hub (unreleased): two end dates warn the operator 14 days before — an operator token and a box's web certificate (R-1379, D77); the other sixteen as a design table
Claude-Session: https://claude.ai/code/session_0159rPz1ZhFKsS53msqPYxtS
This commit is contained in:
1 parent
68292e40f6
commit
86bb912a4b
11 files changed
+527
-5
No files matched your search
@@ -374,6 +374,8 @@ one info line beside `host_crash_restart`; `operatorOnlyEvents`, pinned by
|
||||
| `tunnel_down` | error | the box's two newest host reports say the tunnel is `not_running` (more than one report cycle, 15 min) and the one before did not | the first `running` after it → `tunnel_recovered` (info) | `api/tunnel_test.go` |
|
||||
| `proxmox_trust_problem` | error (warning for „the root CA copy ends within 30 days”) | **[built, not deployed — hub `main` after v0.181.0]** the host report's `proxmox_trust` stanza names a cause (`03` §12a); once per (box, cause), not at every report | the first report with no problem → `proxmox_trust_ok` (info), one for all causes | `api/proxmox_trust_test.go` |
|
||||
| `box_clock_wrong` | warning | **[built, not deployed — hub `main` after v0.181.0; needs an agent that stamps `sent_at`, built and not released]** the box's clock at the moment of sending is more than **one minute** from the hub's clock at arrival, in either direction (D79, `09` §3 decision 300); told once per direction. The hub cannot tell whose clock is wrong: a second box off in the same direction makes the alarm say „check the hub's own clock first”. An agent without `sent_at` is never judged | the first report within **30 seconds** → `box_clock_ok` (info) | `api/box_clock_test.go`; agent `hub/client_sentat_test.go` |
|
||||
| `operator_token_ends_soon` | warning; error when the date passed | **[built, not deployed — hub `main` after v0.181.0; D77, `09` §3 decision 298, §12 below]** an operator access token (hub §17) that is not revoked ends within **14 days**; checked hourly; once per token, and once more when the date passes with the token still there | none (a new token is a new token) | `monitor/end_dates_test.go` |
|
||||
| `web_cert_ends_soon` | warning; error when the date passed | **[built, not deployed — hub `main` after v0.181.0; D77, `09` §3 decision 298, §12 below]** needs a controller after 0.352.1 (built, not released): its report's `web_certs` list names a web certificate that ends within **14 days** — the box renews 30 days before the end, so this means the renewal has failed for about two weeks; once per (household, name, end date) | none (a renewed certificate has a new end date) | `monitor/end_dates_test.go`; controller `infra/webcert_test.go` |
|
||||
| `hub_daily_job_skipped` | warning | **[built, not deployed — hub `main` after v0.181.0; D78, `05` §16.6]** at a hub start: the newest slot of `deadline-check` or `offsite-key-audit` passed while the hub was not running; once per missed slot | none (the next slot runs) | `monitor/daily_ledger_test.go`, `cmd/hub/r1383_wiring_test.go` |
|
||||
| `operator_mail_lost` | error | **[built, not deployed — hub `main` after v0.181.0; D78, `05` §16.6]** at a hub start: operator mails of the last 48 h whose newest delivery record is `failed` — one mail for all, with each lost mail's own sentence; and a row (no mail) when three retries gave up | none | `notify/lost_operator_mail_test.go` |
|
||||
| `signed_job_not_taken` | warning | **[built, not deployed — hub `main` after v0.181.0; D78, `05` §16.6]** a signed job is still in a box's queue after two hours; once per job | none (the job leaves the queue when the box takes it) | `monitor/signed_job_waits_test.go` |
|
||||
@@ -772,3 +774,39 @@ only a word (`WORD_ONLY`: „security announcements", „no longer receives secu
|
||||
another program in the repo (`PACKAGE_NOT_OURS`: Gitea's runner, Home Assistant's phone apps, Uptime Kuma's rootless image) is
|
||||
not listed; a name not in the table is listed — the safe direction. A `YY.MM.YYYYMMDD` tag is not „newer". Test W16.
|
||||
|
||||
|
||||
|
||||
## 12. End dates — who knows the date, and who is warned [R-1379, D77, `09` §3 decision 298]
|
||||
|
||||
The End-dates list (`audits/end-dates-list-2026-11-11/`) found 18 end dates with no warning before them. The ruling:
|
||||
the product warns 14 days before the two it can read itself; the rest stays a list the operator reads.
