Extra repeats

src/mesh/FloodingRouter.cpp

@@ -4,6 +4,7 @@
 #include "configuration.h"
 #include "mesh-pb-constants.h"
 #include "meshUtils.h"
+#include "modules/RepeatScalingModule.h"
 #include "modules/TextMessageModule.h"
 #if !MESHTASTIC_EXCLUDE_TRACEROUTE
 #include "modules/TraceRouteModule.h"
@@ -137,10 +138,16 @@ bool FloodingRouter::roleAllowsCancelingDupe(const meshtastic_MeshPacket *p)
 void FloodingRouter::perhapsCancelDupe(const meshtastic_MeshPacket *p)
 {
     if (p->transport_mechanism == meshtastic_MeshPacket_TransportMechanism_TRANSPORT_LORA && roleAllowsCancelingDupe(p)) {
-        // cancel rebroadcast of this message *if* there was already one, unless we're a router!
-        // But only LoRa packets should be able to trigger this.
-        if (Router::cancelSending(p->from, p->id))
-            txRelayCanceled++;
+        // RepeatScalingModule owns the "how many duplicates should we tolerate before giving up"
+        // decision (and its bookkeeping/logging) - see RepeatScalingModule::shouldCancelDupe. When
+        // the module is compiled out (MESHTASTIC_EXCLUDE_REPEATSCALING), fall back to the historical
+        // behavior of cancelling on the first duplicate rather than never cancelling at all.
+        if (!repeatScalingModule || repeatScalingModule->shouldCancelDupe(p)) {
+            // cancel rebroadcast of this message *if* there was already one, unless we're a router!
+            // But only LoRa packets should be able to trigger this.
+            if (Router::cancelSending(p->from, p->id))
+                txRelayCanceled++;
+        }
     }
     if (config.device.role == meshtastic_Config_DeviceConfig_Role_ROUTER_LATE && iface) {
         iface->clampToLateRebroadcastWindow(getFrom(p), p->id);

src/mesh/FloodingRouter.h

@@ -70,7 +70,9 @@ class FloodingRouter : public Router
     // the same packet
     bool roleAllowsCancelingDupe(const meshtastic_MeshPacket *p);
 
-    /* Call when receiving a duplicate packet to check whether we should cancel a packet in the Tx queue */
+    /* Call when receiving a duplicate packet to check whether we should cancel a packet in the Tx queue.
+     * The "how many duplicates should we tolerate before giving up" decision (and its bookkeeping/
+     * logging) is delegated to RepeatScalingModule - see perhapsCancelDupe's implementation. */
     void perhapsCancelDupe(const meshtastic_MeshPacket *p);
 
     // Return true if we are a rebroadcaster

src/mesh/NextHopRouter.cpp

@@ -11,6 +11,7 @@
 #include "modules/TrafficManagementModule.h"
 #endif
 #include "NodeDB.h"
+#include "modules/RepeatScalingModule.h"
 
 #if USERPREFS_EVENT_MODE
 static void capEventRelayHops(meshtastic_MeshPacket *packet)
@@ -234,6 +235,15 @@ bool NextHopRouter::perhapsRebroadcast(const meshtastic_MeshPacket *p)
                         return true;
                     LOG_INFO("Rebroadcast msg from %x", p->relay_node);
 
+                    // Cache this packet's portnum against (p->from, p->id) so that when a duplicate of
+                    // it arrives from the air later - still encrypted to us, since only this decoded
+                    // copy ever gets decoded - RepeatScalingModule::getDupeCancelThreshold has a
+                    // per-portnum signal to act on instead of always seeing an undecodable packet.
+                    if (repeatScalingModule)
+                        repeatScalingModule->noteScheduled(
+                            p->from, p->id,
+                            p->which_payload_variant == meshtastic_MeshPacket_decoded_tag ? p->decoded.portnum : -1);
+
                     // If exhausting hops, force hop_limit = 0 regardless of other logic
                     if (exhaustHops) {
                         tosend->hop_limit = 0;

src/mesh/RadioLibInterface.cpp

@@ -12,6 +12,7 @@
 #if !MESHTASTIC_EXCLUDE_BEACON
 #include "modules/MeshBeaconModule.h"
 #endif
+#include "modules/RepeatScalingModule.h"
 #include <pb_decode.h>
 #include <pb_encode.h>
 
