USB passthrough (1ea1aa4) made /dev/usb/lp1 visible in the container, but the node
is `crw-rw---- root:lp` (660) and the server runs as the non-root `app` user, not in
`lp` — so open(O_WRONLY) → EACCES → printer still "offline". Add the host's `lp` GID
(7 on this Ubuntu booth, verified `getent group lp` → lp:x:7:) via group_add, so the
app process gains the supplementary group that owns the node. Least-privilege: no
world-writable device, no root, no image rebuild. (If a future booth's lp GID differs,
update the number.)
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
The Hikvision driver's create() had a leftover `console.log(c)` that dumped the
ENTIRE camera config — including the plaintext `password` — to stdout every time
the adapter was built, on every request that resolves a camera. That puts a device
credential in the logs (which get shipped/cached/read — the booth operator is the
adversary). Removed. Swept the rest of the shipped source: no other console.* leaks.
The USB ESC/POS printer is the host's /dev/usb/lpN (usblp char device, major 180),
but the container has its own /dev — `docker exec server ls /dev/usb` → "No such
file or directory", so probeUsb's open() ENOENTs and the printer is always offline
regardless of the path set in setup. Containerization isolates host hardware (same
root cause as the network fix); USB needs explicit passthrough:
- volumes: /dev/usb:/dev/usb → the lpN NODES appear inside the container
- device_cgroup_rules: 'c 180:* rmw' → permit the usblp char major (180), and the
`:*` minor wildcard survives lp0/lp1/lp2 renumbering across replug/boot-order.
Binding the /dev/usb DIR (not a single `devices:` node) is what survives renumber.
Merge verified: parking-data ledger volume preserved (lists append), host net intact,
config valid. Booth prereq: `usblp` loaded at boot + printer attached before start,
else /dev/usb is absent.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
network_mode: host + sysctls: net.ipv4.ping_group_range fails at container create:
"sysctl not allowed in host network namespace" — runc refuses a per-netns sysctl
when there's no separate netns. Remove it; under host net the server uses the HOST's
ping_group_range (set on the booth via /etc/sysctl.d). Fixes the park-buzi-server-1
start failure introduced by c87dcb2.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
In prod the containerized server sat on the Docker bridge (172.18.0.x) and could
only see eth0 — so the setup backend-IP picker (net.ts networkInterfaces) showed
just the Docker IP, the server couldn't reach the relay or fetch Hikvision ISAPI
snapshots, and push devices (readers/cameras) couldn't reach it. The server is the
ONLY container doing device I/O, so put it on the HOST network namespace.
- docker-compose.prod.yml: server + proxy → network_mode: host (server detaches the
base `parking` network via `networks: !reset []`). server VISION_URL=127.0.0.1:8089.
vision stays BRIDGED (it never touches a device — the server hands it JPEG bytes)
but publishes 8089 on 127.0.0.1 only, so the host-net server reaches it over
loopback while the ANPR service stays off the LAN.
- docker-compose.yml: VISION_URL is now ${VISION_URL:-http://vision:8089} so dev keeps
compose-DNS service-name routing; prod overrides to loopback.
- Caddyfile: reverse_proxy 127.0.0.1:3000 (was server:3000 — service DNS doesn't
resolve on host net). Dev doesn't use Caddy, so unaffected.
Merge validated for both envs (booth.sh config, exit 0). Host-net side effect: the
container ping_group_range sysctl is a no-op — the HOST must set it for reader ICMP
liveness (see appliance-provisioning).
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Adopt Komodo Periphery (over the NetBird mesh) as the booth fleet control plane,
superseding SSH-and-booth.sh. The booth runs the SAME compose files; Komodo Core
drives them remotely. booth.sh is demoted to a break-glass local fallback.
- komodo/resources.toml mirrors the working park-buzi Stack (built by hand in the
Core UI, then exported to TOML — field names match the running v2.2). Stack-only:
servers are created by the agent onboarding OUTBOUND (one-time onboarding key →
Periphery self-registers, auto-rotating keys, booth opens no inbound port), so
there is no [[server]] block. Per-booth secrets via [[...]] refs to Core's store.
- komodo/README.md + .env.komodo.example document the flow and the hard rules
(no webhook; onboarding/outbound/mesh-only; per-booth unique secrets; never
down -v the ledger volume).
- wiki/decisions/fleet-deployment-komodo.md records the decision + threat-model
analysis (Periphery is a root agent → mesh-bound; EVENT_SIGNING_KEY-in-Core is a
fraud-root blast radius until ATECC608 signs; Core is now Tier-0; GPL-3.0 is fine
as external ops tooling). container-deployment reframed (booth.sh = fallback);
index + log updated.
