- Add preflight check for 'docker compose' in manage.sh (catches plugin missing)
- Document named Caddy volumes as cert storage, not user data
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Remove all legacy docker run code paths. manage.sh is now a pure
docker compose wrapper with no dual-mode branching.
Removed:
- COMPOSE_MODE flag and all if/else branches
- get_docker_run_args(), get_data_mount_args(), recreate_container()
- get_required_ports(), get_current_ports(), check_port_match()
- CONTAINER_NAME, DATA_VOLUME, CADDY_VOLUME variables
- All direct docker run/stop/start/rm invocations
All commands now delegate to docker compose:
- start → docker compose up -d prod
- stop → docker compose down / docker compose stop
- restart → docker compose up -d --force-recreate
- update → docker compose build prod + up -d --force-recreate
- reset → docker compose down --rmi local
- backup/restore use bind mount path from .env (PROD_DATA_DIR)
- verify_health, mqtt-test, status all use corescope-prod
Net result: -248 lines, zero dual-mode logic, identical behavior
to running docker compose directly.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
- Replace all hardcoded \C:\Users\KpaBap/meshcore-data with \ variable
- \ resolves from \ in .env or defaults to ~/meshcore-data
- Updated get_data_mount_args(), cmd_backup(), cmd_restore(), cmd_reset()
- Enhanced .env.example with detailed comments for each variable
- Both docker compose and manage.sh now read same .env file
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
PROBLEM:
manage.sh was using named Docker volumes (meshcore-data) as the default,
which hides the database and theme files inside Docker's internal storage.
Users couldn't find their DB on the filesystem for backups or inspection.
The function get_data_mount_args() had conditional logic that only used
bind mounts IF it detected an existing ~/meshcore-data with a DB file.
For new installs, it fell through to the named volume — silently hiding
all data in /var/lib/docker/volumes/.
FIXES:
1. get_data_mount_args() — Always use bind mount to ~/meshcore-data
- Creates the directory if it doesn't exist
- Removes all conditional logic and the named volume fallback
2. cmd_backup() — Use direct path C:\Users\KpaBap/meshcore-data/meshcore.db
- No longer tries to inspect the named volume
- Consistent with the bind mount approach
3. cmd_restore() — Use direct path for restore operations
- Ensures directory exists before restoring files
- No fallback to docker cp
4. cmd_reset() — Updated message to reflect bind mount location
- Changed from 'docker volume rm' to '~/meshcore-data (not removed)'
5. docker-compose.yml — Added documentation comment
- Clarifies that bind mounts are intentional, not named volumes
- Ensures future changes maintain this pattern
VALIDATION:
- docker-compose.yml already used bind mounts correctly (\)
- Legacy 'docker run' mode now matches compose behavior
- All backup/restore operations reference the same bind mount path
DATABASE LOCATION:
- Always: ~/meshcore-data/meshcore.db
- Never: Hidden in Docker's volume storage
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Requested-by: Kpa-clawbot
- Load() SQL: keep o.timestamp DESC (consistent with IngestNewFromDB) so
pickBestObservation tie-breaking is identical on both load paths
- GetTimestamps: scan from tail instead of head (was breaking on first item
assuming it was the newest, now correctly reads from newest end)
- QueryMultiNodePackets: apply same DESC/ASC tail-read pagination as
QueryPackets (was sorting for ASC and assuming DESC as-is)
- GetNodeHealth recentPackets: read from tail to return 20 newest items
(was reading from head = 20 oldest items)
- Remove stale "Prepend (newest first)" comments, replace with accurate
"oldest-first; new items go to tail" wording
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
s.packets and s.byPayloadType[t] were prepended on every new packet
to maintain newest-first order, copying the entire slice each time.
With 2-3M packets in memory this meant ~24MB of pointer copies per
ingest cycle, causing sustained high CPU and GC pressure.
