Kubernetes and `docker stop` send SIGTERM and then SIGKILL once the grace
period expires; previously oott ignored it and was always force-killed.
Add a shutdown module that watches for SIGTERM (and SIGINT/Ctrl-C) and
cancels a CancellationToken shared by every long-running task. Each scanner,
the retention cleaner, and the notification delivery loop now select on the
token and stop at their next safe point (abandoning in-flight probes and
skipping inter-scan sleeps), the web server drains in-flight requests via
axum's with_graceful_shutdown, and main checkpoints the WAL via db::close()
once all tasks have stopped.
No internal deadline is added: the orchestrator's grace period already bounds
shutdown, and tasks stop on iteration boundaries so there is no half-written
DB state.
Refs #7
Scanner status handlers all derived operationId "status" from their fn
name, so Swagger UI's "Try it out" executed the first one (ARP) — the
DHCP doc hit /api/arp_scanner/status. Same collision for read/list/
register/unregister. Each path now sets an explicit unique operation_id.
main's tokio::join!(...).0 kept only the ARP result and silently dropped
the other tasks' errors, so a DHCP scanner that failed to bind port 67
just showed "off" with no log. Each task is now wrapped to log its error.
Fixes#5
record_sighting only carried the stored name and IP onto the reconciled
sighting, leaving is_registered, owner, and device_type at their bare
defaults. The change handed to the notification layer therefore rendered a
registered, typed device as "Not registered" with type "-". Carry those
fields over too, mirroring how db::devices::seen() preserves them.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Backend:
- Add db::devices::delete to erase a device and its events atomically
- Expose DELETE /api/devices/{mac}/permanently, wired to OpenAPI
- Cover the new db method and endpoint with tests
Frontend:
- Delete action for not-registered devices (detail screen + list row)
and an opt-in "permanently delete" checkbox in the Forget dialog
- Navigate to the devices list after deleting from the detail screen
- Refine button emphasis to M3: single filled primary, error-colored
text buttons for destructive actions, Test demoted to filled-tonal
- Flash the backend-config Test button red on a failed connection test
- Render snackbars through a top-level ScaffoldMessenger host so they
show above dialogs; keep the built-in SnackBar (with an Overlay host)
Docs:
- CLAUDE.md: rustfmt edition 2024, don't revert formatter-only changes,
prefer built-in Flutter components, follow existing patterns + M3
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
POST /api/notifications/test previously returned a blanket 200 even when there
were no registered devices, so a test that reached nobody looked like a success.
push::send now returns the number of devices the relay confirmed delivery to, and
the endpoint returns it as {"delivered": N}. Settings shows "sent to N device(s)"
on success and an explicit "No devices are registered..." warning when N is 0,
which is the case that previously masqueraded as success.
Backend, API and widget tests updated; clippy and dart analyze clean.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Adds POST /api/notifications/test, which delivers a canned test push to every
registered device through the existing relay path (not persisted to the
notifications list), wired into the router and OpenAPI. In Settings, a "Send test
notification" button appears under the push toggle, only when push is enabled on
this device, so the full backend -> relay -> FCM -> APNs -> device path can be
verified on demand without waiting for a real device event.
Backend, API and widget tests added; clippy and dart analyze clean.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Tier-1 dependency refresh (lockfiles only, no manifest/version-range
changes):
- frontend/pubspec.lock: dio, go_router, shared_preferences and others
to their latest in-range versions.
- backend/Cargo.lock: tokio 1.49->1.52.3, clap 4.5->4.6, tower-http,
serde_json and a batch of transitive crates.
All frontend (152) and backend (165) tests pass.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Add a GET /api/config endpoint exposing the front-end-facing backend
configuration (currently the notification delivery method, grouped under
a nested "notifications" object so the shape can grow). The settings
screen fetches it on init and only shows the per-device push toggle when
the backend method is "push" (and the platform supports push, which keeps
it off the browser).
