- Server tracks callsign replacements in prepare_packet_state, sends
convert_to_historical list so JS skips full markerStates scan
- Historical packets get red dot HTML server-side in build_minimal_packet_data
instead of JS createMarkerIcon generating it
- Add callsign_group field to all packet data output; server pre-sorts
by group with chronological order so JS iterates directly
- Normalize heat map intensity in Clustering.cluster_packets (min/50, cap 1.0)
instead of JS doing it per-point
- Remove buildPopupContent JS fallback and unused escapeHtml import;
server already provides popup in all data paths
- Batch real-time packet push_events: collect marker data across the
packet batch and send a single "new_packets" WebSocket message instead
of one per packet. JS handler processes all live→historical conversions
before adding new markers to avoid intra-batch conflicts.
- Batch RF path station queries: replace N+1 get_latest_packet_for_callsign
calls with a single DISTINCT ON query selecting only callsign + lat/lng.
- Add read_concurrency: true to streaming_packets_subscribers and :aprsme
ETS tables (read-heavy, rarely written).
- Remove dead GC code in PacketConsumer (threshold > 1000 unreachable with
max batch_size of 100).
Delete 5 unused modules: Presence (untracked), LiveView ShutdownHandler
(unused mixin), RateLimiterWrapper (pointless delegation), Passcode
(only used by own test), MapLive.Utils (pure delegates).
Fix ShutdownHandler to not call deleted Presence. Inline ParamUtils
at call sites. Remove cachex dep and 27 stale entries from mix.lock.
Clean commented-out Exq/Redis config from application.ex.
System A (PacketManager + PacketStore ETS) had a fatal ID mismatch:
extract_packet_ids returned Ecto UUIDs but PacketStore stored under
MD5 hashes — cleanup could never find stored packets. Unified all
packet tracking on visible_packets map in LiveView assigns.
Deletes packet_manager.ex and packet_store.ex, removes from
supervision tree and PacketDistributor.
Historical map loading previously fetched all 73 columns (SELECT *) from
the packets table (~2068 bytes/row). Added get_recent_packets_for_map/1
with select_map_fields/1 that fetches only the 22 columns needed for map
rendering (~200 bytes/row), yielding a ~3x query speedup.
Also removes unused all_packets tracking from PacketProcessor and
simplifies visible_packets cleanup in Index.
Reorder icon and callsign in the page header, Other SSIDs table, and
Stations Near Current Position table so the icon appears first. Fix
icon clipping with flex-shrink-0 h-8 and add py-2 padding to SSIDs
table cells for proper vertical alignment.
The object name regex only matched [A-Z0-9 ] which missed names with
hyphens (P-K5SGD), hash symbols (#146.760), and other printable ASCII.
Changed to .{9}[*_] to match all valid object names including killed
objects. Also merged the two cond branches so data_extended is tried
first with information_field parsing as fallback.
Removed dead copy_insert/3 tests from db_optimizer_test.
Move JSON response display outside the max-w-md form container
so API results use the full available width instead of being
constrained to the narrow form width.
Historical loading overrides all markers to historical:true, including
the most-recent marker. When a new live packet arrives, the conversion
check only looked for non-historical markers, missing the existing
"current" marker. Broaden the check to also convert markers with
is_most_recent_for_callsign set.
Use mousemove instead of mouseover so the RF path updates as the
cursor moves along the trail. Send the mouse position rather than
the nearest-position's coordinates so the path line anchors where
the user is actually hovering. Debounce and deduplicate events to
avoid flooding the server.
ConnectionMonitor.gather_cluster_stats was calling :rpc.call(Node.self())
which GenServer.call'd back into itself while blocked in handle_info,
causing a 5-second deadlock timeout every 30 seconds. Fixed by computing
local stats directly from state.
LeaderElection never verified that a leader still held the :global
registration. After :global conflict resolution (two isolated nodes
merging), the loser was silently unregistered but stayed is_leader: true
forever, causing both pods to connect to APRS-IS with the same callsign
in a reconnect loop. Fixed by adding verify_leadership/1 to
check_leadership which detects lost registrations and steps down.
Existing historical loading spinner already covers the slow DB
query path; the only unindicated gap is the 400ms debounce which
is too short for a spinner without causing flicker.
:global.register_name returns :no even when the name is already
registered to the calling PID. This caused the leader to clear its
own leadership state every 5 seconds after cluster formation triggered
a redundant re-election attempt.
Now checks if we already hold the registration before attempting to
re-register, preserving leadership state across cluster topology changes.
The slideover toggle button now uses a fixed light theme (white bg,
dark arrow) so it's always visible on the map regardless of OS dark
mode. Removed prefers-color-scheme overrides since the button overlays
the always-light map tiles.
Fixed map not resizing when closing the sidebar — the #aprs-map div
has phx-update="ignore" so server-rendered classes never updated.
Now uses LiveView JS commands (JS.toggle_class + JS.dispatch) to
toggle classes client-side and trigger map.invalidateSize(), replacing
the custom push_event/handleEvent JS handler.
DNSSRV strategy resolves to IP-based names but nodes were named by
pod name, causing connection failures. Switch to DNS strategy with
IP-based node naming so libcluster discovery matches actual node names.
Exact duplicates, indexes subsumed by better composites,
dead lower() indexes (codebase uses upper()), and indexes
with no matching query pattern in the codebase.
parse_rf_path was filtering out all paths containing qA* constructs,
which is virtually every APRS-IS packet. The qA* element delimits the
RF path from the APRS-IS metadata — it should be split on, not
discarded. Now properly extracts digipeaters (before qA*) and the
igate (after qA*), while filtering APRS aliases like WIDE/RELAY/TRACE.
Also adds hover handlers to trail polylines so hovering anywhere on a
call's track shows the RF path, not just on the dot markers.
When the PacketProducer buffer crosses 80% capacity, signal Aprsme.Is
to switch its TCP socket to passive mode, leveraging TCP flow control
to slow APRS-IS intake. Resume active mode when buffer drains below 30%.
A 30-second safety valve timer auto-resumes if backpressure gets stuck.
Disconnect/reconnect/terminate paths all clear backpressure state.