Verified end-to-end against a real booth (park-buzi): onboarded OK, Stack deployed,
all containers green, admin seeded.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
The booth deploys the compose files FLAT (e.g. /opt/parking_systems/) with
booth.sh next to them, but the script assumed it lived in <repo>/scripts/ and
blindly did `cd ..` — so REPO_DIR resolved to the parent, where there are no
compose files, and every subcommand operated on the wrong dir. `usage()` then
sed-read a relative $0 that no longer existed after the cd ("can't read
booth.sh"). Discover the compose files instead: check the script's own dir,
then ../, then $PWD, and cd to whichever has docker-compose.yml. usage() reads
an absolute $SELF so it survives the cd.
Also: .env.example defaulted TAG=main, but the registry only has dev-* tags
(no main build yet), so `compose pull` 404s. Default to TAG=dev and document
the moving-vs-immutable (dev / dev-<sha>) tag scheme.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
- lpr-camera.md: "503 Device Busy" can be PERSISTENT (main-stream saturation on
the G3H) — the real fix is sub-stream selection, not just retry.
- device-status-monitoring.md: QR reader health was false-healthy (hardcoded
"ready") until the ICMP-ping fix; document the push-device monitoring model.
- log entries for both 2026-06-26 sessions.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Two genuinely-offline QR readers showed GREEN: the adapter's healthCheck was
hardcoded to { ready, "stub" } and never probed. These are PUSH devices (scan →
GET our backend, resolve by serial) with NO TCP port, so a connect probe has
nothing to hit — the stub "solved" that by lying. False-healthy is the worst
failure for a status bar.
- Optional reader IP field (monitor-ONLY; scans still resolve by serial,
operation unchanged).
- Unprivileged ICMP ping (drivers/icmp.ts): shells /bin/ping -c1, exit-0 = reply.
No native dep, no CAP_NET_RAW. docker-compose.prod.yml sets
net.ipv4.ping_group_range so it works for the non-root container user.
- healthCheck: replies → ready, no reply → offline, NO IP → degraded
("set IP to monitor") — never a false green.
Verified on hardware: readers (10.0.10.7/.8) answer ICMP on the device VLAN;
UI Test connection → "● ready — ping 10.0.10.7". Tests: reader.test.ts (4).
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
When a camera has Alarm Server push enabled, the setup form now shows the
camera's Alarm Settings (Destination IP / URL / Protocol / Port) ready to copy,
so the operator never hunts the deviceId or memorises the endpoint.
CRUCIAL: host/port come from the BACKEND address on the camera's subnet
(backendIpForDevice + the server's listen port — the same probe the push-IP
picker uses), NOT window.location.origin (the SPA's dev/proxy origin, which
would wrongly say localhost:5173). Verified live: matches the on-camera config
field-for-field (10.0.10.203 / …/event / HTTP / 3000). Shows a "save first"
(needs a deviceId) then "test first" (needs the resolved backend IP) hint.
i18n keys added to sq + en (parity enforced).
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
A Hikvision DS-2CD1047G3H-LIU returned HTTP 503 (statusCode 2 / deviceBusy)
on EVERY main-stream snapshot — its main encoder is persistently saturated.
Probed on hardware: channels/101/picture → 503 on 5 consecutive tries, while
channels/102/picture (sub stream) → 200 clean JPEG every time. A retry loop
can't fix a persistent busy; the real fix is stream selection.
- Add a `stream` config field to the Hikvision driver (1=main, default for
back-compat; 2=sub). ISAPI channel id is <channel><stream> (101 main, 102 sub).
Verified live: setting the G3H to Sub flips its status degraded→ready (14.7KB
JPEG in ~87ms).
- captureSnapshot also retries the TRANSIENT case (503/500, linear backoff
250/500/750ms ×4) then fails naming it "(device busy)"; does NOT retry 401/404
(config errors won't self-heal). Complements captureSnapshotShared (concurrent
de-dup). healthCheck still reports a live 503 as degraded (surfaces a saturated
main stream rather than hiding it).
Tests: camera.test.ts (10) — retry behaviour + main/sub path selection.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
The dev box runs vision as bare `uv run uvicorn`, and a plain uv run/uv sync
re-resolves the venv to the lockfile DEFAULTS, stripping fast-alpr/onnxruntime.
So after any `pnpm dev` real ANPR silently degraded to "snapshot, no plate"
(diagnosed 2026-06-25: real reads through 06-22, venv frozen lean since 06-19,
no other env with fast_alpr). The BOOTH was never affected — it runs the Docker
image, which bakes `uv sync --frozen --extra alpr` at build (immutable, weights
pre-warmed); a booth ModuleNotFoundError is a STALE image (fix: booth.sh update).
Vision package.json dev/start/recognize now run `uv sync --extra alpr &&` first
so pnpm dev is self-healing; added a dev:stub escape hatch for a lean run.