Fix: store both slices oldest-first (append to tail). Load() SQL
changed to ASC ordering. QueryPackets DESC pagination now reads from
the tail in O(page_size) with no sort; GetChannelMessages switches
from reverse-iteration to forward-iteration.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* refactor: remove all packets_v SQL fallbacks — store handles all queries
Remove DB fallback paths from all route handlers. The in-memory
PacketStore now handles all packet/node/analytics queries. Handlers
return empty results or 404 when no store is available instead of
falling back to direct DB queries.
- Remove else-DB branches from handlePacketDetail, handleNodeHealth,
handleNodeAnalytics, handleBulkHealth, handlePacketTimestamps, etc.
- Remove unused DB methods (GetPacketByHash, GetTransmissionByID,
GetPacketByID, GetObservationsForHash, GetTimestamps, GetNodeHealth,
GetNodeAnalytics, GetBulkHealth, etc.)
- Remove packets_v VIEW creation from schema
- Update tests for new behavior (no-store returns 404/empty, not 500)
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* fix: address PR #220 review comments
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
---------
Co-authored-by: Kpa-clawbot <259247574+Kpa-clawbot@users.noreply.github.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: KpaBap <kpabap@gmail.com>
- Add pull_request trigger for PRs against master
- Add 'if: github.event_name == push' to build/deploy/publish jobs
- Test jobs (go-test, node-test) now run on both push and PRs
- Build/deploy/publish only run on push to master
This fixes the chicken-and-egg problem where branch protection requires
CI checks but CI doesn't run on PRs. Now PRs get test validation before
merge while keeping production deployments only on master pushes.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
The glob trick COPY .git-commi[t] only works with BuildKit.
manage.sh uses legacy docker build. Just create a default via RUN.
Commit hash comes through --build-arg ldflags anyway.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Server defaults to 6060, ingestor to 6061. Removed shared PPROF_PORT
env var. Bind failure logs warning instead of log.Fatal killing the process.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Sensor nodes embed telemetry (battery_mv, temperature_c) in their advert
appdata after the null-terminated name. This commit adds decoding and
storage for both the Go ingestor and Node.js backend.
Changes:
- decoder.go/decoder.js: Parse telemetry bytes from advert appdata
(battery_mv as uint16 LE millivolts, temperature_c as int16 LE /100)
- db.go/db.js: Add battery_mv INTEGER and temperature_c REAL columns
to nodes and inactive_nodes tables, with migration for existing DBs
- main.go/server.js: Update node telemetry on advert processing
- server db.go: Include battery_mv/temperature_c in node API responses
- Tests: Decoder telemetry tests (positive, negative temp, no telemetry),
DB migration test, node telemetry update test, server API shape tests
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Fresh Go installs failed with 'no such table: packets_v' because the
ingestor created tables but never the VIEW that the Go server queries.
Add DROP VIEW IF EXISTS + CREATE VIEW packets_v to applySchema(), using
the v3 definition (observer_idx → observers.rowid JOIN). The view is
rebuilt on every startup to stay current with any definition changes.
Add tests: verify view exists after OpenStore, and verify it returns
correct observer_id/observer_name via the LEFT JOIN.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Add net/http/pprof support to both Go server (default port 6060) and
ingestor (default port 6061). Profiling is off by default — only
starts the pprof HTTP listener when ENABLE_PPROF=true.
PPROF_PORT env var overrides the default port for each binary.
Enable on staging-go in docker-compose with exposed ports 6060/6061.
Not enabled on prod.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
SQLite stores these as REAL on some instances. Go *int scan silently
fails, dropping the entire observer row (404 on detail, missing from list).