@@ -799,6 +800,12 @@ bool RadioLibInterface::startSend(meshtastic_MeshPacket *txp)
             enableInterrupt(isrTxLevel0);
             lastTxStart = millis();
             printPacket("Started Tx", txp);
+            if (repeatScalingModule) {
+                uint8_t dupesTolerated = repeatScalingModule->getToleratedDupeCount(txp->from, txp->id);
+                if (dupesTolerated > 0)
+                    LOG_DEBUG("[REPEATSCALE] Transmitting 0x%08x from=0x%08x after tolerating %u duplicate(s)", txp->id,
+                              txp->from, dupesTolerated);
+            }
 #ifdef LED_LORA
             digitalWrite(LED_LORA, LED_STATE_ON);
 #endif

src/modules/HopScalingModule.h

@@ -204,6 +204,10 @@ class HopScalingModule : private concurrency::OSThread
     uint16_t getHashSeed() const { return hashSeed; }
     /// Expose hashNodeId for tests that need to compute which node IDs pass a given denominator.
     uint16_t hashNodeIdPublic(uint32_t nodeId) const { return hashNodeId(nodeId); }
+    /// Directly inject the last-rollover per-hop counts, bypassing the histogram/rollHour()
+    /// machinery. For tests (e.g. FloodingRouter's dupe-repeat gate) that only need a specific
+    /// getLastPerHopCounts() result and don't care how it was derived.
+    void setLastPerHopCountsForTest(const PerHopCounts &counts) { lastPerHopCounts = counts; }
 #endif
 
   protected:

src/modules/Modules.cpp

@@ -50,6 +50,9 @@
 #if HAS_VARIABLE_HOPS
 #include "modules/HopScalingModule.h"
 #endif
+#if !MESHTASTIC_EXCLUDE_REPEATSCALING
+#include "modules/RepeatScalingModule.h"
+#endif
 #include "modules/TextMessageModule.h"
 #if !MESHTASTIC_EXCLUDE_TRACEROUTE
 #include "modules/TraceRouteModule.h"
@@ -143,6 +146,10 @@ void setupModules()
     hopScalingModule = new HopScalingModule();
 #endif
 
+#if !MESHTASTIC_EXCLUDE_REPEATSCALING
+    repeatScalingModule = new RepeatScalingModule();
+#endif
+
 #if !MESHTASTIC_EXCLUDE_ADMIN
     adminModule = new AdminModule();
 #endif

