Next hop reliability

src/Mesh.cpp

@@ -9,6 +9,7 @@ void Mesh::begin() {
 
 void Mesh::loop() {
   Dispatcher::loop();
+  processNextHopRetries();
 }
 
 bool Mesh::allowPacketForward(const mesh::Packet* packet) { 
@@ -30,6 +31,12 @@ uint32_t Mesh::getCADFailRetryDelay() const {
   return _rng->nextInt(1, 4)*120;
 }
 
+uint32_t Mesh::getNextHopConfirmTimeout(const Packet* packet) const {
+  // allow for the next hop's own (possibly randomised) forwarding delay, its airtime to repeat, plus margin
+  uint32_t airtime = _radio->getEstAirtimeFor(packet->getRawLength());
+  return airtime * 3 + 2000;
+}
+
 int Mesh::searchPeersByHash(const uint8_t* hash) {
   return 0;  // not found
 }
@@ -39,6 +46,11 @@ int Mesh::searchChannelsByHash(const uint8_t* hash, GroupChannel channels[], int
 }
 
 DispatcherAction Mesh::onRecvPacket(Packet* pkt) {
+  if (pkt->isRouteDirect()) {
+    // any overheard direct packet may be the next hop repeating one of ours -- check before anything else
+    checkNextHopConfirm(pkt);
+  }
+
   if (pkt->isRouteDirect() && pkt->getPayloadType() == PAYLOAD_TYPE_TRACE) {
     if (pkt->path_len < MAX_PATH_SIZE) {
       uint8_t i = 0;
@@ -102,6 +114,10 @@ DispatcherAction Mesh::onRecvPacket(Packet* pkt) {
         _tables->markSeen(pkt);
         removeSelfFromPath(pkt);
 
+        if (pkt->getPathHashCount() > 0) {   // only worth tracking if there is a further hop to overhear
+          registerNextHopConfirm(pkt);
+        }
+
         uint32_t d = getDirectRetransmitDelay(pkt);
         return ACTION_RETRANSMIT_DELAYED(0, d);  // Routed traffic is HIGHEST priority 
       }
@@ -341,6 +357,64 @@ void Mesh::removeSelfFromPath(Packet* pkt) {
   }
 }
 
+void Mesh::registerNextHopConfirm(const Packet* pkt) {
+  if (!getNextHopReliabilityEnabled() || getNextHopMaxRetries() == 0) return;
+
+  for (int i = 0; i < MAX_PENDING_NEXTHOP_CONFIRMS; i++) {
+    auto& e = _pending_confirms[i];
+    if (!e.active) {
+      e.active = true;
+      e.retries = 0;
+      e.pkt = *pkt;   // keep a copy, so it can be resent unchanged if not confirmed
+      pkt->calculatePacketHash(e.hash);
+      e.deadline = futureMillis(getNextHopConfirmTimeout(pkt));
+      return;
+    }
+  }
+  MESH_DEBUG_PRINTLN("%s Mesh::registerNextHopConfirm(): pending table full, skipping reliability tracking", getLogDateTime());
+}
+
+void Mesh::checkNextHopConfirm(const Packet* pkt) {
+  uint8_t hash[MAX_HASH_SIZE];
+  bool calculated = false;
+
+  for (int i = 0; i < MAX_PENDING_NEXTHOP_CONFIRMS; i++) {
+    auto& e = _pending_confirms[i];
+    if (!e.active) continue;
+
+    if (!calculated) {
+      pkt->calculatePacketHash(hash);
+      calculated = true;
+    }
+    if (memcmp(hash, e.hash, MAX_HASH_SIZE) == 0) {
+      e.active = false;   // next hop has repeated it -- confirmed, no retry needed
+    }
+  }
+}
+
+void Mesh::processNextHopRetries() {
+  for (int i = 0; i < MAX_PENDING_NEXTHOP_CONFIRMS; i++) {
+    auto& e = _pending_confirms[i];
+    if (!e.active || !millisHasNowPassed(e.deadline)) continue;
+
+    if (e.retries >= getNextHopMaxRetries()) {
+      MESH_DEBUG_PRINTLN("%s Mesh::processNextHopRetries(): giving up, no confirm heard after %d retries", getLogDateTime(), (uint32_t)e.retries);
+      e.active = false;
+      continue;
+    }
+
+    Packet* retry_pkt = obtainNewPacket();
+    if (retry_pkt == NULL) {
+      e.deadline = futureMillis(100);   // packet pool busy, back off briefly and try again
+      continue;
+    }
+    *retry_pkt = e.pkt;
+    e.retries++;
+    e.deadline = futureMillis(getNextHopConfirmTimeout(&e.pkt));
+    sendPacket(retry_pkt, 0);   // resend immediately, same priority as a fresh direct forward
+  }
+}
+
 DispatcherAction Mesh::routeRecvPacket(Packet* packet) {
   uint8_t n = packet->getPathHashCount();
   if (packet->isRouteFlood() && !packet->isMarkedDoNotRetransmit()
@@ -709,6 +783,10 @@ void Mesh::sendDirect(Packet* packet, const uint8_t* path, uint8_t path_len, uin
     } else {
       pri = 0;
     }
+
+    if (packet->getPathHashCount() > 0) {   // there's a next hop to listen for repeating this
+      registerNextHopConfirm(packet);
+    }
   }
   _tables->markSeen(packet); // mark this packet as already sent in case it is rebroadcast back to us
   sendPacket(packet, pri, delay_millis);

