Keep Alive timeout with auto reconnect

.changeset/keepalive-timeout-reconnect.md

@@ -0,0 +1,5 @@
+---
+'firmware': minor
+---
+
+feat: add a keepalive ping timeout that forces a reconnect on backend connection loss

include/GatewayClient.h

@@ -1,4 +1,4 @@
 #pragma once
 
 #include "Common.h"
 #include "GatewayClientState.h"
@@ -13,22 +13,24 @@
 
 namespace OpenShock {
   class GatewayClient {
     DISABLE_COPY(GatewayClient);
     DISABLE_MOVE(GatewayClient);
 
   public:
     GatewayClient(const std::string& authToken);
     ~GatewayClient();
 
     inline GatewayClientState state() const { return m_state; }
 
     void connect(const std::string& host, uint16_t port, const std::string& path);
     void disconnect();
 
     bool sendMessageTXT(std::string_view data);
     bool sendMessageBIN(tcb::span<const uint8_t> data);
 
+    void markPingReceived();
+
     bool loop();
 
   private:
     void _setState(GatewayClientState state);
@@ -37,5 +39,6 @@
 
     WebSocketsClient m_webSocket;
     GatewayClientState m_state;
+    int64_t m_lastPingTimestamp;
   };
 }  // namespace OpenShock

include/GatewayConnectionManager.h

@@ -1,4 +1,4 @@
 #pragma once
 
 #include "AccountLinkResultCode.h"
 #include "span.h"
@@ -19,5 +19,7 @@
   bool SendMessageTXT(std::string_view data);
   bool SendMessageBIN(tcb::span<const uint8_t> data);
 
+  void MarkPingReceived();
+
   void Update();
 }  // namespace OpenShock::GatewayConnectionManager

src/GatewayClient.cpp

@@ -1,3 +1,3 @@
 #include "GatewayClient.h"
 
 const char* const TAG = "GatewayClient";
@@ -14,11 +14,14 @@
 
 using namespace OpenShock;
 
+const int64_t GATEWAY_PING_TIMEOUT = 90'000;
+
 static bool s_bootStatusSent = false;
 
 GatewayClient::GatewayClient(const std::string& authToken)
   : m_webSocket()
   , m_state(GatewayClientState::Disconnected)
+  , m_lastPingTimestamp(0)
 {
   OS_LOGD(TAG, "Creating GatewayClient");
 
@@ -90,6 +93,11 @@
   return m_webSocket.sendBIN(data.data(), data.size());
 }
 
+void GatewayClient::markPingReceived()
+{
+  m_lastPingTimestamp = OpenShock::millis();
+}
+
 bool GatewayClient::loop()
 {
   if (m_state == GatewayClientState::Disconnected) {
@@ -104,6 +112,13 @@
     return true;
   }
 
+  if (m_lastPingTimestamp != 0 && (OpenShock::millis() - m_lastPingTimestamp) > GATEWAY_PING_TIMEOUT) {
+    OS_LOGW(TAG, "No ping received from gateway for %lld ms, forcing reconnect", GATEWAY_PING_TIMEOUT);
+    m_webSocket.disconnect();
+    _setState(GatewayClientState::Disconnected);
+    return false;
+  }
+
   return true;
 }
 
@@ -160,6 +175,7 @@
       _setState(GatewayClientState::Disconnected);
       break;
     case WStype_CONNECTED:
+      m_lastPingTimestamp = 0;
       _setState(GatewayClientState::Connected);
       _sendBootStatus();
       break;

src/GatewayConnectionManager.cpp

@@ -1,3 +1,3 @@
 #include "GatewayConnectionManager.h"
 
 const char* const TAG = "GatewayConnectionManager";
@@ -194,6 +194,16 @@
   return client->sendMessageBIN(data);
 }
 
+void GatewayConnectionManager::MarkPingReceived()
+{
+  auto client = GetClient();
+  if (client == nullptr) {
+    return;
+  }
+
+  client->markPingReceived();
+}
+
 bool FetchHubInfo(std::string authToken)
 {
   // TODO: this function is very slow, should be optimized!

src/message_handlers/websocket/gateway/Ping.cpp

@@ -18,5 +18,6 @@ void _Private::HandlePing(const OpenShock::Serialization::Gateway::GatewayToHubM
     return;
   }
 
+  GatewayConnectionManager::MarkPingReceived();
   Serialization::Gateway::SerializePongMessage(GatewayConnectionManager::SendMessageBIN);
 }