10/TABLE.md`. (CC)
|
||
- **2026-11-14 — the clock is right again, repaired by the product.** When the box took agent 0.174.0 and its root files (11:36 and 11:51 UTC), the new „profile refresh" found the old security profile of the time service, loaded the right one and started the service: the clock stepped forward 32.7 s (offset to DooPlex −32.55 s → +0.18 s; still so 2.5 hours later). Nine seconds after the step the controller read a new boot and recreated Radarr, Sonarr and qBittorrent once (each down for a few seconds; all healthy). A dry run says tonight's night removes 6 of the 8 kernels. Register: R-1301 and R-1388 closed, R-1395 (the controller and the clock step), R-1368 (the kernels — read the night of 11-15). Evidence: `audits/release-2a-2026-11-14/D/` (`09`, `13`, `13b`, `14`). (CC)
|
||
- **2026-11-15 — the box removed its old kernels by itself, the first time.** At 05:58 CET, at the end of the night's update run (the Proxmox step applied 12 packages first), the clean-up purged 6 of 8 kernel packages in 35 s and kept 7.0.14-26 and the running 7.0.14-28: 6.81 GB gained, 65.1 GB free after, 0 left-over paths. The default kernel and the pinned file did not change; 0 failed units, 9 of 9 containers healthy, the clock +0.06 s to DooPlex; the hub's „Kernel step” cell reads „cleaned”. Nothing was done by hand. Register: R-1368 (D82: 1 of 2 nights). Evidence: `audits/night-build-2026-11-14/E/`. (CC, read-only)
|