src/modules/RepeatScalingModule.cpp

@@ -0,0 +1,185 @@
+#include "RepeatScalingModule.h"
+#include "DebugConfiguration.h"
+#include "airtime.h"
+#include "configuration.h"
+#include "modules/HopScalingModule.h"
+
+RepeatScalingModule *repeatScalingModule;
+
+// Design notes for getDupeCancelThreshold()'s policy (kept here rather than inline):
+//
+// Historically we cancel our own queued rebroadcast the instant we hear one duplicate from another
+// node. This module lets selected packet types instead tolerate a few heard duplicates first,
+// trading a little extra airtime for better delivery odds. Thresholds live in a compile-time
+// per-portnum switch (not runtime config) because the right values aren't yet settled.
+//
+// The switch applies uniformly to broadcasts and DMs - NextHopRouter routes duplicates through the
+// same FloodingRouter::perhapsCancelDupe -> shouldCancelDupe path - so a DM of a listed portnum
+// gets the same tolerance as a broadcast of it, and only ever for a *decodable* DM (a DM not
+// addressed to us can't be decoded by us).
+//
+// When the portnum can't be determined (packet still encrypted, no noteScheduled() cache hit), we
+// fall back to a next_hop gate: NO_NEXT_HOP_PREFERENCE (flood-relayed) gets text-message tolerance
+// since it has the same uncertain single-path shape; a specific next_hop does not, as delivery
+// there is already backed by the sender's end-to-end ACK/retry.
+//
+// Either way, meshTooBusyForExtraRepeats() forces the threshold back to 1 on a busy/dense mesh.
+
+namespace
+{
+// Thresholds above which the mesh is busy/dense enough that extra repeats aren't worth the airtime.
+constexpr float BUSY_CHANNEL_UTIL_PERCENT = 10.0f;
+constexpr float BUSY_AIR_UTIL_TX_PERCENT = 4.0f;
+constexpr uint16_t BUSY_DIRECT_ACTIVE_NODES = 10;
+
+// True if channel/air utilization or direct-neighbor density says the mesh is too busy for extra
+// repeats. Logs which condition tripped.
+bool meshTooBusyForExtraRepeats()
+{
+    if (airTime && airTime->channelUtilizationPercent() > BUSY_CHANNEL_UTIL_PERCENT) {
+        LOG_DEBUG("[REPEATSCALE] Mesh busy: chUtil=%.1f%% > %.1f%%", airTime->channelUtilizationPercent(),
+                  BUSY_CHANNEL_UTIL_PERCENT);
+        return true;
+    }
+    if (airTime && airTime->utilizationTXPercent() > BUSY_AIR_UTIL_TX_PERCENT) {
+        LOG_DEBUG("[REPEATSCALE] Mesh busy: airUtilTX=%.1f%% > %.1f%%", airTime->utilizationTXPercent(),
+                  BUSY_AIR_UTIL_TX_PERCENT);
+        return true;
+    }
+#if HAS_VARIABLE_HOPS
+    // perHop[0] is HopScalingModule's estimate of active direct (hop_away == 0) neighbors.
+    if (hopScalingModule && hopScalingModule->getLastPerHopCounts().perHop[0] > BUSY_DIRECT_ACTIVE_NODES) {
+        LOG_DEBUG("[REPEATSCALE] Mesh busy: directActiveNodes=%u > %u", hopScalingModule->getLastPerHopCounts().perHop[0],
+                  BUSY_DIRECT_ACTIVE_NODES);
+        return true;
+    }
+#endif
+    return false;
+}
+} // namespace
+
+// Per-portnum duplicate-tolerance threshold; see the design notes above for the full rationale.
+uint8_t RepeatScalingModule::getDupeCancelThreshold(const meshtastic_MeshPacket *p)
+{
+    const int32_t portnum = resolvePortnum(p);
+
+    uint8_t threshold;
+    if (portnum >= 0) {
+        switch (portnum) {
+        case meshtastic_PortNum_TEXT_MESSAGE_APP:
+        case meshtastic_PortNum_TEXT_MESSAGE_COMPRESSED_APP:
+            // User-visible chat: no broadcast ACK/retry safety net, so tolerate one heard repeat.
+            threshold = 2;
+            break;
+        default:
+            threshold = 1;
+            break;
+        }
+    } else {
+        // Portnum unknowable (undecodable packet): fall back to the plaintext next_hop header.
+        threshold = (p->next_hop == NO_NEXT_HOP_PREFERENCE) ? 2 : 1;
+        LOG_DEBUG("[REPEATSCALE] portnum unknown for 0x%08x from=0x%08x; next_hop=0x%x -> threshold=%u", p->id, p->from,
+                  p->next_hop, threshold);
+    }
+
+    // A busy/dense mesh overrides any extra tolerance decided above.
+    if (threshold > 1 && meshTooBusyForExtraRepeats()) {
+        LOG_DEBUG("[REPEATSCALE] portnum=%d wanted threshold=%u but mesh is busy; falling back to 1", portnum, threshold);
+        return 1;
+    }
+
+    return threshold;
+}
+
+uint8_t RepeatScalingModule::registerDupeHeard(NodeNum sender, PacketId id)
+{
+    for (auto &entry : dupeCounts) {
+        if (entry.id == id && entry.sender == sender) {
+            if (entry.count < UINT8_MAX)
+                entry.count++;
+            return entry.count;
+        }
+    }
+    // Not tracked yet: claim the next ring slot, evicting whatever was there.
+    DupeCountEntry &slot = dupeCounts[dupeCountsNextSlot];
+    dupeCountsNextSlot = (dupeCountsNextSlot + 1) % DUPE_COUNT_TRACKER_SIZE;
+    slot.sender = sender;
+    slot.id = id;
+    slot.count = 1;
+    slot.portnum = -1; // no noteScheduled() preceded this
+    return slot.count;
+}
+
+void RepeatScalingModule::clearDupeCount(NodeNum sender, PacketId id)
+{
+    for (auto &entry : dupeCounts) {
+        if (entry.id == id && entry.sender == sender) {
+            entry.sender = 0;
+            entry.id = 0;
+            entry.count = 0;
+            entry.portnum = -1;
+            return;
+        }
+    }
+}
+
+void RepeatScalingModule::noteScheduled(NodeNum sender, PacketId id, int32_t portnum)
+{
+    for (auto &entry : dupeCounts) {
+        if (entry.id == id && entry.sender == sender) {
+            entry.portnum = portnum;
+            return;
+        }
+    }
+    // Not tracked yet: claim the next ring slot. count starts at 0 (not 1) since scheduling our
+    // own rebroadcast is not itself a heard duplicate.
+    DupeCountEntry &slot = dupeCounts[dupeCountsNextSlot];
+    dupeCountsNextSlot = (dupeCountsNextSlot + 1) % DUPE_COUNT_TRACKER_SIZE;
+    slot.sender = sender;
+    slot.id = id;
+    slot.count = 0;
+    slot.portnum = portnum;
+}
+
+int32_t RepeatScalingModule::lookupNotedPortnum(NodeNum sender, PacketId id) const
+{
+    for (const auto &entry : dupeCounts) {
+        if (entry.id == id && entry.sender == sender)
+            return entry.portnum;
+    }
+    return -1;
+}
+
+int32_t RepeatScalingModule::resolvePortnum(const meshtastic_MeshPacket *p) const
+{
+    return (p->which_payload_variant == meshtastic_MeshPacket_decoded_tag) ? p->decoded.portnum
+                                                                           : lookupNotedPortnum(p->from, p->id);
+}
+
+uint8_t RepeatScalingModule::getToleratedDupeCount(NodeNum sender, PacketId id) const
+{
+    for (const auto &entry : dupeCounts) {
+        if (entry.id == id && entry.sender == sender)
+            return entry.count;
+    }
+    return 0;
+}
+
+bool RepeatScalingModule::shouldCancelDupe(const meshtastic_MeshPacket *p)
+{
+    const uint8_t threshold = getDupeCancelThreshold(p);
+    const uint8_t dupesHeard = registerDupeHeard(p->from, p->id);
+    const int32_t portnum = resolvePortnum(p); // for logging only
+
+    if (dupesHeard >= threshold) {
+        LOG_INFO("[REPEATSCALE] Giving up own rebroadcast of 0x%08x from=0x%08x portnum=%d: heard %u/%u duplicate(s)", p->id,
+                 p->from, portnum, dupesHeard, threshold);
+        clearDupeCount(p->from, p->id);
+        return true;
+    }
+
+    LOG_DEBUG("[REPEATSCALE] Tolerated duplicate %u/%u of 0x%08x from=0x%08x portnum=%d: will still transmit our own "
+              "rebroadcast",
+              dupesHeard, threshold, p->id, p->from, portnum);
+    return false;
+}