Documented in wiki/decisions/vision-service-packaging.md ("Two runtimes, one
fragile") + a log entry.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
On a vehicle entry, two paths captured the SAME Hikvision camera within ~1s —
the ANPR bridge (barrier-driving) and the advisory snapshotAsync (evidence/
telemetry) — each from a separate adapter instance. Hikvision serves snapshots
single-threaded, so the second concurrent GET returned HTTP 503; the bridge
then fail-softed and burned its 12s debounce, producing a ~74s "slow" subscriber
entry (observed 2026-06-25, Qazim Mulleti / AB816NN — plate read was instant at
conf 1.000; the delay was the 503/debounce churn, not recognition).
Add captureSnapshotShared() in snapshot.ts: a module-level, deviceId-keyed cache
that both paths call. It coalesces in-flight captures (the 2nd caller awaits the
1st's pull → no concurrent 503), serves a brief freshness window (1500ms) so the
bridge→advisory sequence for one vehicle reuses one frame, never caches a failure
(next caller retries), and keys by deviceId (no cross-camera/stale-vehicle reuse).
Wired into anpr-entry.ts (bridge) and snapshot.ts (advisory).
Tests: snapshot.test.ts (concurrent coalescing, TTL reuse, TTL-lapse re-pull,
failure-not-cached, per-camera keying); anpr-entry.test.ts mock updated. 168
server tests green.
NOTE: this removes the latency (the 503 collision). The separate double-entry
(two signed vehicle_entry for one car) — debounce-too-short / stamp-before-
success — is still open; less likely now but not eliminated.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
The booth PC (Ubuntu) needs one command instead of the long
`docker compose -f docker-compose.yml -f docker-compose.prod.yml --env-file .env …`
over the three compose files.
scripts/booth.sh — prod by default (ENV=dev for the dev override):
up/down/restart/status/logs/pull/config/exec, plus the requested `update` =
pull the moving branch tag → up -d --remove-orphans (recreates only
digest-changed services; named volumes / the SQLite ledger are preserved) →
docker image prune. Prod refuses to run without .env (no safe JWT_SECRET
default); dev with no .env injects the documented benign local secret (the
base file makes JWT_SECRET shell-required via ${JWT_SECRET:?}). down never
passes -v (would wipe the signed-ledger volume); help/unknown-command
short-circuit before any Docker/.env requirement.
.env.example — the vars the compose files consume (REGISTRY, TAG, JWT_SECRET,
EVENT_SIGNING_KEY, COOKIE_SECURE=0, WS_ALLOWED_ORIGINS). .env stays gitignored.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
New concepts/printer-usb-transport.md (the seam, usblp char device,
reachability-only status, threat model). open-questions #14: confirm the
on-site printer is USB and bake the usblp + udev write-access rule into the
appliance image (provisioning, not app code; unverified on hardware). Updated
rongta-printer.md (USB transport note), index.md, log.md.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
The wizard rendered every configField in a flat loop, so the USB device path
showed under a Network printer (and host/port would show under USB) — the
form could mislead. Add a transport-aware filter (mirroring the existing
pulseMs/inputRestingHigh skip): when Connection=USB hide host/port/httpPort,
otherwise hide devicePath. Verified live (Playwright): each transport shows
only its own fields and toggling swaps them.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
The ESC/POS printer drivers were TCP-only — every path went through
sendRaw/probe to a raw socket on port 9100. Add a USB transport behind
the existing render layer without touching a single render*() function.
- printer-escpos.ts: sendRawUsb/probeUsb write the same ESC/POS bytes to a
kernel usblp char device (/dev/usb/lp0) via a plain fs write — no
libusb/CUPS/native dep (keeps MIT-only + minimal-deps appliance). A
discriminated Transport + transportFromConfig/sendTo/probeTo dispatch the
wire; anything not transport:"usb" is TCP, so existing host-only configs
need no migration. Shared transportField/devicePathField config fields.
- cashino + rongta resolve a Transport once; both are reachability-only over
USB, and the Rongta's HTTP status page degrades to the open-the-node probe
over USB (no guessed paper/cover — the standing honesty rule). host/port
made not-required so a USB printer needs neither.
- Tests: printer-escpos.test.ts (USB writes the exact rendered bytes; probe
present/absent; transportFromConfig TCP back-compat) + printer-cashino.test.ts
(USB-configured driver prints to the node, ready/offline).
USB itself is unverified on hardware (the on-site printers are networked);
the appliance-side usblp + udev provisioning is tracked as open-questions #14.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
- dingtian-relay: the "offline despite ping" gotcha (relay_pw in every binary frame,
missing form field → Test connection sent 0 → timeout) + the identity-gated secret
re-merge that stops a redirected probe exfiltrating the password.
- button-light-indicator: serialized desired-state worker (UDP is unordered → the lamp
stuck on/off) and hot-reload of the lamp config (no restart).
- log entry for the three fixes (commits 420542c / fd15988 / 830993b).
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Two bugs in the button-light controller:
1. Stuck relay (random on/off). The blink fired fire-and-forget setAux every 500ms over
UNORDERED UDP with no serialization — concurrent on/off packets reordered/overlapped,
so the relay latched on whichever packet the device processed last. Replace with a
desired-state + serialized worker (#pump): the blink timer only flips desiredOn; a
single in-flight send per lamp is guaranteed, and on completion it re-converges to the
latest desired state — so the final state is always authoritative and a lost/stale
packet self-corrects.
