Long-lived SSE connections prevent the gevent worker from exiting on
SIGINT. --graceful-timeout 5 force-kills the worker after 5 seconds.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Gunicorn's gevent worker (-k gevent) handles monkey-patching internally.
The manual patch_all() in app.py ran in the master process before worker
fork, preventing the worker from booting (no 'Booting worker' log line,
server unreachable).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Resolves ModuleNotFoundError when running outside a venv by auto-detecting
the venv/bin/python relative to the script, falling back to VIRTUAL_ENV
or system python3.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add start.sh as the recommended production entry point with:
- gunicorn + gevent worker for concurrent SSE/WebSocket handling
- CLI flags for port, host, debug, HTTPS, and dependency checks
- Auto-fallback to Flask dev server if gunicorn not installed
- Conditional gevent monkey-patch in app.py via INTERCEPT_USE_GEVENT env var
- Docker CMD updated to use start.sh
- Updated all docs, setup.sh, and requirements.txt accordingly
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Change 'already_running' to 'already_scanning' status in bluetooth_v2
so frontend recognizes the response and connects the SSE stream
- Hide pagerScopePanel and sensorScopePanel in switchMode() to prevent
audio waveform bars leaking into other modes
- Clear devices Map, pendingDeviceIds Set, and UI in BluetoothMode.destroy()
to prevent memory accumulation on repeated mode switches
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Previously all HackRF devices were hardcoded as "HackRF One" regardless
of actual hardware variant. Now parses the Board ID line from hackrf_info
to correctly identify HackRF Pro, HackRF One, and other variants.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Pass observer location and gpsd status to radiosonde_auto_rx station config
- Add station marker on radiosonde map with GPS live position updates
- Display distance from station to each balloon in cards and popups
- Update aircraft database
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Resolved conflicts:
- routes/acars.py: keep /messages and /clear endpoints for history reload
- routes/vdl2.py: keep /messages and /clear endpoints for history reload
- templates/adsb_dashboard.html: keep removal of hardcoded device-1
defaults for ACARS/VDL2 selectors (users pick their own device)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Mode modules were leaking EventSource connections, setInterval timers,
and setTimeout timers on every mode switch, causing progressive browser
sluggishness. Added destroy() to 8 modules missing it (meshtastic,
bluetooth, wifi, bt_locate, sstv, sstv-general, websdr, spy-stations)
and centralized all destroy calls in switchMode() via a moduleDestroyMap
that cleanly tears down only the previous mode.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Re-implements envelope detection on top of the rewritten Morse decoder.
Addresses PR #160 review feedback:
- Rebase: rebuilt on current upstream/main (lifecycle state machine)
- Gap thresholds: 2.0/5.0 for envelope only; goertzel keeps 2.6/6.0
- Frequency validation: max_mhz=1766 for envelope, 30 for goertzel
- Tests: EnvelopeDetector unit tests + envelope-mode decoder test
- Envelope uses direct magnitude threshold (no SNR/noise ref)
- Goertzel path completely unchanged
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
rtl_test opens the USB device during probing. After killing the
process, the kernel may not release the USB interface immediately.
dump1090 then fails with usb_claim_interface error -6. Add a 0.5s
delay after probe cleanup to allow the kernel to fully release the
device before the actual decoder opens it.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
rtl_test -t often exits non-zero after finding a device (e.g.
"No E4000 tuner found, aborting" with R820T tuners). The return
code fallback was firing even when the "Found N device(s)" success
message had already been matched. Track device_found separately
and only use return code as fallback when no success was seen.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The success check ('Found' in line and 'device' in line) matched
"No supported devices found" since both keywords appear. Add a
pre-check for negative device messages, a return code fallback,
and a clearer error message.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The stop button appeared unresponsive because UI updates waited for the
server response. If the fetch hung or errored, the user saw nothing.
