Detect stuck outgoing transactions and re-relay after daemon restart

Field report: a send sat 'in mempool' for two days while the network had
never seen it (Tor down at send time defeats both the multi-broadcast and
the embedded daemon's exclusive --tx-proxy relay), locking the change.

Three additive changes, all strictly local (no txid is ever queried
against an external service):

- onMultiBroadcast now counts relay responses; if no node accepted the
  broadcast within 45s the user is told immediately and offered the
  existing rebroadcast path, instead of silence.
- A periodic local history scan warns once per outgoing tx that has been
  pending for more than 25 minutes, again offering rebroadcast (which
  already reconnects to a fresh node first).
- On DaemonManager::synced, still-pending outgoing txs are re-announced
  from the wallet's own tx cache; a restarted embedded node's pool has
  forgotten them.

CMakeLists.txt

@@ -1,7 +1,7 @@
 cmake_minimum_required(VERSION 3.18)
 
 project(feather
-  VERSION "5.0.2"
+  VERSION "5.0.3"
   DESCRIPTION "A free Monero desktop wallet"
   LANGUAGES CXX C ASM
 )

src/MainWindow.cpp

@@ -542,6 +542,24 @@ void MainWindow::initWalletContext() {
     connect(m_wallet, &Wallet::deviceError,         this, &MainWindow::onDeviceError);
     
     connect(m_wallet, &Wallet::multiBroadcast,      this, &MainWindow::onMultiBroadcast);
+
+    // wowlet: count multi-broadcast relay results so a total failure (Tor not
+    // bootstrapped, every node unreachable) is surfaced instead of silent.
+    connect(m_rpc, &DaemonRpc::ApiResponse, this, [this](const DaemonRpc::DaemonResponse &resp){
+        if (resp.endpoint == DaemonRpc::Endpoint::SEND_RAW_TRANSACTION && resp.ok) {
+            m_broadcastBatchSuccesses++;
+        }
+    });
+
+    // wowlet: when the embedded daemon reports synced after a (re)start,
+    // re-announce any still-pending outgoing txs; a restarted pool has
+    // forgotten them.
+    connect(daemonManager(), &DaemonManager::synced, this, &MainWindow::relayPendingTransactions);
+
+    // wowlet: periodic local check for outgoing txs stuck unconfirmed.
+    m_stuckTxTimer = new QTimer(this);
+    connect(m_stuckTxTimer, &QTimer::timeout, this, &MainWindow::checkStuckTransactions);
+    m_stuckTxTimer->start(2 * 60 * 1000);
 }
 
 void MainWindow::menuToggleTabVisible(const QString &key){
@@ -760,6 +778,7 @@ void MainWindow::onSubtractFeeFromAmountEnabled(bool enabled) {
 }
 