Reported for YC-Base-Repeater and YC-Work-Repeater.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Observability:
- Add DBStats struct with atomic counters for tx_inserted, tx_dupes,
obs_inserted, node_upserts, observer_upserts, write_errors
- Log SQLite config on startup (busy_timeout, max_open_conns, journal)
- Periodic stats logging every 5 minutes + final stats on shutdown
- Instrument all write paths with counter increments
Tests:
- TestConcurrentWrites: 20 goroutines × 50 writes (1000 total) with
interleaved InsertTransmission + UpsertNode + UpsertObserver calls.
Verifies zero errors and data integrity under concurrent load.
- TestDBStats: verifies counter accuracy for inserts, duplicates,
upserts, and that LogStats does not panic
Three changes to eliminate concurrent write collisions:
1. Add _busy_timeout=5000 to ingestor SQLite DSN (matches server)
- SQLite will wait up to 5s for the write lock instead of
immediately returning SQLITE_BUSY
2. Set SetMaxOpenConns(1) on ingestor DB connection pool
- Serializes all DB access at the Go sql.DB level
- Prevents multiple goroutines from opening overlapping writes
3. Change SetOrderMatters(false) to SetOrderMatters(true)
- MQTT handlers now run sequentially per client
- Eliminates concurrent handler execution that caused
overlapping multi-statement write flows
Root cause: concurrent MQTT handlers (SetOrderMatters=false) each
performed multiple separate writes (transmission lookup/insert,
observation insert, node upsert, observer upsert) without transactions
or connection limits. SQLite only permits one writer at a time, so
under bursty MQTT traffic the ingestor was competing with itself.
#210: Add role="img" aria-label to 9 Chart.js canvases in node-analytics.js
and observer-detail.js with descriptive labels.
#211: Add scope="col" to all <th> elements across analytics.js, audio-lab.js,
compare.js, node-analytics.js, nodes.js, observer-detail.js, observers.js,
and packets.js (40+ headers).
#212: Add aria-label to packet filter input and time window select in
packets.js. Add for/id associations to all customize.js inputs: branding,
theme colors, node/type colors, heatmap sliders, onboarding fields, and
export controls.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
#203: Live page node detail panel becomes a bottom-sheet on mobile
(width:100%, bottom:0, max-height:60vh, rounded top corners).
#204: Perf page reduces padding to 12px, perf-cards stack in 2-col
grid, tables get smaller font/padding on mobile.
#205: Nodes table hides Public Key column on mobile via .col-pubkey
class (same pattern as packets page .col-region/.col-rpt).
Cache busters bumped in index.html.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Build args ensure version badge shows correctly. Health timeout
bumped from 20s to 90s for Go store loading time.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Changed all 'Node.js' references to generic 'Server' in:
- verify_health() - health check messages
- show_container_status() - stats display comment
- cmd_status() - service health output
The Go backend runs behind Caddy just like the Node version did,
so the health checks via docker exec localhost:3000 remain correct.
Only the messaging needed updating.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
- app.js: render engine badge with .engine-badge span (was plain text)
- test: fix #pktRight waitForSelector to use state:'attached' (hidden by detail-collapsed)
- test: fix map heat persist race — wait for async init to restore checkbox state
- test: fix live heat persist race — test via localStorage set+reload instead of click
- test: fix live matrix toggle race — wait for Leaflet tiles before clicking
- test: increase packet detail timeouts for remote server resilience
- test: make close-button test self-contained (navigate if #pktRight missing)
- bump cache busters
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Problem 1 (Go staging timeout): Increased healthcheck from 60s to 120s to allow 50K+ packets to load into memory.
Problem 2 (Node staging timeout): Added forced cleanup of stale containers, volumes, and ports before starting staging containers to prevent conflicts.
Problem 3 (Proto validation WS timeout): Made WebSocket message capture non-blocking using timeout command. If no live packets are available, it now skips with a warning instead of failing the entire proto validation pipeline.
Problem 4 (Playwright E2E failures): Added forced cleanup of stale server on port 13581 before starting test server, plus better diagnostics on failure.
All health checks now include better logging (tail 50 instead of 30 lines) for debugging.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