Now the UI updates immediately (matching the pager stop pattern) and
the server request happens in the background.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Auto_rx reads many config keys without defaults and crashes if they're
missing, even for disabled features like email. Include every section
and key from the example config to prevent missing-key errors.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The config format changed significantly: SDR settings moved to [sdr_1],
[positioning] became [location], and many sections are now required.
Also enable payload_summary UDP output so telemetry reaches our listener.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Pass config file path (not directory) to auto_rx -c flag
- Use absolute paths in generated station.cfg since auto_rx runs
with cwd set to its install directory
- Teach dependency checker about auto_rx.py at /opt install path
so the "missing dependency" banner no longer appears
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The setup.sh skip check only looked for auto_rx.py, so a previous
incomplete install (Python files but no compiled binaries) would be
treated as fully installed. Now also checks for dft_detect binary.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
setup.sh and Dockerfile were installing the Python package and copying
files but skipping the build.sh step that compiles the C decoders.
This caused "Binary dft_detect does not exist" at runtime.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The subprocess was launched with bare 'python' which on Debian doesn't
exist (python3 only) and wouldn't have access to the venv-installed
radiosonde dependencies anyway. Using sys.executable ensures the same
interpreter (with all installed packages) is used.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Move adsb_active_device/sdr_type assignment to immediately after
claim_sdr_device so stop_adsb() can always release the device, even
during startup. Sync sdr_type_str after SDRType fallback to prevent
claim/release key mismatch. Clear active device on all error paths.
Replace blind 3s sleep for dump1090 readiness with port-polling loop
(100ms intervals, 3s max). Replace subprocess.run() in rtl_test probe
with Popen + select-based early termination on success/error detection.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Avoids PEP 668 externally-managed-environment error on Debian Bookworm
by using the project's venv/bin/pip instead of system pip3.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Integrate radiosonde_auto_rx for automatic weather balloon detection and
decoding on 400-406 MHz. Includes UDP telemetry parsing, Leaflet map with
altitude-colored markers and trajectory tracks, SDR device registry
integration, setup script installation, and Docker support.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The registry used plain int keys (device index), so HackRF at index 0
and RTL-SDR at index 0 would collide. Changed to composite string keys
("sdr_type:index") so each SDR type+index pair is tracked independently.
Updated all route callers, frontend device selectors, and session restore.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Switch direwolf subprocess output from PIPE to PTY (pseudo-terminal),
forcing line-buffered output so packets arrive immediately instead of
waiting for a 4-8KB pipe buffer to fill. Matches the proven pattern
used by pager mode.
Also enhances direwolf config with FIX_BITS error correction and
disables unused AGWPE/KISS server ports.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The globe wasn't rendering because initGlobe() used setTimeout(100)
which can race with the display:none removal by switchMode(). Both
GPS and WebSDR modes use requestAnimationFrame to wait for the browser
to compute layout before initializing Globe.gl.
- Replace setTimeout with RAF-based retry loop (up to 8 frames)
- Add try-catch around Globe() init with fallback message
- Match the proven pattern from GPS and WebSDR modes
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Fix globe destroyed on re-render by preserving canvas DOM node across
renderLocationCard() calls instead of recreating from scratch
- Reduce globe.gl camera minDistance (180->120) so globe is visible in
200px container
- Clear stale globeInstance ref when canvas is gone
- Enlarge CPU gauge (90->110px), percentage label (18->22px), core bars
(24->48px height), and detail text (11->12px)
- JS fetchLocation() now supplements server response with client-side
ObserverLocation.getShared() from localStorage when server returns
'default' or 'none', picking up manual coordinates from settings modal
- Location priority: GPS > config env vars > manual (localStorage) > default
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add third location fallback to utils/constants (London 51.5074/-0.1278)
so location always resolves even without GPS or env vars configured
- Remove min-height from sys-card to eliminate wasted space
- Switch System Info to vertical key-value layout filling the card
- Clean up OS string (strip glibc suffix), use locale date for boot time
- Bump info grid font size from 11px to 12px for readability
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace broken app.gps_state lookup with utils.gps.get_current_position()
and return GPS metadata (fix quality, satellites, accuracy). Shrink location
card to single-column with 200px globe, move System Info into row 2.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add SVG arc gauge, per-core CPU bars, temperature sparkline, network
interface monitoring with bandwidth deltas, disk I/O rates, 3D globe
with observer location, weather overlay, battery/fan/throttle support,
and process grid layout. New /system/location and /system/weather endpoints.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Clear sidebar highlights and ACARS message timer when stale selected
aircraft is removed in cleanupOldAircraft()
- Escape all user-controlled strings in renderAcarsCard(),
addAcarsMessage(), and renderAcarsMainCard() before innerHTML insertion
- Remove dead duplicate H1 check in classify_message_type
- Move _d label from link_test set to handshake return path
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Real-time dashboard for host metrics (CPU, memory, disk, temperatures),
active decoder process status, and SDR device enumeration via SSE streaming.
