From 014798e50b2de37ed5243c7e622fc290dfda8606 Mon Sep 17 00:00:00 2001
From: bitflash-maintainer <noreply@bitflash.invalid>
Date: Wed, 2 Sep 2026 09:03:33 +0000
Subject: worker: fail over across seed relays when the advertised meeting node
 keeps failing

---
 src/main.cpp | 29 +++++++++++++++++++++++++++++
 1 file changed, 29 insertions(+)

(limited to 'src')

diff --git a/src/main.cpp b/src/main.cpp
index c0a8825..0414714 100644
--- a/src/main.cpp
+++ b/src/main.cpp
@@ -3739,6 +3739,33 @@ static bool PoolParticipantMiner()
     }
     LogPrint("worker", "[worker] pool resolved to meeting node %s\n", meetingHostPort.c_str());
 
+    // The descriptor names ONE meeting node, but the pool parks a
+    // registration at every relay it knows (one accept thread per relay,
+    // rpc.cpp:AcceptOneFn), so any seed relay can pair us. Insisting on the
+    // advertised entry forever turns one bad path into a miner that idles
+    // indefinitely while a working path exists (issue #17); after a few
+    // consecutive tunnel failures, rotate the seed relays into the attempts
+    // too. Only ONE participant session runs (StartMinerThreads), so these
+    // statics are single-threaded.
+    static const int TUNNEL_FAILOVER_AFTER = 3;
+    static int nTunnelFails = 0;      // consecutive tunnel failures, this pool
+    static size_t iFailoverRelay = 0; // rotation cursor over vBtfMeetingRelays
+    static std::string strFailPool;   // pool the counters belong to
+    if (strFailPool != strParticipantPool) {
+        strFailPool = strParticipantPool;
+        nTunnelFails = 0;
+        iFailoverRelay = 0;
+    }
+    if (nTunnelFails >= TUNNEL_FAILOVER_AFTER && !vBtfMeetingRelays.empty()) {
+        std::string alt = vBtfMeetingRelays[iFailoverRelay++ % vBtfMeetingRelays.size()];
+        if (alt != meetingHostPort) {
+            LogPrint("worker", "[worker] advertised meeting node %s failed %d"
+                     " times in a row -- trying seed relay %s instead\n",
+                     meetingHostPort.c_str(), nTunnelFails, alt.c_str());
+            meetingHostPort = alt;
+        }
+    }
+
     size_t colon = meetingHostPort.rfind(':');
     if (colon == std::string::npos) {
         LogPrint("worker", "[worker] bad meeting node format '%s'\n", meetingHostPort.c_str());
@@ -3762,12 +3789,14 @@ static bool PoolParticipantMiner()
     s = btf::BtfClientTunnel(host.c_str(), (unsigned short)port,
                              target_pubkey, service_enc_pub, &tunnelFail);
     if (s == INVALID_SOCKET) {
+        nTunnelFails++;
         LogPrint("worker", "[worker] tunnel connect FAILED via %s: %s\n",
                  meetingHostPort.c_str(), btf::TunnelFailText(tunnelFail));
         SetParticipantMiningStatus(std::string("tunnel connect failed: ")
                                    + btf::TunnelFailText(tunnelFail));
         return false;
     }
+    nTunnelFails = 0;
     LogPrint("worker", "[worker] tunnel open via %s\n", meetingHostPort.c_str());
     SetParticipantMiningStatus("tunnel open");
     }
-- 
cgit v1.2.3