Also fold in related push-notifications cleanups: fix the Android app
label ("frontend" -> "OOTT") so the notification permission dialog reads
correctly, remove the completed push_notifications.md plan, and update
TODO.md.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- flake.nix: add nodejs_22, firebase-tools, and google-cloud-sdk to the dev
shell so the relay can be tested, built, deployed, and administered locally
- backend settings: point default_relay_url at the deployed relay
- rename the relay liveness route /healthz -> /health: Google Front End
reserves /healthz and returns its own 404 before the request reaches Cloud
Run, so the probe was unreachable (verified live; /v1/push and the FCM path
work end-to-end)
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Backend (Rust):
- push_tokens migration, model (PushToken/PushPlatform), and db layer
(upsert/list/delete/delete_many)
- PUT/DELETE /api/push_tokens endpoints wired into the router + OpenAPI
- "push" notification method: relay sender (reqwest) that forwards only the
sanitized title/body and prunes dead tokens, plus settings with a default
relay_url
- 164 tests pass, clippy clean
Relay (push_relay/, TypeScript Firebase Cloud Function):
- POST /v1/push (firebase-admin sendEach + per-token status mapping),
GET /healthz, payload validation, per-IP Firestore rate limiting
- Jest tests + README documenting the manual project-owned setup
Frontend (Flutter):
- oott_api_push.dart (register/unregister), push_service.dart behind a
PushService abstraction, and a per-device push toggle in settings
- firebase_core/firebase_messaging/flutter_local_notifications deps
- 145 tests pass, dart analyze clean
Dev shell:
- add nodejs_22 to the Nix dev shell so the relay tests/build run locally
Remaining (manual, project-owned): create the Firebase project, deploy the
relay and set the real default_relay_url, add native Firebase config, and test
on real devices.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Without an explicit directive ServeDir let browsers heuristically cache
the Flutter bundle, and the service worker kept serving stale assets, so
new icons/images did not appear after an upgrade (even on hard reload).
Force revalidation so users always get the latest assets; unchanged files
still return a cheap 304.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The events.rs file mixed three domains: device-event recording, change
detection, and the entire notification pipeline (rendering + delivery +
sending). This made it long, gave functions side effects beyond their
stated goal (classify_* silently recorded events), and intertwined the
events and notifications logic.
Split along domain boundaries:
- model::device_events now owns DeviceChange, the shared contract.
- events records device events only (record_new_device/record_known_device);
events/detection.rs holds pure change detection.
- new notifications module owns rendering, delivery, and sending
(notifications.rs + delivery.rs + render.rs); pushover/error moved here.
Data now flows one way: events produces DeviceChange, notifications
consumes it, both depend only on model. Scanners/pipeline/main orchestrate.
classify_* renamed to record_* so the write is the stated goal; send and
send_notification collapsed into persist_and_deliver.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Record what happened on each known-device sighting instead of only
"seen": a baseline DeviceSeen (history heartbeat, no notification) plus
DeviceChanged and DeviceBackOnline events, each deduplicated
independently so a recent routine sighting no longer suppresses a
genuine change or return notification. The frontend chart now trusts the
event type for its marker and tooltip rather than comparing each event's
snapshot against the device's current state.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Gate notification persistence and delivery on the device-event
deduplication window, not just the device_events table: record_event now
reports whether it recorded, and classify_new_device/classify_existing_device
suppress their changes when the sighting is a duplicate.
Treat an empty->value IP fill as not a change via a new ip_changed helper
(mirroring vendor_changed), so a device gaining its first address raises no
"changed" notification.
Key deduplication on (mac, scanner, event_type) instead of (mac, ip,
scanner), so repeated sightings of the same kind collapse regardless of the
reported address.
Also delete stale oott.db-wal/-shm in run_tests.sh to avoid init_db panics.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
New devices are never registered, so the status line only ever read
"Not registered" and added no information. Remove it from the single
new-device notification body (the consolidated summary never had one).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The root URI had no content of its own and only showed a pointer
message; send visitors straight to the front-end instead.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Backend: notifications now show a plain "-" for an absent name, vendor, or
device type (was empty string / "(unknown)" / "Unknown"), via a single
UNKNOWN_PLACEHOLDER constant.
