From 0102f595517c277ca320b3c0aacdbc12855fd8a6 Mon Sep 17 00:00:00 2001
From: Thomas Winget <tewinget@gmail.com>
Date: Thu, 3 Oct 2024 18:23:20 -0400
Subject: [PATCH] mainnet service node detect and exit bad state

If a node is deregistered while running (or is started before being registered),
it can fail to properly connect to the network once registered (again).  This does
not *fix* that, but these symptoms should be sufficient to detect such a state and
the node should recover upon restart.
---
 llarp/router/router.cpp             | 12 ++++++++++++
 llarp/router/router.hpp             |  1 +
 llarp/util/thread/queue_manager.hpp |  1 +
 3 files changed, 14 insertions(+)

diff --git a/llarp/router/router.cpp b/llarp/router/router.cpp
index 6a4db8c5ae..2318eee72f 100644
--- a/llarp/router/router.cpp
+++ b/llarp/router/router.cpp
@@ -634,6 +634,7 @@ namespace llarp
     log::debug(logcat, "Network ID set to {}", conf.router.m_netId);
     if (!conf.router.m_netId.empty() && strcmp(conf.router.m_netId.c_str(), llarp::DEFAULT_NETID))
     {
+      is_default_netID = false;
       const auto& netid = conf.router.m_netId;
       llarp::LogWarn(
           "!!!! you have manually set netid to be '",
@@ -928,6 +929,17 @@ namespace llarp
       return;
     // LogDebug("tick router");
     const auto now = Now();
+
+    // If a node is deregistered while running (or is started before being registered),
+    // it can fail to properly connect to the network once registered (again).  This does
+    // not *fix* that, but these symptoms should be sufficient to detect such a state and
+    // the node should recover upon restart.
+    if (is_default_netID and IsServiceNode() and uptime() > 1h and nodedb()->NumLoaded() < 10)
+    {
+      Stop();
+      return;
+    }
+
     if (const auto delta = now - _lastTick; _lastTick != 0s and delta > TimeskipDetectedDuration)
     {
       // we detected a time skip into the futre, thaw the network
diff --git a/llarp/router/router.hpp b/llarp/router/router.hpp
index 16efb1fccc..f20c4fa326 100644
--- a/llarp/router/router.hpp
+++ b/llarp/router/router.hpp
@@ -53,6 +53,7 @@ namespace llarp
   {
     llarp_time_t _lastPump = 0s;
     bool ready;
+    bool is_default_netID{true};
     // transient iwp encryption key
     fs::path transport_keyfile;
 
diff --git a/llarp/util/thread/queue_manager.hpp b/llarp/util/thread/queue_manager.hpp
index 6389d9e787..f9a55b3f5e 100644
--- a/llarp/util/thread/queue_manager.hpp
+++ b/llarp/util/thread/queue_manager.hpp
@@ -3,6 +3,7 @@
 #include <algorithm>
 #include <atomic>
 #include <cassert>
+#include <cstdint>
 #include <iostream>
 #include <limits>
 #include <string>