2. Lamp ignored until restart. The lamp map was built once at start(); a button light
added/changed via the UI never took effect without a server restart. #reconcile now
re-reads the device config (at start and before each event, like DeviceMonitor),
adding/updating/dropping lamps live — so a just-saved lamp blinks on the next radar
edge.
Tests assert confirmedOf() (the device's latched state); +1 reconcile-after-start case.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
The controller editor mixed outputs and inputs in one flat "Relays" block — relay
direction, the entry-button terminal, and the presence/radar terminal all on the same
row, with the lamp orphaned below. Reorganize into two labelled sections:
- Outputs — relays (barriers + lamp): relay # + direction, the button-light relay, and
"Pulse open (ms)" (a relay hold-time, NOT an input setting — answers a recurring
confusion).
- Inputs — terminals (button, sensor): per entry relay, the button + presence/radar
terminals (kind, active-low) and cooldown, each labelled "For relay N", plus the
board-wide "Inputs idle HIGH".
UI-only: storage stays config.relays[] (+ config.buttonLight), so saved booth configs
keep working with no migration. pulseMs/inputRestingHigh are pulled out of the generic
field loop and rendered in their section. i18n parity (sq + en).
Also passes the device id to testDevice() so an edited device's stored relay password
re-merges on Test connection (pairs with the secure-merge server change).
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
The relay control password (relay_pw) was read by the driver but had NO form field,
so Test connection sent it as 0 → the device ignored the probe → a controller showed
"offline" even though it pinged. Add a "Relay control password" config field (secret;
blank keeps the stored value).
Because relayPassword is redacted from the client, the edit form can't resend it — so
the test endpoint now re-merges the stored secret by device id (mirroring save). It is
re-merged ONLY when the submitted config addresses the SAME device: matching driverId
and every connection-identity field it sets (host/port/binaryPort/httpPort/serial). A
redirected host/port or mismatched driver yields NO secret, so a probe can't exfiltrate
the password to an attacker host (the booth operator is the threat-model adversary).
testDevice() now passes the device id; setup-secrets.test.ts covers the identity guard.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Model the entry button (I1) and a Hikvision radar (I2) as named children of the
access controller, and drive the button's 12V lamp on a spare relay.
- Radar = the existing relays[].presenceInput one-car-one-ticket gate, now labelled
presenceKind: loop|radar. A radar may idle opposite the button, so add a per-input
active-level override: relays[].presenceActiveLow -> driver inputActiveLow set,
inverting just that terminal (pure helper inputActive()). The Dingtian has one
board-wide resting level otherwise.
- AuxOutputDevice.setAux(channel,on) capability on the device interface (Dingtian
latch) so business logic drives a NON-barrier lamp through the interface. Barriers
still only pulseOpen — barrier-not-a-door preserved.
- ButtonLightController: subscribes to the radar input edge + the camera lane status
and drives a 3-state lamp — radar+car=solid, radar-only=blink (~1Hz), else off.
Fails OFF on host loss/error; de-duped. A radar detection never opens a barrier on
its own (advisory; threat model).
- SetupWizard: presence kind + active-low + a button-light relay picker; sq+en i18n.
Tests: button-light.test.ts (truth table + blink + fail-OFF + de-dupe),
access-dingtian.test.ts (active-level inversion). Workspace build+lint+test green
(158 server tests). Wiki: hikvision-radar, button-light-indicator + updates.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
actions/upload-artifact@v4's backend fails on the Gitea runner (Upload installers
step errored). Mirror release.yml's proven path instead: curl + the built-in token
to the Releases API, into a ROLLING per-branch prerelease (tag desktop-<branch>,
deleted+recreated each push). Installers renamed space-free
(parking-desktop-<branch>-<sha>.{deb,AppImage}). Signed v* releases unchanged.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
createUpdaterArtifacts:true (for release.yml's .sig signing) makes `tauri build`
demand TAURI_SIGNING_PRIVATE_KEY and fail without it — even though the .deb/.AppImage
built fine. Override it off for the unsigned per-commit build via
--config '{"bundle":{"createUpdaterArtifacts":false}}'. release.yml keeps signing.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Self-service profile: any signed-in user edits their OWN fullName/email and
changes their OWN password (proving the current one), without any user:*
permission. New routes PUT /api/auth/profile + /api/auth/password act only on
req.user.sub (cannot touch username/role), CSRF-guarded; SPA screen at /profile
reachable from the header username chip. email added to the session view +
SessionUser. 7 tests (routes/profile.test.ts); 148 server tests green.
Desktop in CI: new .gitea/workflows/build-desktop.yml builds .deb + .AppImage
on every push to dev/main and uploads them as unsigned workflow artifacts
(per-commit test build). Signed/versioned release stays on release.yml (tag v*).