Frontend:
- Empty/unknown values render as an em dash everywhere, centralised in a new
Placeholders.emptyValue constant (replaces inline '—' and '(unknown)').
- Route paths moved to a new Routes class, used by the router and every
navigation call site.
- Device event type modelled as a DeviceEventType enum mirroring the backend
(NewDevice/DeviceSeen) instead of bare string comparisons.
- Hardcoded EdgeInsets/SizedBox spacing replaced with existing Insets tokens.
Tests and formatting updated; all backend and frontend tests pass.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The Flutter web bundle was built with the default base href of "/" but is
served under "/web/", so index.html loaded while every asset reference
resolved to the site root and 404'd, leaving a blank page. Build the bundle
with --base-href=/web/ so asset URLs match the mount point.
Also correct the root guidance text (the API explorer is at /api/docs, not
/api) and add an /api -> /api/docs redirect for convenience.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Active scanners (ARP, SNMP) now accumulate every change across a whole
scan and emit one notification per type via events::notify: a single
device produces the usual single-device notification (carrying its MAC),
while two or more produce one consolidated summary with an empty
mac_address. Device events are still recorded per device.
Notification bodies no longer include MAC or IP addresses; the title
MAC fallback is masked to the last two octets. Summaries list up to
three devices then "…and N more devices".
Split sighting handling so record_sighting persists + records the event
and returns Vec<DeviceChange>; passive listeners (mDNS, SSDP, DHCP) use
record_and_notify since they see one device per event.
Also fixes NotificationType::from_str never mapping "DeviceChanged",
which made those notifications round-trip from the DB as Other.
Frontend: the card already hides the device link when mac_address is
null; added widget tests for the present/absent link cases.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Enabling WAL per pooled connection raced several concurrent journal-mode
switches when r2d2 eagerly opens the pool at startup, producing transient
"disk I/O error" and "database is locked" failures. WAL is a persistent
property of the database file, so enable it once in init_db (before the
scanners and web server start) and keep only the genuinely per-connection
pragmas (busy_timeout, synchronous, foreign_keys) in the pool initializer.
Also remove the -wal/-shm sidecars in run.sh: deleting oott.db but leaving
a stale write-ahead log behind makes SQLite open the fresh database against
an orphaned log and fail with "disk I/O error".
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Tune SQLite and remove blocking calls from async/scan paths:
- Enable WAL + synchronous=NORMAL + busy_timeout + foreign_keys on each
pooled connection, so the five scanners, web server, and retention no
longer contend on the default rollback journal / FULL fsync.
- Run DB work in axum handlers via spawn_blocking (db::run_blocking) so
synchronous rusqlite calls no longer block tokio worker threads.
- Deliver notifications on a dedicated task fed by a bounded channel; the
blocking Pushover HTTP call runs in spawn_blocking, so a slow or
unreachable Pushover can never stall device discovery.
- Make get_db_connection() return Result instead of panicking, so pool
exhaustion surfaces as a 500 rather than crashing the process.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Add ActiveStatusCell/PassiveStatusCell wrappers in the scanners common
module so the five per-scanner status.rs files reduce to a single static;
replace parse_parameter_bool/int/string with one generic parse_parameter
over FromStr; extract the shared LIMIT/OFFSET paging clause into
db::apply_paging; and drop a no-op for-loop in the ARP sender.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The ARP and SNMP scanners reported every online sighting, so a device
seen on multiple IPs or via duplicate ARP replies was counted more than
once. Fold the dedup into ActiveStatus::record_scan, which now takes the
device slice and reports the number of distinct MAC addresses.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The list endpoints now return a total count alongside the page so the
front-end can show how many pages exist and offer a last-page jump.
Backend: add count(is_new) and count_devices(...) (sharing a WHERE-builder
with list_devices so page and count can't drift), wrap both list responses
in {items, total_count} structs, and register them with utoipa.
Front-end: parse the wrapper shape (dropping the fetch-one-extra trick),
add a Last-page button and a responsive "Page X of Y" / "X / Y" label to
the shared PaginationBar, and track the total in both lists. Notifications
re-sync the count on every fetch and decrement it locally on mark-read/
unread removals so the count stays accurate without a re-fetch.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The pushover config section is now only required when notifications.method
is "pushover". Validation at startup rejects the missing-section case so a
misconfiguration fails fast instead of erroring on every notification.
