Document the multi-pattern presence model and migration 9

Describe presence_patterns (zero or more home/away patterns per child,
each with its own cycle length and anchor date), the first-match-wins
evaluation order, and migration 9 replacing the one-per-child schedule
table. Update the Admin Schedule spec for the patterns list, pattern
editor, combined month grid and Sunday-first week; replace the
set_presence_pattern service row with the admin pattern routes; and add
the pre-upgrade backup and rollback step for migration 9 to the runbook.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
2026-09-24 20:30:20 -04:00
co-authored by Claude Opus 5.5
parent 4172b286ff
commit 94c4617459
5 changed files with 173 additions and 15 deletions
+69
View File
@@ -551,3 +551,72 @@ The file provider reloads the dynamic file on save.
| 5 | Panel refused externally | section 9 check 4 | `403` |
| 6 | Non-Traefik source refused | from a LAN machine other than `10.60.1.33`/`10.60.1.14`: `curl -s -o /dev/null -w '%{http_code}\n' http://10.60.1.14:8081/` | `403` (nginx `deny all`) |
| 7 | Proxy source allowed | on `10.60.1.14`: `curl -s -o /dev/null -w '%{http_code}\n' http://10.60.1.14:8081/` | `200` |
## 11. Upgrades and rollback
The API applies pending schema migrations when it starts (FastAPI
lifespan, before `/health` answers), so **the first start of a new
version is the point of no return** for any migration it carries. A
build refuses a database whose schema version is newer than it
understands ("newer than this integration understands … downgrades are
not supported") and does not start.
### Migration 9 (multiple presence patterns)
Migration 9 creates `presence_patterns`, copies every row of the old
one-per-child `presence_schedules` table into it as a `home` pattern
named `Home schedule`, and **drops `presence_schedules`**. Version 0.7.1
refuses the resulting version-9 database, and could not read
`presence_patterns` anyway, so the only way back to 0.7.1 is a copy of
the database taken **before** the first start on the new version.
**Before the first start on the new version**, with the old version
still deployed:
1. Stop writers (as section 8 step 1):
dc stop api litestream
2. Copy the database file beside the live one with SQLite's online
backup (includes anything still in the WAL), then check it:
ts=$(date -u +%Y%m%dT%H%M%SZ)
dc run --rm --no-deps --entrypoint sqlite3 litestream \
/var/lib/nestquest/nestquest.db \
".backup /var/lib/nestquest/pre-v9-$ts.db"
dc run --rm --no-deps --entrypoint sqlite3 litestream \
/var/lib/nestquest/pre-v9-$ts.db \
'PRAGMA integrity_check;' \
'SELECT version FROM nestquest_schema_version;' \
'SELECT count(*) FROM presence_schedules;'
Expect `ok`, schema version `8`, and the number of presence
schedules. Write down `$ts`.
3. Also record the Litestream replica position and the UTC time, so a
point-in-time restore (section 8, `-timestamp`) can reach the
pre-upgrade state even if the local copy is lost:
dc up -d litestream
dc exec litestream litestream status # note the txid
date -u +%Y-%m-%dT%H:%M:%SZ # note the time
dc stop litestream
The replica only covers the 7-day retention window (section 7).
4. Deploy the new version (sections 1–2) and run section 9. Confirm the
live database is at version `9` (the section 8 step 3 `sqlite3`
command against `nestquest.db`, with
`'SELECT version FROM nestquest_schema_version;'`) and that the
Schedule tab lists each child's old schedule as a `Home schedule`
pattern.
**Rollback to 0.7.1:** check out and build 0.7.1 (section 1), then put
the pre-upgrade database back with section 8: either steps 1 and 4–6
with `pre-v9-<ts>.db` in place of `nestquest.db.restored`, or the full
restore with `-timestamp` set to the time noted in step 3. Presence
pattern changes made after the upgrade are not in that copy and are
lost. Once the new version has proven itself, remove the copy:
`dc run --rm --no-deps --entrypoint rm litestream -f /var/lib/nestquest/pre-v9-<ts>.db`.
This rollback has not yet been rehearsed.