Auto-connects when entering the mode with graceful psutil fallback.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Switch decoder subprocess from text mode to binary mode and decode
each line with errors='replace' so corrupted radio bytes (e.g. 0xf7)
are substituted instead of raising UnicodeDecodeError after long runs.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The satellite dropdown had no change listener, so selecting a different
satellite never updated the pass list, timeline, countdown, or polar plot.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The .btn, .btn-sm, and .btn-ghost classes used by morse mode buttons
(TXT, CSV, Copy, Clear) were defined in core/components.css but the
stylesheet was never loaded in index.html, causing unstyled buttons.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
When the stop POST timed out (5s), lifecycle was set to 'idle' on error,
allowing checkStatus to see running=true and reconnect SSE. Now:
- stop .then() stays in 'stopping' on timeout/error instead of going idle
- checkStatus skips reconnect when lifecycle is 'stopping' post-timeout
but still transitions to idle when server confirms running=false
- LOCAL_STOP_TIMEOUT_MS raised from 5s to 12s to match server cleanup time
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The previous stopPromise guard only prevented new polls from being
dispatched. Polls already in-flight before stop was clicked could still
return with running=true and override the stopping lifecycle, causing
SSE reconnection and an apparent restart loop. Add a second guard in
the .then() handler to check stopPromise/lifecycle before acting.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Guard checkStatus() against in-flight stop to prevent status poller
from overriding stopping state and reconnecting SSE. Lower SNR floor
from 1.3 to 1.15 to accommodate weaker CW signals. Add SNR/noise_ref
to scope events and metrics for real-time threshold debugging.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Widen noise detector offset from ±100Hz to ±200Hz to reduce spectral
leakage into the noise reference, and scale threshold_multiplier for
SNR space (2.8 → 1.54) so real CW signals reliably trigger tone
detection instead of producing all-E's at 60 WPM.
Fix misleading "decoder startup" timeout message on stop requests and
increase stop timeout from 2.2s to 5s.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The previous magnitude-based threshold couldn't distinguish CW tone from
AGC-amplified inter-element silence — the Goertzel level stayed above
threshold permanently, preventing any tone OFF transitions and thus zero
character decodes.
Switch tone detection to use SNR (tone_mag / adjacent_band_noise_ref).
Both bands are equally amplified by AGC, so the ratio is gain-invariant.
Also replace the conditional noise_ref guard with unconditional blending
so the noise floor tracks actual ambient levels continuously.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Filter decoder-thread 'stopped' status events that race with the route
lifecycle, causing the frontend to drop back to idle on first start.
Pull noise floor upward using adjacent-frequency Goertzel reference when
warmup calibration runs before AGC converges, preventing permanent
tone-on with zero character decodes.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The multimon-ng MORSE_CW decoder never reliably decoded characters.
Switch live decode to use the existing morse_decoder_thread() which
wraps MorseDecoder with Goertzel tone detection, adaptive thresholds,
and proper timing estimation — eliminating multimon-ng, PTY plumbing,
and the relay thread from the CW pipeline entirely.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
