Skip to content

docs(adr): the estate's backup sets ride on the second recipient's medium, and the visit that proves the key proves the copy - #588

Merged
Gerrrt merged 2 commits into
mainfrom
gerrrt/estate-backup-offsite-2e1180
Sep 21, 2026
Merged

Gerrrt merged 2 commits into
mainfrom
gerrrt/estate-backup-offsite-2e1180

Conversation

@Gerrrt

@Gerrrt Gerrrt commented Sep 20, 2026 •

Copy link
Copy Markdown
Owner

The decision

ADR-0048: the estate's three backup sets — the firewall export, the volume sets, the NAS set — go beyond the shelf on the medium that holds the second age recipient (ADR-0024, #294), carried there on the ninety-day visit that medium already owes this host, and proved by the same visit. Of #573's three answers this is the second. The first, #455's drive, has the wrong reader and the wrong key and does not exist yet; the third, the shelf, is defensible for the volume sets and wrong for the firewall export, which is the one artefact with a real rebuild cost and exactly what a successor standing up a replacement morpheus wants first.

The cost is stated and accepted: that medium becomes the one place where the estate's ciphertext sits beside a key that opens it. ADR-0015's refusal to give oracle that property stands — its argument turned on a powered, networked host in the same room. docs/security.md records the disclosure delta on losing the medium (the firewall's rule bodies and WAN address, the export's password hashes, grafana.db) beside ADR-0023's residual.

What is built

  • scripts/backup-offsite.sh and make backup-offsite DEST=…. Refuses a destination on this host's own filesystem by device number (a copy on the disk being insured is off-host to nowhere) and one inside this repository. Re-verifies every complete set already on the medium against its own MANIFEST before writing — the proof that last visit's bytes survived ninety days — then copies the newest complete set of each kind the medium lacks, MANIFEST last, syncs, hashes, and applies OFFSITE_KEEP (default 1, never the newest, only names it wrote). Layout mirrors oracle's. --self-test runs 22 fixtures against a fake tree and a /dev/shm destination under make validate; benched against the real sets (1.7 GB) to tmpfs: copy, verify, list.
  • The third human job with a deadline and no timer. offsite-copy row in the JOBS table beside the two key proofs, same ninety days; excluded from ScheduledJobStale; OffsiteCopyStale with "mount the medium" advice; unit tests pin that a stale job fires the new rule and not the generic one. --list, --verify-only and --prune run unwrapped so looking at the medium never resets the clock. A bare make backup-offsite exits 2 before reaching the wrapper.
  • copy-the-backups-offsite.md — the visit's procedure, what the proof does and does not prove, restoring from the medium, and the failure table.
  • The residual is owned. roadmap.md (Rehearse the firewall restore on trinity #92's paragraph, a Decided but not built entry, the Off-estate copy: ADR-0023's second-person proof #455 row), build-the-nas.md §8, both restore runbooks, schedule-maintenance.md, observability.md, successor-handover.md (says where the sets are without saying where the medium is), the three scripts' "see roadmap" lines, ADR-0015's forward note. Rule and ADR counts follow main, which claimed 0046 and 0047 while this was open.

What is not done here, on purpose

The copy. It needs the second recipient's medium at the host, which has not yet made its first visit — that recipient's proof series reads 0. #573 stays open until the first proved copy; ScheduledJobNeverRan{homelab_job="offsite-copy"} is the expected reading after make install-timers until then. The visit is: mount, make backup-offsite DEST=… ARGS=--list, make backup-offsite DEST=…, then make secrets-verify-backup KEY=… against the same medium.

Verification

  • scripts/backup-offsite.sh --self-test — 22/22.
  • make validate from the worktree, output to a file, no FAIL.
  • promtool check rules and promtool test rules on backup.rules.yaml / backup.test.yaml via the pinned image.
  • install-timers.sh --check: offsite-copy: no timer, 7776000s deadline declared for the human.
  • make backup-offsite with no DEST exits 2 and writes nothing.

Refs #573, #92, #455, #294.

🤖 Generated with Claude Code

…dium, and the visit that proves the key proves the copy

ADR-0047 decides where the firewall export, the volume sets and the NAS set
go beyond the shelf (#573): on the offline medium that holds the second age
recipient, carried there by `make backup-offsite` on the ninety-day visit that
medium already owes, and proved there. One copy of record, the newest complete
set of each kind, in oracle's layout so a restore is "copy the directory back".

scripts/backup-offsite.sh refuses a destination on this host's own filesystem
by device number, re-verifies what the medium already holds before writing,
copies MANIFEST-last, hashes the copy, keeps OFFSITE_KEEP per kind and never
the newest, and names a set that differs without deleting it. Its --self-test
runs 22 fixtures under make validate. The offsite-copy row joins the two key
proofs in install-timers.sh with no unit and the same ninety days; it is
excluded from ScheduledJobStale and OffsiteCopyStale reads it with the right
advice; the unit tests pin that. --list, --verify-only and --prune run outside
the wrapper so inspecting the medium never resets the deadline.

The residual four documents named and no issue owned is now owned: roadmap,
build-the-nas §8, both restore runbooks and the three scripts' "see roadmap"
lines point at the ADR and the runbook; successor-handover says where the
sets are without saying where the medium is; security.md records the accepted
disclosure delta of ciphertext beside a key that opens it; ADR-0015 gains a
forward note.

The copy itself is a visit only the holder of the medium can make, so #573
stays open until the first one; ScheduledJobNeverRan names offsite-copy until
then, which is the honest state.

Refs #573, #92, #455, #294.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…nder it, and the counts move with main

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant