CHANGELOG.md

ACARS Hub Change Log

ACARS Hub v4.1.5

v4.1.5 Bug Fixes

• Alerts: Alert term updates could desynchronise the in-memory alert cache from the database in several ways: adding or removing terms did not purge stale alert_matches rows for terms that no longer exist; the message ring buffer was not reheated after term changes, so recently received messages were not re-evaluated against the new term list; and the prune callback could race with a concurrent regeneration, leaving the ring buffer in an inconsistent state. All six desync paths have been fixed and the backend now emits an alerts_refreshed event so connected clients know when to reload alert state.
• Live Messages: Unsaved messages (those that arrive between database writes) were all assigned a placeholder uid of undefined, which React used as a duplicate key, causing rendering bugs when multiple unsaved messages appeared in the same list. Each unsaved message now receives a unique generated UID based on timestamp and a random suffix.
• Live Messages: The virtualised message list height was being reduced by a fixed 200 px offset that caused content to be cut off at the bottom of the viewport. The list now uses the full available window height.
• Database: Migrations 13–14 remove the uid TEXT column from the messages table and switch alert_matches.message_uid to alert_matches.message_id (INTEGER), aligning all foreign-key references with the autoincrement id primary key. This eliminates a redundant UUID column and the index that supported it.
• Database: The backup DB writer no longer attempts to persist unsaved messages (those with negative synthetic IDs), which previously caused constraint violations.
• Backend: ICAO hex field enrichment no longer attempts to guess whether an ICAO string is hexadecimal or decimal by inspecting its character content. Strings of six or fewer characters are treated as hex (uppercased, zero-padded); strings longer than six characters are treated as decimal and converted. This removes a class of misclassification bugs where all-numeric hex addresses (e.g. "400123") were incorrectly converted as decimal (3).
• Backend: Database error handlers no longer call process.exit() directly. The main application code is now responsible for process lifecycle, preventing abrupt termination that could skip cleanup.
• Statistics: The Message Count bar chart on the Stats page was displaying index numbers (0, 1, 2) on the x-axis instead of category labels ("Good Messages", "Errors", "Total"). The x and y scale configurations were swapped — the numeric formatter and axis title were applied to the category axis rather than the value axis. Tooltip values also read from the wrong axis. Fixed by placing the grid, tick formatter, and "Count" title on the y (value) axis and the category labels on the x axis.
• CI: GitHub Actions update workflows (FIR, TRACON, sprites) are now hardened against API rate limits. Unauthenticated GitHub API calls (60 req/hr) could return responses without tag_name, causing jq to output "null" and trigger bogus PRs with corrupted data. Workflows now use GITHUB_TOKEN for authenticated requests (5,000 req/hr), validate extracted versions, and use curl -f to fail loudly on HTTP errors.

v4.1.5 Performance

• Backend: The timeseries statistics cache has been restructured into a tiered architecture. Instead of holding all eight time-period datasets in memory simultaneously, only the active period requested by connected clients is kept fully materialised. Cache warm-up and refresh cycles are staggered so that startup memory consumption is spread over time rather than spiking on first connection.
• Backend: A message ring buffer replaces the previous unbounded in-memory message list for on-connect state delivery. The buffer holds a fixed window of recent messages and recent alerts, capping memory usage regardless of message volume. New clients receive the ring buffer contents on connect rather than triggering a database query.
• Backend: High-frequency per-message debug log statements (message received, formatted, saved) have been moved to trace level, reducing log volume and I/O overhead when running at the default info or debug log levels.
• Database: regenerateAllAlertMatches now uses batch reads and wraps all inserts in a single transaction, reducing I/O and lock contention on large databases.
• Database: Unnecessary indexes on the timeseries_stats table are no longer created during migrations, reducing migration time and database size.
• Database: Migration 15 drops six redundant B-tree indexes on the messages table (dsta, depa, tail, flight, label, freq). These columns are already covered by the FTS5 full-text search index, so the separate B-tree indexes consumed disk space and slowed writes without being used by any query path.
• Database: ICAO search queries now use exact equality (=) instead of LIKE '%…%' when the search term is a full six-character hex address, allowing SQLite to use an index seek instead of a full table scan. Wildcard searches (* or %) and partial-length terms still use LIKE (3).
• Docker: The Dockerfile now copies only package.json manifests before npm ci, so the dependency layer survives source-only edits and Docker can cache it independently (1).

v4.1.5 New

• Settings: Alert term management (add, remove, ignore, regenerate) has been split out of the Notifications tab into its own dedicated Alerts tab in the Settings modal. The Notifications tab now contains only notification delivery settings (desktop notifications, sound, volume, on-page alerts).
• Messages: Bump @airframes/acars-decoder to 1.8.13 (2).
• Data: TRACON boundaries updated to v1.2.6; FIR boundaries updated to v2602.2.
• Docker: Log timestamps now use s6-wrap instead of piping to awk, reducing process overhead and simplifying the logging pipeline (1).
• Backend: Database query elapsed time is now logged at debug level for search and message retrieval queries, aiding performance diagnosis.

v4.1.5 Improvements

• Backend: RRD migration memory footprint reduced — the importer now streams data in smaller batches rather than loading entire RRD files into memory at once (1).
• Frontend: The Stats page layout has been tightened for smaller screens — reduced padding, spacing, and margins throughout; chart wrappers are capped at 600 px height to maintain readable proportions on wide monitors; the tab switcher uses more compact padding at narrow widths.
• Frontend: On viewports shorter than 800 px, the page title header is hidden to reclaim vertical space for content. On viewports shorter than 820 px, the navigation bar logo image is also hidden. This benefits laptop screens and browser windows that are not full-height.
• Frontend: Chart minimum heights reduced from 250 px to 200 px across all chart components, allowing charts to fit better on constrained viewports without forcing a scrollbar.
• Dependencies: npm packages updated across backend, frontend, and shared types.

v4.1.5 Notes

1. Credit to @wiedehopf for the Dockerfile caching improvement in PR #1652, the s6-wrap logging change, and the RRD migration memory reduction.
2. Credit to @makrsmark for updating @airframes/acars-decoder to 1.8.13 in PR #1650.
3. Credit to @wiedehopf for the ICAO hex handling fix and ICAO search optimization.

ACARS Hub v4.1.4

v4.1.4 Bug Fixes

• Stats page: Signal Level, Alert Terms, and Frequency Distribution charts now correctly flex-fill the available viewport height without overflowing the page. All three chart components were bypassing the fill-height flex chain — Signal Level had no wrapper class, and Alert Terms / Frequency Distribution used fixed inline pixel heights with contain: layout preventing the flex shrink mechanism from clamping them. All three now use className="chart__canvas-wrapper" and the inline styles have been removed.
• Live Messages / Alerts: When a duplicate message is promoted to the front of its card group, the incoming message's matched/alert metadata is now correctly merged forward. Previously the promoted card retained the existing message's matched=false state even if the newer copy carried matched=true, causing alert styling, has_alerts, and alert counts to be incorrect. Matched state can only transition false → true, never true → false. Matched term arrays (matched_text, matched_icao, matched_tail, matched_flight) are deduplicated and merged across both copies. If the promoted message had previously been marked as read, its read state is also reset so the newly-relevant card re-appears as unread.
• Alerts: The per-card Mark Read button on mobile now renders in a correct two-row CSS Grid layout: the button appears top-right, and the message timestamp wraps to a full-width second row. Non-alert cards (which have no button) are unaffected by the grid rule.
• Alerts: The page-level Mark All Read button is now placed in the title row on mobile, appearing as a direct flex sibling of the page title. On mobile, space-between aligns the title left and the button right on the first row, with the message statistics wrapping to a full-width row below. On desktop the original end-of-stats-row grouping is preserved via CSS order.
• Database: IMSL frequency distribution on the Status page was permanently empty. normalizeMessageType("IMSL") returns "IMS-L" (matching Python's DB storage format), but updateFrequencies() in helpers.ts only had "IMSL" as a lookup key — not "IMS-L". Every IMSL message silently fell through to the unknown-type warn branch without writing to freqs_imsl. Fixed by adding "IMS-L": freqsImsl to the map, mirroring the existing "VDL-M2" / "VDLM2" dual-entry pattern. Regression test added that fails without the fix.
• Navigation: The "ACARS Hub" text next to the logo image is now hidden between 768 px and 850 px viewport widths, where the number of nav bar elements would otherwise cause it to word-wrap. The image logo alone identifies the app at this range; the text reappears at 851 px and above.
• Status page: The rolling messages-per-minute rate was showing a coarse per-minute counter (reset at each minute boundary) instead of the rolling 60-second rate, and the value only refreshed on the 10-second request_status poll rather than the 5-second message_rate emit. Fixed by reading getRollingRates() for the initial system_status response, emitting message_rate alongside system_status on request_status, and having the Status page use the store's live messageRate value (updated every 5 seconds) as its primary source for the Rate (1 min) display.
• Database: Running several migrations in sequence triggered multiple VACUUM stalls. Consolidated VACUUM and ANALYZE so they run exactly once at the end of runMigrations(), only when at least one migration step executed or the FTS startup repair rebuilt the virtual table. ANALYZE now runs after VACUUM so the query planner sees the final compacted page layout.
• Live Map: Collapsing the aircraft-list sidebar at tablet viewport widths (768 px–1023 px) left the sidebar at ~280 px instead of the expected 40 px collapsed strip. A min-width: 280px rule applied to the sidebar in that breakpoint range was overriding the JS-driven --map-sidebar-width: 40px CSS custom property. Fixed by adding a &--collapsed override inside the tablet breakpoint block that resets min-width to 0 and max-width to none, allowing the CSS variable to take full effect.
• Status page: Value labels on the Alert Terms and Frequency Distribution horizontal bar charts no longer overlap the row (Y-axis) labels when a bar represents a very small count. Labels now use dynamic positioning: bars whose value is less than 15% of the axis maximum place their label outside and to the right of the bar tip (using the theme text colour, no pill background); taller bars continue to place the label inside the bar at its right end (dark text on the coloured pill). This preserves the original intent of keeping labels for large-value bars inside the chart boundary while eliminating the overlap for small-value bars.
• Statistics: The RRD → SQLite import pipeline had two bugs that caused the Reception Over Time charts to show incorrect zero values and a data gap after migration. (1) Coarse archive points (5-min, 1-hour, 6-hour) were expanded forward in time instead of backward: rrdtool returns the timestamp at the end of a consolidation period, so the old code placed historical data at wrong future timestamps, permanently squatting slots the live stats-writer needed. (2) Time slots with no RRD source data (the gap between the RRD's last_update and the import time) were inserted as all-zero rows instead of being skipped, producing false zero counts. Both bugs are fixed: coarse points now expand backward, and all-NaN rows are skipped entirely. Future-dated rows are clamped to Date.now() as an additional safeguard. On the next startup after this update the importer automatically detects and repairs databases affected by the old code: rows in the corrupted time ranges are deleted and a clean re-import is performed.

v4.1.4 Performance

• Statistics: The Stats page time-series graphs (Reception Over Time) no longer query the database on every user request or period change. The backend now warms all eight time-period cache entries in memory at startup and refreshes each on a wall-clock-aligned schedule (every 1 minute for the 1 hr/6 hr/12 hr windows, scaling up to every 12 hours for the 1-year window). Refreshed results are broadcast to all connected clients automatically — no client request is needed. The frontend receives these pushes through the global useSocketIO handler and stores all eight period results in a Zustand in-memory cache; switching between time periods on the Stats page is now instant with no loading delay once the initial warm-up response has arrived (within one round-trip of connecting).

• Live Messages: The message list is now rendered as a virtual windowed list using @tanstack/react-virtual. Only the ~7 message cards visible in the viewport are mounted in the DOM at any time, down from ~90 fully-mounted trees previously. On busy stations this eliminates the UI lag that accumulated as the message list grew. Theme switching, which previously had to cascade CSS variable changes through every mounted card, is now instant.

• Alerts: The alert message list (both live and historical modes) is now also virtualised, using the same architecture as Live Messages.

• Search: Search results are now rendered as a virtual list. The search form itself remains fully scrollable on all screen sizes.

v4.1.4 New

• Scroll to top: A floating action button (FAB) appears on mobile devices after scrolling 150 px and smoothly returns to the top of the current page. The FAB is suppressed on the ADS-B map page, which manages its own scroll context. On desktop, clicking the currently-active navigation link also scrolls back to the top of the page, giving keyboard and mouse users an equivalent shortcut. Pages with virtualised lists (Live Messages, Alerts) register their custom scroll containers so both the FAB and the nav re-click always target the correct element.

• Database: Migration 12 — the timeseries_stats table has been rebuilt to remove three dead-weight columns (id, resolution, created_at) and replace them with timestamp INTEGER PRIMARY KEY. resolution was a non-nullable constant ('1min') on every row; id is superseded by the timestamp rowid alias; created_at was set on insert and never read. The new schema saves ~20 bytes per row with zero index overhead beyond the table B-tree itself.

• Docker: The frontend now shows a "Database migration in progress" banner while the backend applies SQLite migrations at startup. The HTTP and Socket.IO servers start before migrations begin; clients that connect during the migration window receive migration_status { running: true } and are held in a pending queue until all initialisation is complete, then receive the full connect sequence.

• Message rate widget: A rolling messages-per-minute counter is now displayed in the navigation bar. On desktop, hovering or focusing the widget expands a tooltip showing the rate broken down by decoder type (ACARS, HFDL, VDL-M2, etc.). On mobile devices with a screen width of 375 px or wider the total rate is shown directly in the nav bar.

• Status page: A "Rolling Rate" row has been added to the Message Statistics card for each enabled decoder, showing the same rolling 60-second rate as the nav bar widget.

• Search: The search form now auto-collapses when you scroll more than 80 px into the results. A sticky header pins to the top of the viewport while collapsed, showing a summary of the active search criteria and a button to expand the form again. Clicking the expand button scrolls back to the top and reopens the form.

• Search: All text search inputs are now automatically normalised to uppercase as you type, matching the way messages are stored in the database and eliminating missed results caused by case differences.

• Messages: Any portion of a message that the decoder recognised but could not fully decode is now shown as "Remaining Text" in the message detail view, rather than being silently discarded (1)

• Messages: Bump acars-decoder version to 1.8.8

• Label the y axis on Reception over time graphs with the message count per time slot.

v4.1.4 Improvements

• Navigation: The mobile and desktop navigation bars are now conditionally rendered — only the layout appropriate for the current screen size is mounted. Previously both trees were always present in the DOM with CSS toggling visibility, which meant React was maintaining two full navigation trees, their event listeners, and active-link tracking simultaneously.

v4.1.4 Notes

1. Credit to @makrsmark for the remaining text feature in PR #1637.
2. Credit to @makrsmark for updating acars-decoder to 1.8.8 in PR #1639

ACARS Hub v4.1.3

v4.1.3 Bug Fixes

• Database: FTS5 tombstone accumulation on high-volume installs (HFDL + VDL-M2) caused the search index shadow tables to grow to hundreds of thousands of segments and several gigabytes. Every new message insertion then triggered FTS5 automerge to block the database thread for seconds, stalling message ingestion entirely. Root cause: optimizeDbMerge() was calling merge(-16) which performs only ~64 KB of consolidation work per call — completely unable to keep pace with tombstone generation at high message rates. Fixed by: (1) adding database migration 10 which drops and recreates all FTS tables and triggers and rebuilds the index from scratch (one-time startup cost; expect 5–30 minutes on a large database — no data is lost); (2) increasing the merge work per call from merge(-16) (~64 KB) to merge(500) (~2 MB); (3) adding a new optimizeDbFts() function that runs FTS5 optimize — a closed-loop operation that consolidates all b-tree levels until fully done, regardless of how much work is required. merge(500) runs every 5 minutes for cheap bounded housekeeping; optimize runs every 30 minutes as a correctness guarantee. Together they ensure segment count stays bounded on any deployment size.
• Database: WAL checkpoint threshold lowered further from 400 pages (~1.6 MB) to 200 pages (~800 KB) to keep the WAL file small between scheduled TRUNCATE checkpoints on high-volume installs (HFDL can sustain 1,000+ messages/minute). A TRUNCATE-mode checkpoint now also runs at startup in addition to the existing 15-minute scheduled run.
• Database: Added checkpointBackup() — the scheduled WAL checkpoint was only applied to the primary database connection, leaving the backup WAL (DB_BACKUP) to grow without bound. The backup connection is now checkpointed on the same schedule as the primary.
• Database: WAL mode activation is now verified on startup. On file systems that do not support shared-memory files (some NFS mounts, certain bind-mounts), SQLite silently falls back to DELETE journal mode. ACARS Hub now detects this and logs an error so operators know the WAL strategy is inactive before a disk-full condition occurs.
• Database: Timeseries statistics data (timeseries_stats) could be imported from RRD more than once, doubling (or further multiplying) the historical graph data on each restart after the initial migration. Fixed by database migration 11 which: (1) deduplicates any existing duplicate rows, keeping the highest-ID survivor per time slot; (2) upgrades the non-unique index on (timestamp, resolution) to a UNIQUE INDEX so the database itself enforces one data point per time slot going forward; (3) creates a new rrd_import_registry table that stores a SHA-256 content fingerprint of each imported RRD file. On startup, if the fingerprint of the .rrd file is already registered the import is skipped entirely — this catches re-imports regardless of filename. All batch inserts now use INSERT OR IGNORE so even if the registry check is somehow bypassed, the unique constraint silently discards duplicate rows.
• Statistics: Frequency and alert term charts now render bars sorted from highest to lowest count. Previously bars were rendered in data-arrival order, making the charts difficult to read.