Wiki: local-jwt-auth (self-service routes), desktop-shell-tauri (two-workflow CI
split), log entry.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Operators/admins reach the booth at http://<name-or-ip>/ (no :3000). Adds a caddy:2-alpine
proxy to the prod override that reverse-proxies :80 → server:3000 (the /api/ws WebSocket
upgrades pass through natively); the server is now `expose: 3000` (internal, no published
port), vision stays internal. The Caddyfile binds `:80` so it matches ANY hostname/IP —
works for the booth IP, localhost, AND parksystems.msai.al (pointed at the booth via
hosts/DNS on-site; no domain baked into any image). TLS later = swap `:80` for the real
hostname + uncomment :443 → Caddy auto-provisions HTTPS.
Pairs with the relative-/api SPA fix (77b2acb): together verified end-to-end locally —
through Caddy on :80 with Host: parksystems.msai.al, GET / serves the SPA, assets/health
200, and POST /api/auth/login reaches the server (real 401, no CORS/connection error).
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
apps/web/.env.production sets VITE_API_BASE=http://127.0.0.1:3000 for the TAURI
desktop build (which loads from tauri://localhost and needs an absolute backend
origin). But Vite auto-loads .env.production for ANY `vite build`, so the server
image baked 127.0.0.1:3000 into the browser bundle — loading the UI from a real
host (e.g. http://parksystems.msai.al) then made the browser call 127.0.0.1:3000
cross-origin and fail the Same-Origin Policy on /api/auth/login.
Fix: the server Dockerfile writes apps/web/.env.production.local with an empty
VITE_API_BASE before the web build (.local has higher Vite precedence), so the SPA
served by Fastify stays relative/same-origin (/api/...). The desktop build is
unaffected (it doesn't use this Dockerfile). Verified: 127.0.0.1:3000 no longer in
the built bundle; /api/auth/login is relative.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
The base compose only forwarded DATABASE_URL/VISION_URL/JWT_SECRET, so a booth deploy
was missing the vars that actually make it usable on the plain-HTTP LAN:
- COOKIE_SECURE (default 0) — without it auth cookies are HTTPS-only and operators
CANNOT log in over http. The #1 booth-deploy footgun.
- WS_ALLOWED_ORIGINS — the live-feed WS rejects the browser Origin without it.
- EVENT_SIGNING_KEY — dedicated ledger key (falls back to JWT_SECRET if empty).
- VISION_ENABLED=1 — the server's ANPR master switch.
All driven from .env; verified via `docker compose config` that the seven vars resolve.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
The Gitea runner can't reliably resolve the astral-sh/setup-uv@v5 action — the
"Set up uv" step failed (exit 1) in build-images.yml (and the same step exists in
ci.yml). Replace the action with uv's official standalone install script
(`curl -LsSf https://astral.sh/uv/install.sh | sh`) + add $HOME/.local/bin to
$GITHUB_PATH, matching how the rest of the pipeline provisions tools (apt, corepack).
No third-party action dependency. Verified the install method yields a working uv on
a clean HOME. Same fix in both workflows.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
New wiki/decisions/appliance-provisioning.md: the hardware-verified step-by-step for
provisioning a booth PC (Dell OptiPlex 7070, i5-8500, discrete Nuvoton TPM 2.0) from
factory Windows to a hardened Ubuntu 26.04 LTS appliance. Every command was run on the
first real unit (2026-06-23). Captures the firmware-specific gotchas: Ventoy → 0x1A under
Secure Boot (flash ISO directly); the 7070 BIOS can't view db (verify via live USB); the
installer's hardware-backed encryption fails with PCR_UNUSABLE/dbt (use passphrase LUKS +
manual systemd-cryptenroll PCR-7 seal); GRUB password must be edit-only (--unrestricted)
to keep unattended boot.
OS hardening on unit 1 is COMPLETE + verified: LUKS FDE + TPM auto-unlock (PCR 7,
unattended) + Secure Boot (Deployed) + GRUB edit-lock (closes the init=/bin/bash root-shell
hole that PCR-7 sealing does not cover). Resolves the implementation half of
open-questions #12 for unit 1.
Cross-linked from disk-os-hardening; index + log updated. Still TODO on the box: Docker +
run the stack.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Containerize the two non-desktop apps for the booth appliance. The desktop app stays
on its own tag-only release.yml.
- apps/server/Dockerfile: multi-stage node:22-alpine. `pnpm deploy --legacy --prod`
(NOT prune — the monorepo native better-sqlite3 won't resolve under a root prune)
yields a self-contained bundle; build stage adds node-gyp toolchain, runtime adds
libstdc++; non-root, healthcheck. Migrates the mounted DB on boot via a drizzle-kit-
free runtime migrator (packages/db/scripts/migrate-runtime.mjs) — drizzle-kit is a
devDep, pruned from prod.
