Files
esh-pfi-infrastructure/stacks/park/README.md
T
vh b637947ffd docs(park): record the API-key ruling — leave it as is
Operator ruled 2026-08-18 on the 5-character PARK_API_KEY flagged during
the v1.0.0-beta.2 deploy: leave it. The henge is LAN/WG-internal and
never internet-exposed. Written down so the next audit does not re-raise
a question that has already been answered.
2026-08-18 05:49:10 -07:00

74 lines
4.2 KiB
Markdown

# park (stonehenge-park — "the henge")
Self-contained needs-attention tracker: FastAPI + SQLite (FTS5) + an in-process
scheduler, one container. Source: `gitea.phasefinal.com/vh/stonehenge-park`,
deployed at tag **v1.0.0-beta.2** (commit `2c258f7`). Owned by `park-dev`.
- **Host:** ana-docker (`10.250.50.70`), compose home `/opt/docker/compose/park/`.
Permanent home since 2026-08-13 (migrated off nh3-docker before the v1.0.0 final cut).
- **Stable name:** `park.phasefinal.com``10.250.50.70` (Cloudflare, DNS-only/internal) — clients
and Homepage point at the name so future moves need no client edits.
- **Port:** `8420` (host + container). **LAN/WG-internal only — never internet-exposed.**
- **Data:** named volume `park-data``/data`; SQLite single file `/data/park.db`
(the sole source of truth). Covered by restic via the host's `/var/lib/docker/volumes`
backup source — no per-volume add needed. (⚠ nh3-docker's restic repo was broken as of
2026-08-12 — see fleet infra notes.)
- **Secret:** `PARK_API_KEY` from the vault (`secret get nh3-dev/park-api-key`), in the
host `.env` (chmod 600). `/healthz` and the UI shell are unauthenticated, asset GETs
under `/ui/assets/` are public, and `/park/*` needs the bearer key.
**The key is 5 characters, and that is a settled decision — leave it.** Surfaced to
the operator 2026-08-18 with a rotation offer; ruled *leave as is*. The service is
LAN/WG-internal and never internet-exposed, so the exposure is "already on the
network", not the open web. Do not re-flag this on the next audit.
- **Health:** `GET /healthz``{"status":"ok"}` (image `HEALTHCHECK` built in).
## Image is built locally (no registry yet)
```bash
# ⚠ on ana-docker (10.250.50.70) — NOT nh3-docker, decommissioned for this stack 2026-08-13.
# The source is exported per-tag from a checkout that has the tag, so the previous
# tag's tree stays on the host as a rollback. From a box with the repo:
git -C ~/development/stonehenge-park archive --format=tar v1.0.0-beta.2 | \
ssh infra-ops@10.250.50.70 'mkdir -p ~/deploy-src/stonehenge-park-v1.0.0-beta.2 &&
tar xf - -C ~/deploy-src/stonehenge-park-v1.0.0-beta.2'
# then on ana-docker (infra-ops is in the docker group — no sudo needed):
cd ~/deploy-src/stonehenge-park-v1.0.0-beta.2
docker build -t park:v1.0.0-beta.2 .
cd /opt/docker/compose/park && docker compose up -d park
```
**Never `docker compose down -v`** — that removes `park_park-data` and every parked
idea with it. `up -d` and plain `down` both preserve it.
Swap `image:` for a registry ref once one exists (park-dev left the door open).
## althing wiring (TODO)
The scheduler posts needs-attention snapshots to althing handle `henge-crow` by
shelling out to `althing-cli post` inside the container. `althing-cli` is **not yet
wired into the container**, so `PARK_ALTHING_CHANNEL` is set **empty** in the host
`.env` (push disabled; REST endpoints fully functional — graceful by design). To
enable: get `althing-cli` reachable in-container (bind-mount + config / sidecar /
host-run), set `PARK_ALTHING_CHANNEL=henge-crow`, recreate.
## Verify
```bash
curl -s http://park.phasefinal.com:8420/healthz # {"status":"ok"}
curl -s http://park.phasefinal.com:8420/ # UI shell
curl -s http://park.phasefinal.com:8420/ui/assets/favicon.svg # brand assets (public GET)
curl -s -H "Authorization: Bearer $KEY" \
"http://park.phasefinal.com:8420/park?limit=100" # the parked items themselves
```
**Use GET, never `curl -I`.** Every route on this app 401s on **HEAD** — including
`/healthz` and `/`, so it is app-wide and not an asset or auth-scope problem. `curl -sI`
therefore reports a false failure on a perfectly healthy deploy. Verified 2026-08-18:
`HEAD /healthz -> 401` while `GET /healthz -> 200`. Reported upstream to park-dev.
**`/park/due-count` returning 0 is not a data-loss signal.** It counts what is *due
now*; a henge holding 28 items can legitimately report `{"count":0}` with `/park/due`
empty across overdue/today/stale. To check items actually survived a redeploy, hit
`GET /park` (or count rows in `/data/park.db`), not the due counters.