Updates the sample TOML, README and nix module for consistency.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Several settings were required at parse time even though they had a
sensible default, causing the backend to panic on startup when omitted:
- notifications.notify_when_not_seen_for now defaults to "1w"
- the whole [networking] section is now optional (no mandatory fields)
- scanner duration/timeout fields (ARP, mDNS, SSDP, SNMP) now fall back
to their defaults when the section is present but the field is omitted.
serde only applies a field default when marked #[serde(default)], so
the per-field attributes were added and the Default impls now share the
same default functions as the single source of truth.
Updates the sample TOML and README accordingly and adds tests covering
each defaulting case.
Fall back to "0.0.0.0" and 3000 when these fields are omitted from the
[web_server] section (api_key remains required). Update the sample TOML
and README to reflect the new defaults; the Nix module already used them.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Make the [log] section and its level field optional, falling back to
"warn" when omitted. Update the sample TOML, Nix module and README to
reflect the new default.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The version string was duplicated across six places. Collapse it to two
ecosystem sources of truth and derive the rest:
- backend/src/web_server.rs: omit the OpenAPI info.version so utoipa fills
it from CARGO_PKG_VERSION (backend/Cargo.toml); add a test pinning this.
- nix/package.nix: read the version from backend/Cargo.toml via fromTOML.
- nix/frontend.nix: read the version from frontend/pubspec.yaml by splitting
into lines (a whole-file regex triggers catastrophic backtracking in Nix's
regex engine).
- frontend/lib/about/about.dart: read the version at runtime via
package_info_plus instead of a hardcoded constant.
Also drop frontend/pubspec.lock.json: it is unreferenced (Nix's
autoPubspecLock generates its own JSON from pubspec.lock) and was going stale.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
When the same scanner sees the same device (same MAC and IPv4) again within
a configurable window (default 1 minute), only one device_events row is
recorded, keeping the events table from filling with near-identical rows.
Device last_seen updates and notifications are unaffected.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The five scanners (ARP, SNMP, mDNS, SSDP, DHCP) duplicated their
persist-and-notify pipeline, device enrichment, and status state
machines across same-family files. Extract the shared logic so a change
lands in one place instead of three to five.
- scanners/common/pipeline.rs: single record_sighting() persist+notify
path, replacing the per-scanner match blocks. ARP/SNMP now use the
same merge rules as the passive scanners (keep a stored hostname,
never overwrite a known IP with an empty one).
- scanners/common/enrichment.rs: build_device() for vendor/device-type
lookup with the privacy-MAC service fallback.
- scanners/common/{active,passive}_status.rs: the two status state
machines plus their tests, written once. Each scanner status.rs is now
a thin wrapper over its own static.
- utils/network::format_mac(): replaces three identical copies.
- web_server/scanner_status.rs: Active/Passive response types and two
handler helpers, replacing five near-identical structs+handlers. JSON
field names are unchanged so the frontend is unaffected; only OpenAPI
schema names change.
27 files changed, ~900 lines net removed. Build, clippy and all 113
tests (4 new) pass.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Replace the lifetime "devices seen since start" counter in the mDNS, SSDP
and DHCP scanners with a rolling count of distinct devices (deduped by
MAC) seen within the last hour.
Each scanner's status now tracks a MAC -> last-seen-time map; the snapshot
prunes entries older than 60 minutes and reports the remaining count. The
API field name (devices_seen) is unchanged, so only its meaning and the
frontend labels ("N devices in the last hour") are updated.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The ARP and SNMP scanner status endpoints and front-end cards now expose
the number of devices found by the most recent successful scan, following
the existing mDNS device-count pattern. The count persists across the
running/waiting transitions, and for SNMP a failed poll keeps the last
good count rather than overwriting it.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Add an ArpScanner Default impl (30m/1m/10m, enabled) and mark the
arp_scanner field with serde default, so the [arp_scanner] section can
now be omitted entirely and fall back to code defaults — matching the
pattern used by the SNMP scanner. Previously these three durations were
mandatory and the backend would not start without them.