- apps/server/src/static-spa.ts: Fastify serves the built React SPA (one container
serves API + UI). GET-only fallback to index.html, excludes /api + /health so it never
shadows the backend; a no-op in dev (no dist). Registered last in server.ts.
- apps/vision/Dockerfile: uv base, --extra alpr, model weights PRE-WARMED into the image
as the runtime user so fast_alpr boots offline (0 downloads at runtime). Engine env-
selected (VISION_RECOGNIZER stub|fast_alpr).
- Branch-aware: docker-compose.yml (base) + .dev.yml (build local, stub, ports) +
.prod.yml (pull pinned, fast_alpr, vision internal, restart always); REGISTRY/TAG from
env so a branch deploy pulls that branch's image.
- .gitea/workflows/build-images.yml: on push to dev/main, run the full turbo build+lint+
test gate, then buildx push both images to git.infra.msai.al/mca/parking_solution with
branch + branch-<sha> tags (registry cache; optional Komodo webhook behind KOMODO_ENABLED).
- .dockerignore excludes **/parking.sqlite* so the signed ledger is NEVER baked.
Verified locally (Docker 29): server image migrates + serves API+SPA (/health 200, /
+ /booth HTML, /api/nope JSON 404, no sqlite outside /data); vision image boots fast_alpr
with 0 runtime downloads; compose stack healthy with server→vision over the private network.
Wiki: new container-deployment.md; vision-service-packaging open Qs resolved; index + log.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
`@parking/server#lint` (tsc --noEmit) failed in CI with "Cannot find module
'@parking/db' / '@parking/shared'" + a cascade of implicit-any errors. Root cause:
the root turbo `lint` task had no dependsOn, but those packages expose their types via
"./dist/index.d.ts" — only present after their `build` runs. In a clean CI tree lint
ran before the deps were built, so tsc couldn't resolve them. Locally it passed only
because a prior `dist/` happened to exist. Make `lint` depend on `^build`, exactly
like `typecheck` and `test`. Verified from a fully clean tree (rm dist + .turbo +
*.tsbuildinfo): `turbo run build lint` → 14/14, 0 cached.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Cancel a misprinted/test/wrong-vehicle ticket via a SIGNED `void` event — the
vehicle_entry is never edited/deleted (append-only). VoidFlow appends void{
voidedEntryRef, voidReason, operator, reasonCode:"void.ticketCancelled" }; route
POST /api/tickets/void gated event:void + open shift; reason REQUIRED. Refuses a
subscription / already-exited / already-voided / paid ticket (refund out of scope).
The void folds the session CLOSED everywhere it's counted — occupancy (count +
reserved spots), pay-station (lookup/activeSessions), exit-flow (#sessionFor), and
reports (excluded from entries) — so a voided car stops occupying a spot, can't be
paid/exited, and doesn't inflate "cars entered". No barrier action. Booth UI: a
"Cancel ticket" action in the pay/exit lookup modal (transient + unpaid + open;
gated on event:void) with a preset-or-free reason prompt.
Reclassify the Live feed: refused-action events (exitRefused/entryRefused/
permitRefused — e.g. a double card-scan, at-capacity subscriber, exit on a closed
session) are benign warnings, not red anomalies. event-detail.tsx now shows them as
amber REFUZUAR/REFUSED, reserving red ANOMALI for genuine red-flags. Display-only —
no ledger change, so historical events reclassify too.
CI: install uv + sync vision deps before the Turbo run. @parking/vision's lint/
typecheck/test shell to `uv run …`, but CI set up only Node+pnpm, so `uv run ruff`
failed ("uv not found") and broke the whole Turbo run. The Python checks pass once
uv provisions the toolchain.
- new: void-flow.ts (+ tests, 8) ; occupancy void-fold test
- shared: reason code void.ticketCancelled ; both web catalogs (sq/en parity)
- wiki: parking-session (ticket-void folds + guards, refused/anomaly split), log
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Wire the lane camera's vehicle event into the gated subscription flow: on a
vehicle/active push from an opt-in (config.anpr) camera, AnprBridge pulls a fresh
snapshot, runs ANPR, applies a stricter entry confidence floor, debounces, and —
matching the plate to a subscription BEFORE emitting — emits a kind:"plate" read.
The existing ReadDispatcher -> SubscriptionFlow then signs the entry/exit and opens
the barrier. A plate is never the sole authority: it routes through the same gate
(active/window/blocklist/car-count) as any credential. Fail-soft, fire-and-forget,
subscriber-only by construction. Field-verified end to end (plate AA504LX opened the
entry barrier and appended a signed vehicle_entry).
Add an admin master switch (site_config.anpr_entry_enabled, default ON) in Site
Settings that disables ONLY the barrier-driving bridge; advisory snapshot-ANPR and
lane busy/free are unaffected. Read live per event, so toggling takes effect with no
restart. Migration 0013 (additive ALTER ADD COLUMN, default 1).