src/modules/RepeatScalingModule.h

@@ -0,0 +1,60 @@
+#pragma once
+
+#include "MeshTypes.h"
+#include "mesh/mesh-pb-constants.h"
+
+/**
+ * RepeatScalingModule owns the "how many duplicate rebroadcasts of a packet we ourselves have
+ * queued to rebroadcast should we tolerate before giving up" decision (see
+ * FloodingRouter::perhapsCancelDupe, its sole caller).
+ *
+ * The historical behavior is to cancel our own queued rebroadcast as soon as we hear the very
+ * first duplicate from another node. This module allows some packet types (see
+ * getDupeCancelThreshold in RepeatScalingModule.cpp) to instead tolerate a configurable number of
+ * heard duplicates first, trading a little extra airtime for better delivery odds - unless the
+ * mesh is already busy/dense, in which case it always falls back to the historical behavior.
+ */
+class RepeatScalingModule
+{
+  public:
+    RepeatScalingModule() = default;
+    virtual ~RepeatScalingModule() = default;
+
+    // Note a heard duplicate for (p->from, p->id); returns true (and clears tracking) once the
+    // per-portnum threshold is reached, meaning the caller should cancel its own rebroadcast.
+    // Virtual so FloodingRouter role-gating tests can substitute a double.
+    virtual bool shouldCancelDupe(const meshtastic_MeshPacket *p);
+
+    // Cache the portnum of a rebroadcast we've scheduled, so later encrypted duplicates of it can
+    // still be classified by getDupeCancelThreshold(). Pass -1 if it couldn't be decoded.
+    void noteScheduled(NodeNum sender, PacketId id, int32_t portnum);
+
+    // Duplicates heard (and tolerated) so far for (sender, id), or 0. For logging at TX time.
+    uint8_t getToleratedDupeCount(NodeNum sender, PacketId id) const;
+
+  protected:
+    // Duplicates to tolerate before cancelling our own rebroadcast. Virtual so tests can inject a
+    // threshold without relying on a real portnum case.
+    virtual uint8_t getDupeCancelThreshold(const meshtastic_MeshPacket *p);
+
+    // Ephemeral ring buffer of per-(sender, id) heard-duplicate counts (not persistent state).
+    uint8_t registerDupeHeard(NodeNum sender, PacketId id);
+    void clearDupeCount(NodeNum sender, PacketId id);
+    int32_t lookupNotedPortnum(NodeNum sender, PacketId id) const;
+
+  private:
+    // Decoded portnum if available, else the one cached by noteScheduled() (or -1).
+    int32_t resolvePortnum(const meshtastic_MeshPacket *p) const;
+
+    static constexpr uint8_t DUPE_COUNT_TRACKER_SIZE = 8;
+    struct DupeCountEntry {
+        NodeNum sender = 0;
+        PacketId id = 0;
+        uint8_t count = 0;
+        int32_t portnum = -1;
+    };
+    DupeCountEntry dupeCounts[DUPE_COUNT_TRACKER_SIZE];
+    uint8_t dupeCountsNextSlot = 0;
+};
+
+extern RepeatScalingModule *repeatScalingModule;