Reconcile the documentation to the canonical 30m/1m/10m: fix the README
NixOS example (was 15m/20m/30m) and options table (was 15m), and note
that the section is optional in both the README and sample TOML. The
NixOS module and sample TOML already used these values.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
A typical router keeps active devices' ARP entries fresh continuously,
so a 10-minute poll interval stays well within common ARP cache TTLs
while keeping device-event churn modest (each poll records a DeviceSeen
event per device). A 5s per-request timeout adds margin for a busy
agent or large ARP table at no cost on the happy path.
Updates the default in settings.rs and the sample TOML, README and
NixOS module. TODO timing-review item narrowed to the ARP scanner.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Introduce a sixth scanner that periodically queries an SNMP agent
(typically the router/firewall) for its ARP/neighbour cache via
SNMPv2c and feeds discovered devices into the shared devices, events
and notifications pipeline. Unlike the ARP scanner it generates no
traffic on the local segment and can surface devices across all
subnets the agent routes.
Scope is intentionally minimal: SNMPv2c only, a single target, and the
ipNetToMediaTable (ARP) only. SNMPv3 and switch MAC/forwarding-table
polling are left as follow-ups in TODO.md.
- backend: csnmp dependency; SnmpScanner config (opt-in, off unless a
[snmp_scanner] section is present); DeviceEventScanner::Snmp;
scanners/snmp/{finder,scanner,status}; main.rs wiring; status API
endpoint wired into OpenAPI
- frontend: SNMP scanner status model, card, API method, status screen
and summary card rows, and device-event label
- docs/config: sample TOML, README options, NixOS module option, and
setup notes for enabling SNMP on pfSense/OPNsense
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- Build the Flutter web app at release time (nix/frontend.nix) and bundle
it next to the binary at $out/share/oott/web; resolve it at runtime
relative to the executable. Remove the prebuilt backend/web from git.
- Add web_server.{ip_address,port,api_key} options to the NixOS module so
the generated config deserializes (was missing, causing a startup panic).
- Docker image: set SSL_CERT_FILE for outbound TLS, drop the heavy
nixos/nix base image, trim contents to [oott cacert], and ensure /tmp
exists.
- Remove the unused "nix" flake input and commit flake.lock.
- Provide Swagger UI to utoipa-swagger-ui offline via a pinned fetchurl so
the package builds in the Nix sandbox; skip the redundant check phase
(tests run via backend/run_tests.sh).
- sample_oott.toml: set database.path to /db/oott.db for the Docker image.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Each scanner (ARP, mDNS, SSDP/UPnP, DHCP) can now be turned off via an
`enabled` flag in its config section, defaulting to true so existing
deployments are unchanged. A disabled scanner's entry function returns
early and never starts.
Documented in the README options table (also fixing the stale `timings.*`
key names to the actual `arp_scanner.*` keys), and added to the TOML
samples and the NixOS module.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Listen for DHCP DISCOVER/REQUEST broadcasts on UDP 67 to catch devices as
early as possible — a device must request an address before doing almost
anything else, often before it has an IP.
Follows the mDNS/SSDP scanner pattern (finder/scanner/status modules) and
feeds the shared devices/events/notifications pipeline. The client MAC is
taken directly from the packet's chaddr, so no ARP probe is needed; a
DISCOVER with no assigned IP reuses any previously recorded address rather
than clobbering it. Exposes GET /api/dhcp_scanner/status, wired into the
OpenAPI generation, and adds a Dhcp variant to DeviceEventScanner.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Introduce a new SSDP/UPnP network scanner alongside the existing ARP and
mDNS scanners:
- New scanner module under scanners/ssdp (finder, scanner, status)
- Wire the scanner into the main scan loop
- Add Ssdp variant to DeviceEventScanner
- Add SsdpScanner settings with a configurable probe timeout
- Expose /api/ssdp_scanner/status and wire it into OpenAPI generation
- Add a lint.sh helper and point CLAUDE.md at it
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
rzuasti