v4.1.3 New

• Alerts: VOMIT and HAZMAT added as default alert terms

ACARS Hub v4.1.2

v4.1.2 Bug Fixes

• Database: The WAL (Write-Ahead Log) file could grow unboundedly because SQLite's default PASSIVE auto-checkpoint is silently skipped whenever any read transaction is open. With the system-status emitter creating short read transactions every 30 seconds there was almost always a recent read mark, so un-checkpointed FTS5 writes accumulated in the WAL indefinitely. Fixed by lowering the auto-checkpoint threshold from 1000 pages (~4 MB) to 400 pages (~1.6 MB), and by adding a scheduled TRUNCATE-mode checkpoint every 15 minutes. TRUNCATE mode checkpoints every pending frame and truncates the WAL file to zero bytes, immediately reclaiming disk space.
• Database: pruneDatabase() loaded all alert-protected message UIDs into a JavaScript array and passed them as SQL bind parameters via NOT IN (?, ?, …). With long DB_ALERT_SAVE_DAYS values and active alert terms, this list could exceed SQLite's SQLITE_MAX_VARIABLE_NUMBER limit (default 999), causing an SQLITE_ERROR on every prune run and preventing the database from ever shrinking. Fixed by replacing the two-step fetch-then-bind approach with a single DELETE … WHERE uid NOT IN (SELECT message_uid FROM alert_matches WHERE …) subquery that SQLite resolves entirely in-engine with no parameter-count ceiling.
• Database: checkpoint() parsed PRAGMA wal_checkpoint results incorrectly in two ways. First, better-sqlite3's { simple: true } option returns only the first column of the first row as a scalar (the busy flag), not an array — so framesCheckpointed and framesRemaining were both undefined at runtime, making the scheduled checkpoint warning impossible to trigger. Second, even if indexing had been correct, the column mapping was inverted: framesCheckpointed was reading the log column (total frames written to WAL) and framesRemaining was reading the checkpointed column (frames already moved to the main DB — the opposite of remaining). Fixed by dropping { simple: true }, accessing columns by name, and computing framesRemaining = row.log - row.checkpointed.
• Database: The backup database connection (DB_BACKUP) was missing the wal_autocheckpoint = 400 pragma that is applied to the primary connection. Without it, the backup WAL defaulted to SQLite's 1000-page threshold and could grow unbounded under the same workload conditions that triggered the original SQLITE_FULL errors. The pragma is now applied to both connections during initialisation.
• Healthcheck: Every decoder socket check has always reported UNHEALTHY due to a wrong process name in the ss(8) filter. The Node.js worker thread that owns decoder sockets appears in the process table as node-MainThread, not node, so grep '"node"' never matched and the socket check silently returned nothing regardless of actual connection state. Additionally, even a correctly named filter can fail in container environments where the kernel does not expose the users:((...)) column to the caller. Fixed by checking for the bound/connected port only (no process name filter — the decoder ports are container-specific and no other process will hold them). The socket check is also now advisory: a failed socket check no longer sets an unhealthy exit code on its own. EXITCODE=1 is only raised when both the socket check and the message-activity check fail simultaneously, so a container receiving messages is never incorrectly marked unhealthy.

v4.1.2 New

• Alerts: drunk added as a default alert term

ACARS Hub v4.1.1

v4.1.1 Improvements

• Docker image: reduced size by approximately 850 MB compared to v4.1.0
  • Build stage: compiler toolchain (make, python3, g++, cmake) no longer leaves artifacts in the final image — tools are installed and used in the build stage where they are already present, then the compiled output is copied across (1)
  • Runtime stage: the backend is now bundled with esbuild into a single file (server.bundle.mjs). All pure-JS dependencies (fastify, socket.io, drizzle-orm, pino, zod, etc.) are inlined into the bundle. Only the two native addons (better-sqlite3 and zeromq) remain in node_modules at runtime, reducing the runtime node_modules footprint from ~66 MB to ~11 MB
  • Runtime stage: npm is no longer included in the image — it is not needed at runtime
• nginx: eliminated startup warnings about duplicate text/html MIME type in the compression configuration (text/html is always compressed by nginx and does not need to be listed explicitly)
• Healthcheck: rewritten for the Node.js backend architecture