 void MainWindow::onMultiBroadcast(const QMap<QString, QString> &txHexMap) {
+    m_broadcastBatchSuccesses = 0;
     QMapIterator<QString, QString> i(txHexMap);
     while (i.hasNext()) {
         i.next();
@@ -770,6 +789,99 @@ void MainWindow::onMultiBroadcast(const QMap<QString, QString> &txHexMap) {
             m_rpc->sendRawTransaction(i.value());
         }
     }
+
+    // wowlet: if not a single node accepted the relay, say so. The commit to
+    // the local daemon may still confirm on its own, but when Tor is down and
+    // the daemon runs with an exclusive --tx-proxy, nothing will ever leave
+    // this machine; the user should know within a minute, not after two days.
+    const QStringList txids = txHexMap.keys();
+    QTimer::singleShot(45 * 1000, this, [this, txids]{
+        if (m_broadcastBatchSuccesses > 0 || txids.isEmpty()) {
+            return;
+        }
+        QMessageBox box(this);
+        box.setWindowTitle("Transaction relay may have failed");
+        box.setText("No node accepted the transaction broadcast. If Tor is "
+                    "still starting (or unavailable), the transaction may not "
+                    "reach the network at all.");
+        box.setInformativeText("Rebroadcasting later is always safe: a "
+                               "transaction can never be spent twice.");
+        QPushButton *retry = box.addButton("Rebroadcast now", QMessageBox::YesRole);
+        box.addButton("Later", QMessageBox::RejectRole);
+        box.exec();
+        if (box.clickedButton() == retry) {
+            this->onResendTransaction(txids[0]);
+        }
+    });
+}
+
+void MainWindow::checkStuckTransactions() {
+    // wowlet: a pending outgoing tx unconfirmed for this long has, in
+    // practice, not reached the network (Tor down at send time, embedded
+    // node restarted mid-relay). Local history scan only; no network calls,
+    // and no txid is ever looked up against any external service. The remedy
+    // reuses the existing rebroadcast path, which reconnects to a fresh node
+    // first.
+    constexpr qint64 kStuckSecs = 25 * 60;
+    if (m_wallet == nullptr || m_wallet->history() == nullptr) {
+        return;
+    }
+    TransactionHistory *history = m_wallet->history();
+    for (quint64 i = 0; i < history->count(); i++) {
+        const TransactionRow &row = history->transaction(static_cast<int>(i));
+        if (!row.pending || row.direction == TransactionRow::Direction_In) {
+            continue;
+        }
+        if (row.timestamp.secsTo(QDateTime::currentDateTime()) < kStuckSecs) {
+            continue;
+        }
+        if (m_stuckTxWarned.contains(row.hash)) {
+            continue;
+        }
+        m_stuckTxWarned.insert(row.hash);
+        QMessageBox box(this);
+        box.setWindowTitle("Transaction appears stuck");
+        box.setText(QString("An outgoing transaction has been unconfirmed for "
+                            "more than %1 minutes:\n\n%2")
+                        .arg(kStuckSecs / 60).arg(row.hash));
+        box.setInformativeText("It may never have reached the network. "
+                               "Rebroadcasting is safe: a transaction can "
+                               "never be spent twice, and your change stays "
+                               "locked until this either confirms or fails.");
+        QPushButton *retry = box.addButton("Rebroadcast now", QMessageBox::YesRole);
+        box.addButton("Later", QMessageBox::RejectRole);
+        box.exec();
+        if (box.clickedButton() == retry) {
+            this->onResendTransaction(row.hash);
+        }
+    }
+}
+
+void MainWindow::relayPendingTransactions() {
+    // wowlet: rebuild the multi-broadcast map from the wallet's own tx cache
+    // for every pending outgoing tx and re-announce it. Strictly local
+    // sources; the announcement itself goes through the normal (Tor-routed)
+    // broadcast path.
+    if (m_wallet == nullptr || m_wallet->history() == nullptr) {
+        return;
+    }
+    TransactionHistory *history = m_wallet->history();
+    QMap<QString, QString> txHexMap;
+    for (quint64 i = 0; i < history->count(); i++) {
+        const TransactionRow &row = history->transaction(static_cast<int>(i));
+        if (!row.pending || row.direction == TransactionRow::Direction_In) {
+            continue;
+        }
+        QString txHex = m_wallet->getCacheTransaction(row.hash);
+        if (!txHex.isEmpty()) {
+            txHexMap[row.hash] = txHex;
+        }
+    }
+    if (txHexMap.isEmpty()) {
+        return;
+    }
+    qDebug() << QString("Re-relaying %1 pending transaction(s) after daemon sync").arg(txHexMap.size());
+    this->onMultiBroadcast(txHexMap);
 }
 
 void MainWindow::onSyncStatus(quint64 height, quint64 target, bool daemonSync) {

src/MainWindow.h

@@ -51,6 +51,9 @@ namespace Ui {
     class MainWindow;
 }
 
+#include <QSet>
+#include <QTimer>
+
 class ToggleTab : QObject {
 Q_OBJECT
 
@@ -214,12 +217,23 @@ private:
     QIcon hardwareDevicePairedIcon();
     QIcon hardwareDeviceUnpairedIcon();
 