src/Mesh.h

@@ -20,6 +20,22 @@ class MeshTables {
   virtual void clear(const Packet* packet) = 0;    // remove this packet hash from table
 };
 
+#ifndef MAX_PENDING_NEXTHOP_CONFIRMS
+  #define MAX_PENDING_NEXTHOP_CONFIRMS  4   // max concurrent direct packets awaiting next-hop confirmation
+#endif
+
+/**
+ * \brief  Tracks a direct (path-routed) packet this node has repeated, while it waits to
+ *     overhear the next hop repeating it in turn (as implicit confirmation of receipt).
+*/
+struct PendingNextHopConfirm {
+  bool active;
+  uint8_t retries;
+  uint32_t deadline;      // millis() at which to retry (or give up if retries exhausted)
+  uint8_t hash[MAX_HASH_SIZE];
+  Packet pkt;             // copy of the packet as it was (re)transmitted, for resending
+};
+
 /**
  * \brief  The next layer in the basic Dispatcher task, Mesh recognises the particular Payload TYPES,
  *     and provides virtual methods for sub-classes on handling incoming, and also preparing outbound Packets.
@@ -28,12 +44,30 @@ class Mesh : public Dispatcher {
   RTCClock* _rtc;
   RNG* _rng;
   MeshTables* _tables;
+  PendingNextHopConfirm _pending_confirms[MAX_PENDING_NEXTHOP_CONFIRMS];
 
   void removeSelfFromPath(Packet* packet);
   void routeDirectRecvAcks(Packet* packet, uint32_t delay_millis);
   //void routeRecvAcks(Packet* packet, uint32_t delay_millis);
   DispatcherAction forwardMultipartDirect(Packet* pkt);
 
+  /**
+   * \brief  Start tracking 'pkt' (just repeated by this node) until the next hop is heard repeating it.
+   */
+  void registerNextHopConfirm(const Packet* pkt);
+
+  /**
+   * \brief  Check an incoming direct packet against the pending-confirm table, and mark any
+   *     match as confirmed (the next hop has repeated it, so no retry is needed).
+   */
+  void checkNextHopConfirm(const Packet* pkt);
+
+  /**
+   * \brief  Called each loop(), resends any pending packets whose confirm deadline has passed,
+   *     up to getNextHopMaxRetries() times, then drops them.
+   */
+  void processNextHopRetries();
+
 protected:
   DispatcherAction onRecvPacket(Packet* pkt) override;
 
@@ -71,6 +105,22 @@ class Mesh : public Dispatcher {
    */
   virtual uint8_t getExtraAckTransmitCount() const;
 
+  /**
+   * \returns  true if 'next-hop reliability' (listen-for-repeat retry) is enabled for repeated
+   *     direct packets. Default is enabled wherever allowPacketForward() also permits forwarding.
+   */
+  virtual bool getNextHopReliabilityEnabled() const { return true; }
+
+  /**
+   * \returns  max number of retries (resends) attempted, if no repeat from the next hop is heard.
+   */
+  virtual uint8_t getNextHopMaxRetries() const { return 3; }
+
+  /**
+   * \returns  number of milliseconds to wait for the next hop to repeat 'packet', before retrying.
+   */
+  virtual uint32_t getNextHopConfirmTimeout(const Packet* packet) const;
+
   /**
    * \brief  Perform search of local DB of peers/contacts.
    * \returns  Number of peers with matching hash
@@ -169,6 +219,7 @@ class Mesh : public Dispatcher {
   Mesh(Radio& radio, MillisecondClock& ms, RNG& rng, RTCClock& rtc, PacketManager& mgr, MeshTables& tables)
     : Dispatcher(radio, ms, mgr), _rng(&rng), _rtc(&rtc), _tables(&tables)
   {
+    memset(_pending_confirms, 0, sizeof(_pending_confirms));
   }
 
   MeshTables* getTables() const { return _tables; }