v4.1.1 Documentation

• README and setup guide: tone, accuracy, and clarity improvements

v4.1.1 Notes

1. Credit to @wiedehopf for the initial compiler and node_modules pruning work in PR #1632 that this builds on.

ACARS Hub v4.1.0

v4.1.0 New

• Backend: completely rewritten in Node.js
• Backend: Connect to acars_router or decoders directly via TCP/ZMQ (2)
• Front End: Reduced initial load time and bandwidth for all deployment types
• Database: Time series data is now stored in the main SQLite database and migrated automatically from RRD on first run
• Message Groups: Instead of showing a generic "ADS-B" label, the actual source type is now shown (ADS-B/UAT/TIS-B/ADSC etc.)
• Live Messages: Filter by station ID
• Live Map: ADS-B source type is displayed on mouse hover of an aircraft
• Live Map: Updated sprites to latest from Plane Watch (BL8 and C206 added)
• Live Map: Aircraft markers outside the viewport are no longer rendered — improves performance on HFDL and other long-range deployments
• Live Map: Side bar filter option to show only aircraft currently visible on the map
• Live Map: Side bar is now resizable and collapsible
• Live Map: Side bar badges indicate which decoder type(s) received each aircraft, replacing the generic green checkmark with a colour-coded badge. At minimum width only the most recent decoder type is shown; expand the sidebar to see all badges for aircraft received on multiple decoder types
• Live Map: Worldwide TRACON boundary overlay (1)
• Live Map: Worldwide FIR boundary overlay (1)
• Live Map: Hey What's That support. Enable with HEYWHATSTHAT=<token>. Optionally specify the altitudes to display with HEYWHATSTHAT_ALTS=<comma-separated list of altitudes in feet, no units>
• Mobile Live Map: Additional map controls collapse into a flyout menu at smaller breakpoints

v4.1.0 Bug Fixes

• Database: Migration from any version of ACARS Hub prior to v4 incorrectly skipped FTS table rebuilds. New databases created in v4 are unaffected. The database will repair itself automatically if the issue is detected — this may take some time on large databases, but no data is lost.
• Network: Removed IPv6 binding in nginx that caused container startup failures on some host configurations
• Live Map: Panning and zooming no longer hides overlays or re-requests overlay data from the server on every interaction
• Live Map: Zoom In/Out buttons now render above aircraft markers

v4.1.0 Notes

1. Worldwide TRACON and FIR boundary data is sourced from VATSIM — a community of flight simulation enthusiasts who volunteer as virtual ATC controllers. The data appears to be derived from real ATC boundaries: US coverage is largely accurate (Amarillo Approach is close but not quite right), though I cannot independently verify accuracy for non-US regions. As an FAA employee, I won't use my work access to pull official data for comparison in order to avoid any conflict of interest.

2. By default, ACARS Hub will behave exactly as before — no change is required and message ingestion is unaffected.

   The new <ACARS/VDLM/HFDL/IRDM/IMSL>_CONNECTIONS variables optionally allow ACARS Hub to connect out to acars_router or the decoders directly, rather than waiting for them to push data in. Multiple sources per decoder type are supported. This is now the recommended setup.

   Example configuration:

   ACARS_CONNECTIONS=udp                          # default — no change needed
   ACARS_CONNECTIONS=udp://0.0.0.0:42069          # UDP on a custom port
   HFDL_CONNECTIONS=zmq://acars_router:15556      # connect to acars_router over ZMQ
   VDLM_CONNECTIONS=udp;zmq://acars_router:45555  # listen on UDP and also connect via ZMQ

   The _CONNECTIONS variables are ignored if the corresponding ENABLE_ variable is not set. If you migrate to the outbound connection model, remove acarshub from your AR_SEND_UDP variables to avoid log spam from acars_router attempting to push to an offline host.

   Documentation has been updated to reflect the new recommended setup.

ACARS Hub v4.0.1

v4.0.1 Bug Fixes

• Live Map Side bar: Hovering over plane in the side bar no longer causes the plane to change heading
• Live Map: Mobile pinch to zoom is fixed

v4.0.1 New

• Re-add functionality to generate <:port>/data/stats.json

ACARS Hub v4.0.0

v4.0.0 New

• Complete rewrite of the web front end
• Desktop notifications
• Unified Settings, localization, custom map provider(S), and so much more
• Improved statistics
• Live map: layers, pausing, follow aircraft, filters, animated sprites...
• Improved searching/alert matching