+    // wowlet: stuck-transaction detection and pending re-relay. Both are
+    // strictly local: history rows and the wallet's own tx cache; no txid is
+    // ever queried against any external service.
+    void checkStuckTransactions();
+    void relayPendingTransactions();
+
     QScopedPointer<Ui::MainWindow> ui;
     WindowManager *m_windowManager;
     Wallet *m_wallet = nullptr;
     Nodes *m_nodes;
     DaemonRpc *m_rpc;
 
+    // wowlet: see checkStuckTransactions / relayPendingTransactions
+    QTimer *m_stuckTxTimer = nullptr;
+    QSet<QString> m_stuckTxWarned;
+    int m_broadcastBatchSuccesses = 0;
+
     SplashDialog *m_splashDialog = nullptr;
     AccountSwitcherDialog *m_accountSwitcherDialog = nullptr;
     TxPoolViewerDialog *m_txPoolViewerDialog = nullptr;

src/utils/DaemonManager.cpp

@@ -4,6 +4,7 @@
 #include "utils/DaemonManager.h"
 
 #include <QCoreApplication>
+#include <QDateTime>
 #include <QDir>
 #include <QDirIterator>
 #include <QEventLoop>
@@ -62,6 +63,11 @@ void DaemonManager::start() {
     if (state == QProcess::ProcessState::Running || state == QProcess::ProcessState::Starting)
         return;
 
+    // wowlet: a deliberate (re)start clears the stop latch, so a later UNEXPECTED exit can auto-restart
+    // again. Without this, m_stopping stays true after the first node toggle-off and the crash
+    // auto-restart is silently dead for the rest of the session.
+    m_stopping = false;
+
     if (!this->unpackBins()) {
         this->setErrorMessage("Unable to unpack the embedded wownerod binary.");
         return;
@@ -85,6 +91,14 @@ void DaemonManager::start() {
         // and surfaces a clear error via handleProcessError().
     }
 
+    // wowlet: the retry cap guards against a rapid crash-loop, not legit restarts. If the node had been
+    // up a while before this (re)start, clear the budget — otherwise a few node toggles over a session
+    // trip "maximum retries exceeded" and it won't start again until wowlet is relaunched.
+    const qint64 nowMs = QDateTime::currentMSecsSinceEpoch();
+    if (m_lastSpawnMs != 0 && (nowMs - m_lastSpawnMs) > 60000)
+        m_restarts = 0;
+    m_lastSpawnMs = nowMs;
+
     m_restarts += 1;
     if (m_restarts > 4) {
         this->setErrorMessage("wownerod failed to start: maximum retries exceeded");
@@ -108,6 +122,14 @@ void DaemonManager::start() {
     arguments << "--no-igd";                 // no UPnP port mapping
     arguments << "--out-peers" << "16";
     arguments << "--log-file" << QDir(dataDir).filePath("wownerod.log");
+    // wowlet: disable LMDB batch write mode so every block is committed in its own transaction.
+    // Without this, wownero accumulates blocks in a single large write transaction that is ABORTED
+    // (not committed) when the daemon shuts down mid-sync — causing all progress since the last
+    // batch boundary to be silently lost. This is visible in wownerod.log: sessions consistently
+    // revert to a stale height despite a full graceful shutdown sequence. Using "safe" mode is
+    // slower for initial sync but is the only correct choice for an embedded node that starts and
+    // stops with the wallet; the wownero chain is small enough that the overhead is acceptable.
+    arguments << "--db-sync-mode" << "safe";
 
     // Full archival by default (wownero's chain is small); prune is an explicit opt-in.
     if (conf()->get(Config::pruneBlockchain).toBool())

src/utils/DaemonManager.h

@@ -73,6 +73,7 @@ private:
     bool m_synced = false;
     bool m_stopping = false;
     int m_restarts = 0;
+    qint64 m_lastSpawnMs = 0;   // wowlet: wall-clock of last spawn — ages out the crash-retry budget
 
     static QPointer<DaemonManager> m_instance;
 };