add stall detection

Signed-off-by: AndreasK <andreas.kirschner@mailbox.org>

Author: an-dr-eas-k

src/libsync/abstractnetworkjob.cpp

@@ -110,6 +110,40 @@ void AbstractNetworkJob::setupConnections(QNetworkReply *reply)
     connect(reply, &QNetworkReply::downloadProgress, this, &AbstractNetworkJob::networkActivity);
     connect(reply, &QNetworkReply::uploadProgress, this, &AbstractNetworkJob::networkActivity);
     connect(reply, &QNetworkReply::redirected, this, [reply, this] (const QUrl &url) { emit redirected(reply, url, 0);});
+
+    if (_stallDetectionEnabled) {
+        // Reset per-reply state so progress from a previous attempt does not
+        // carry over (e.g. after an HTTP redirect).
+        _lastTransferBytes = -1;
+        connect(reply, &QNetworkReply::uploadProgress, this, [this](qint64 bytesSent, qint64) {
+            if (bytesSent > _lastTransferBytes) {
+                _lastTransferBytes = bytesSent;
+                _stallTimer.start();
+            }
+        });
+        connect(reply, &QNetworkReply::downloadProgress, this, [this](qint64 bytesReceived, qint64) {
+            if (bytesReceived > _lastTransferBytes) {
+                _lastTransferBytes = bytesReceived;
+                _stallTimer.start();
+            }
+        });
+        _stallTimer.start();
+    }
+}
+
+void AbstractNetworkJob::enableStallDetection(int timeoutMs)
+{
+    _stallDetectionEnabled = true;
+    _stallTimer.setSingleShot(true);
+    _stallTimer.setInterval(timeoutMs);
+    connect(&_stallTimer, &QTimer::timeout, this, [this]() {
+        qCWarning(lcNetworkJob) << "Transfer stalled: no bytes transferred for"
+                                << _stallTimer.interval() / 1000 << "seconds on" << path();
+        emit transferStalled();
+        if (reply()) {
+            reply()->abort();
+        }
+    });
 }
 
 QNetworkReply *AbstractNetworkJob::addTimer(QNetworkReply *reply)
@@ -175,6 +209,7 @@ QUrl AbstractNetworkJob::makeDavUrl(const QString &relativePath) const
 void AbstractNetworkJob::slotFinished()
 {
     _timer.stop();
+    _stallTimer.stop();
 
     if (_reply->error() == QNetworkReply::SslHandshakeFailedError) {
         qCWarning(lcNetworkJob) << "SslHandshakeFailedError: " << errorString() << " : can be caused by a webserver wanting SSL client certificates";

src/libsync/abstractnetworkjob.h

@@ -110,6 +110,19 @@ class OWNCLOUDSYNC_EXPORT AbstractNetworkJob : public QObject
     /// Returns a standardised error message in case of HSTS errors
     [[nodiscard]] static std::optional<QString> hstsErrorStringFromReply(QNetworkReply *reply);
 
+public:
+    /**
+     * Enables per-job stall detection for uploads and downloads.
+     *
+     * Once enabled, a timer is started when the job begins transferring data.
+     * The timer is reset whenever progress is made. If no bytes are transferred
+     * within @p timeoutMs milliseconds, the transferStalled() signal is emitted
+     * and the reply is aborted.
+     *
+     * Call this from a subclass's start() before invoking sendRequest().
+     */
+    void enableStallDetection(int timeoutMs = 30 * 1000);
+
 public slots:
     void setTimeout(qint64 msec);
     void resetTimeout();
@@ -121,6 +134,12 @@ public slots:
     void networkError(QNetworkReply *reply);
     void networkActivity();
 
+    /**
+     * Emitted when stall detection is active and no bytes have been transferred
+     * within the configured timeout. The reply is aborted immediately after.
+     */
+    void transferStalled();
+
     /** Emitted when a redirect is followed.
      *
      * \a reply The "please redirect" reply
@@ -220,6 +239,10 @@ private slots:
     //
     // Reparented to the currently running QNetworkReply.
     QPointer<QIODevice> _requestBody;
+
+    QTimer _stallTimer;
+    qint64 _lastTransferBytes = -1;
+    bool _stallDetectionEnabled = false;
 };
 
 /**

src/libsync/propagatedownload.cpp

@@ -119,6 +119,8 @@ void GETFileJob::start()
     req.setPriority(QNetworkRequest::LowPriority); // Long downloads must not block non-propagation jobs.
     req.setDecompressedSafetyCheckThreshold(_decompressionThresholdBase + CustomDecompressedSafetyCheckThreshold);
 
+    enableStallDetection();
+
     if (_directDownloadUrl.isEmpty()) {
         sendRequest("GET", makeDavUrl(path()), req);
     } else {
@@ -745,6 +747,11 @@ void PropagateDownloadFile::startDownload()
     _job->setBandwidthManager(&propagator()->_bandwidthManager);
     connect(_job.data(), &GETFileJob::finishedSignal, this, &PropagateDownloadFile::slotGetFinished);
     connect(_job.data(), &GETFileJob::downloadProgress, this, &PropagateDownloadFile::slotDownloadProgress);
+    connect(_job.data(), &AbstractNetworkJob::transferStalled, this, [this]() {
+        qCWarning(lcPropagateDownload) << "Download stalled for" << _item->_file << "- scheduling retry.";
+        _job->cancel();
+        done(SyncFileItem::SoftError, tr("Download stalled: no data was transferred. The download will be retried."), ErrorCategory::GenericError);
+    });
     propagator()->_activeJobList.append(this);
     _job->start();
 }

src/libsync/propagateupload.cpp

@@ -57,6 +57,8 @@ void PUTFileJob::start()
 
     req.setPriority(QNetworkRequest::LowPriority); // Long uploads must not block non-propagation jobs.
 
+    enableStallDetection();
+
     auto requestID = QByteArray{};
 
     if (_url.isValid()) {

src/libsync/propagateuploadng.cpp

@@ -391,6 +391,10 @@ void PropagateUploadFileNG::startNextChunk()
     connect(job, &PUTFileJob::uploadProgress,
         devicePtr, &UploadDevice::slotJobUploadProgress);
     connect(job, &QObject::destroyed, this, &PropagateUploadFileCommon::slotJobDestroyed);
+    connect(job, &AbstractNetworkJob::transferStalled, this, [this]() {
+        qCWarning(lcPropagateUploadNG) << "Upload stalled for" << _item->_file << "- scheduling retry.";
+        abortWithError(SyncFileItem::SoftError, tr("Upload stalled: no data was transferred. The upload will be retried."));
+    });
     job->start();
     propagator()->_activeJobList.append(this);
     _currentChunk++;

src/libsync/propagateuploadv1.cpp

@@ -160,6 +160,10 @@ void PropagateUploadFileV1::startNextChunk()
     connect(job, &PUTFileJob::uploadProgress, this, &PropagateUploadFileV1::slotUploadProgress);
     connect(job, &PUTFileJob::uploadProgress, devicePtr, &UploadDevice::slotJobUploadProgress);
     connect(job, &QObject::destroyed, this, &PropagateUploadFileCommon::slotJobDestroyed);
+    connect(job, &AbstractNetworkJob::transferStalled, this, [this]() {
+        qCWarning(lcPropagateUploadV1) << "Upload stalled for" << _item->_file << "- scheduling retry.";
+        abortWithError(SyncFileItem::SoftError, tr("Upload stalled: no data was transferred. The upload will be retried."));
+    });
     if (isFinalChunk)
         adjustLastJobTimeout(job, fileSize);
     job->start();