- New: apps/server/src/anpr-entry.ts (AnprBridge) + tests (9)
- hikvision-alarm.ts hands vehicle detections to the bridge (fire-and-forget) + wiring tests (3)
- server.ts reorders the read flows above the hik-alarm registration
- snapshot.ts exports buildCamera for reuse
- env: VISION_ENTRY_MIN_CONFIDENCE (0.85), ANPR_DEBOUNCE_MS (12000)
- site route + SiteSettings checkbox + i18n (sq/en parity)
- wiki: lane-presence-and-anpr-entry / lpr-camera / index / log -> BUILT
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Captures this session's back-and-forth as a new concept page
[[lane-presence-and-anpr-entry]] and cross-links it:
- BUILT: advisory lane busy/free booth lights (LaneStatus + WS), with the
measured camera limits behind the 30s timeout (no leave signal; movement-
driven re-fire; notificationRecurrence locked to "beginning" — ISAPI flip
silently reverts).
- PLANNED: the ANPR "bridge" — explicitly a small apps/server HANDLER (~40
lines), NOT a new service/container. On a camera vehicle event: snapshot ->
ANPR -> high-confidence match -> debounce -> emitRead{kind:"plate"}, then
the existing subscription match/dispatch/gate admits the subscriber. Both
directions, opt-in (config.anpr), plate never the sole authority.
- Records the decisions (high confidence floor, debounce-for-correctness)
and the REJECTED ideas (continuous livestream / per-car queue tracking /
make-model) with why, plus the open hardware question (booth-PC test).
Updates subscription.md (plate matching is built; the live source is this
bridge) and lpr-camera.md (the two consumers of the vehicle event).
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Controlled in/out test on the camera: the `active` re-fire rate is
MOVEMENT-driven, not steady — ~1-3s apart while the car moves, but up to
~15-25s when it sits MOTIONLESS in the zone. A 5s TTL would flicker a
parked car free; the TTL must exceed the still-car gap. The camera has
~no dwell lag (goes silent within ~1s of the car leaving — measured: last
event 16:15:17 vs car-left ~16:15:30), so 30s keeps a motionless car busy
while clearing promptly after departure. This also confirms vision-based
tracking isn't warranted: the camera's leave signal is already tight.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Measured the real re-fire rate on the camera: while a vehicle is in the
zone it POSTs `active` about every ~1 second (not the ~30-80s I'd guessed).
The camera sends no leave signal, so "free" is timeout-driven — but with a
~1s re-fire, 90s made the lane stay red for a minute and a half after the
car left. 5s of silence reliably means the car is gone; the light now
clears within seconds. Still override-able via LANE_BUSY_TTL_MS.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
A Hikvision vehicle detection (eventType=VMD, targetType=vehicle) on a
camera bound to entry/exit now marks that lane "busy" and shows it as a
barrier light beside the scan input on the booth (green=free, red=busy).
Advisory only — it gates nothing (never blocks a ticket or opens a barrier).
- Parse eventState (active/inactive) from the Hik payload.
- LaneStatus tracker: a vehicle `active` event marks the camera's bound lane
busy + arms an auto-clear timer. This camera class sends no leave/`inactive`
signal, so "free" is timeout-driven (LANE_BUSY_TTL_MS, default 90s; the
camera re-fires `active` while a car sits there, refreshing the timer). A
"both"-direction camera marks both lanes.
- Push lane-status over the existing booth WS (+ in the hello snapshot);
live-store holds { entry, exit }; two BarrierLight icons render it.
- i18n booth.laneEntry/laneExit (sq + en).
Tests: lane-status.test.ts (7 — busy/free, TTL auto-clear, timer re-arm,
no re-emit while busy, both/exit direction, unknown device). server 120/120;
web + server build/lint green.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Vite had no host set (localhost only). Bind 0.0.0.0 so the dev booth UI is
reachable from other LAN devices at http://<host-lan-ip>:5173. The SPA
already uses relative paths + the page origin for API and the live WS, so
no app code changes — but loading from a non-localhost origin means the
/api/ws handshake's Origin is the LAN address, which the backend's
WS_ALLOWED_ORIGINS must include (documented in .env.example; the host's own
.env is gitignored).
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
The Hik DS-2CD1043G2-LIU was NOT defective. An earlier wiki entry wrongly
concluded it needed RMA (dead event engine) based on a silent alertStream
+ diskfull/EventScribe:except + dead RTC surviving a full factory reset.
Real cause: no detection AREA was drawn on the frame. With no region, the
camera detects nothing -> generates no event -> posts nothing. The instant
an area was drawn, the first vehicle produced a clean POST.
- Flag "draw the detection area" as the FIRST thing to check.
- Document the confirmed real payload: multipart/form-data (MoveDetection.xml),
EventNotificationAlert with eventType=VMD, eventState=active,
targetType=vehicle (vehicle/human classified on-device), targetRect bbox.
Note the dateTime is garbage (dead RTC) -> use our own receive time.