|
||||
|
||||
### 12.1 The two that are built `[FACT — built, not released: hub `main` after v0.181.0 and controller `main` after 0.352.1, 2026-11-14]`
|
||||
|
||||
| End date | Who knows the date | How it reaches the hub | The warning |
|
||||
|---|---|---|---|
|
||||
| A box's web certificate | **only the box.** Traefik in the guest asks Let's Encrypt and keeps the certificate in its own store (`<stacks>/traefik/acme.json`). The household's public names end at Cloudflare, so nobody outside can read it | the controller reads the store's certificate fields — names and end dates, never a key (`infra.ReadWebCerts`; the store's shape was measured on three boxes on 2026-11-11) — and sends `web_certs` in its report | `web_cert_ends_soon`, 14 days before; an error if the date passes |
|
||||
| An operator access token | the hub (`api_tokens.expires_at`) | — | `operator_token_ends_soon`, 14 days before; an error if the date passes |
|
||||
|
||||
What it does not do: a box with NO certificate in the store (no ACME e-mail: Traefik serves its own certificate) is not
|
||||
told — the field is absent. A box that is switched off is judged on its last report: the warning is still true (an off
|
||||
box cannot renew). A token or certificate that ended more than 14 days ago raises nothing at the first run (no news).
|
||||
**Not proven:** on a box or a deployed hub; a real failed renewal has never been seen.
|
||||
|
||||
### 12.2 The other sixteen `[DESIGN — proposed, not built]`
|
||||
|
||||
| End date (block) | Who knows the date today | Could the product read it? | Proposed |
|
||||
|---|---|---|---|
|
||||
| The hub address's certificate, and the download, ISO, Gitea and website names (DH20) | anybody: readable from outside with no login | yes — the hub can open a TLS connection to its own five names once a day | **build next:** the same checker, a third source. Small. Not built tonight: it is a call from the hub to outside names, and the name list needs the operator's yes |
|
||||
| A node's Tailscale key (DH30) | the Tailscale admin page; one date is in `operations/tailscale.md` | not from a box (the service does not run on two of three) | the operator's list |
|
||||
| The package signing keys (DB16) | measured: the soonest key in use ends 2029-01-15 | yes, by the root wrapper's facts read | no warning now; read again in the yearly end-dates pass |
|
||||
| The 30-day restore copy (DB24) | the box (the copy's own date) | yes | not an alarm: a page fact. The household's restore page shows the date already; nothing for the operator |
|
||||
| The end of security fixes for Debian 13 and Proxmox 9 (DB17) | nobody in this project: no document gives the day | no — it is a published date, not a machine fact | the operator's list, with the two dates written into `11-os-updates.md` once they are read |
|
||||
| App images and the images the controller starts (DB18, DB19) | no end date exists; the end is the day a registry stops serving | no | none; the nightly security watch covers the versions |
|
||||
| The agent's local-API certificate (DB03) | source: 10 years from the install | yes (the agent made it) | no warning before 2035; a row then |
|
||||
| A household's own Cloudflare token (DH23) | only the household (it set the end at Cloudflare) | **maybe:** the hub's zone-reach check calls Cloudflare's zone list (`cloudflare/reach.go`), which gives no end date. Cloudflare has a token-verify call that is said to give one (`expires_on`) — NOT read in this project, NOT seen in an answer | **design question, with a spike first:** one read-only token-verify call on a test household's token, to see the field. Not done tonight: a call to Cloudflare is forbidden in this run |
|
||||
| The mail service key; domains and registrars; the Cloudflare account, the site publish token and the ISO store keys; the Hetzner account; Gitea and registry logins; the tokens in CC's secret store (DH24, DH27, DH28, DH29, DH31, DH32) | the account's own page; no document holds a date | no — each needs a login the product must not hold | **the operator's list:** one table of name, where to read the date, the date, read on. It does not exist yet (R-1381 asks the operator to read the dates only they can read) |
|
||||
|
||||
**A third source that is small and was left out on purpose:** the hub's own five public names (first row). It needs
|
||||
no secret, and its cost is one outbound connection a day; it waits for a yes because tonight's brief forbids outside
|
||||
calls and the list of names is a promise about what the hub watches.
|
||||
Reference in new issue
Block a user