- Reframe the SSH diagnostics: diskfull/EventScribe/RTC are RED HERRINGS,
not proof of a dead camera; don't escalate to hardware fault while a basic
config precondition is unmet.
- Append a log correction (append-only) superseding the earlier conclusion.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Captures the hard-won findings from the field session: the WSL source-IP
rewrite + skipSourceIpCheck fix, the boolean-as-string setup bug, the
unreliable "Test" button, the latching httpBroken flag, and Notify-
Surveillance-Center vs HTTP-Alarm-Server.
Adds a "diagnose a non-pushing camera from its OWN state" runbook
(alertStream heartbeat silence, SSH showStatus EventScribe:except, dmesg
RTC/UBIFS, netstat outbound watch) and documents the verified-dead
DS-2CD1043G2-LIU unit (defective event engine, survives factory reset ->
RMA), with the pull+vision fallback.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Diagnosed why no camera push ever landed: (1) the route only registered
POST/GET, so a probe with another method got a generic 404 the camera
reads as "service available" while our handler never ran; (2) more
fundamentally, WSL mirrored mode REWRITES the inbound source IP to the
host's own address (10.0.10.203), so the camera's real IP (10.0.10.12)
never survives and the source-IP guard rejected every push as a mismatch.
- Register the event route on POST/GET/PUT/PATCH/DELETE/OPTIONS (HEAD comes
with GET) so ANYTHING hitting the path reaches the handler and is recorded.
- Log + store the HTTP method of each hit; log every hit on arrival, before
any guard, so even a rejected probe is visible immediately.
- Add per-device skipSourceIpCheck (a Setup checkbox) to bypass the
source-IP guard where the network rewrites the source (WSL). Digest auth +
the signed ledger remain the real guards.
Tests: hik-alarm 10 (skip-IP accept + method capture). server green;
web build green (new checkbox renderer + this field).
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
The generic config-field loop had no boolean branch, so a type:"boolean"
field (e.g. the camera's alarmPushEnabled) fell through to a TEXT input and
saved the STRING "true" instead of a real boolean. Downstream checks use
=== true, so the feature read as disabled even when the admin ticked it.
- Web: render type:"boolean" config fields as a real checkbox; store/merge
a true/false boolean (and persist false on edit so toggling off sticks);
normalize a legacy string "true"/"false" on load.
- Server: isOn() coerces the flag when reading config (accepts true/"true"/
1/"yes"/"on") so an existing row saved as the string "true" still works
without a re-save, and no other boolean field hits the same trap.
Tests: hik-alarm accepts string "true" for alarmPushEnabled. server
112/112; web typecheck + build green.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Debugging "is the camera event coming or not?" was painful: a rejected
push only logged a warning and recorded nothing, so "no event" was
ambiguous (never sent vs sent-and-refused), and the only durable record
was an unreadable device_events row.
- Record EVERY push, accepted or rejected: accepted -> kind:"alarm",
rejected -> kind:"alarm-rejected" with the precise reason (unknown
device / not-hikvision / push-disabled / source-IP mismatch / digest
fail). The 404 body now also returns the reason.
- New GET /api/devices/hikvision/alarms (device:read): the recent pushes
newest-first as JSON (accepted+rejected, with ip/reason/eventType/
target/plate/rawHead) so you can SEE arrivals in the browser instead of
grepping the dev log or querying SQLite.
Tests: hikvision-alarm.test.ts now 8 (rejection-recorded + read-endpoint
list + gating). server 111/111; build+lint green.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
Newer Hik firmware can PUSH events to us: Event -> Smart/VCA with
"Detection Target: Human/Vehicle" + Notify Surveillance Center + Alarm
Settings -> Alarm Server makes the camera HTTP-POST an
EventNotificationAlert on each detection.
- New POST /api/devices/hikvision/:deviceId/event (routes/hikvision-alarm.ts):
same machine-push pattern as the Dingtian Input Link — source-IP guarded
+ optional HTTP Digest, not behind the SPA cookie/CSRF.
- Discovery-first / permissive: a wildcard content-type parser accepts ANY
body as raw bytes (event XML, multipart+JPEG, or JSON — Hik varies by
firmware), records it verbatim as a kind:"alarm" device_event, and
best-effort extracts eventType/target/plate/dateTime/channelID for the
summary + a loud log line. The point is to SEE exactly what a camera
sends before wiring it further.
- hikvision driver gains alarmPushEnabled + pushUser/pushPassword config and
pushesToBackend:true (setup offers the backend push IP).
- NOT yet a barrier trigger / DeviceReadEvent — records only. A plate read
is advisory, never the sole reason a barrier opens; the read-bus/ANPR
wiring is a deliberate next step once the real payload is known.
Tests: hikvision-alarm.test.ts (6: vehicle XML summary, ANPR plate, raw
JSON, wrong-IP 404, disabled 404, unknown-device 404). server 109/109;
build+lint 14/14. Wiki: lpr-camera.md + log.
Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
julian