Port changes to fix the keep alive timeout (#9483)

CHANGELOG

@@ -4,6 +4,8 @@ v3.5.0-rc.5 (2019-XX-XX)
 * coordinator code was reporting rocksdb error codes, but not the associated detail message.
   Corrected.
 
+* The keep alive timeout specified via --http.keep-alive-timeout is now being honored
+
 * Replication requests on Document API are now on higher priority then client-triggered requests.
   This should help to keep in sync replication up and running even if the server is overloaded.
 

arangod/GeneralServer/HttpCommTask.cpp

@@ -109,10 +109,11 @@ void HttpCommTask::addResponse(GeneralResponse& baseResponse, RequestStatistics*
 #endif
 
   finishExecution(baseResponse);
-  resetKeepAlive();
 
   // response has been queued, allow further requests
   _requestPending = false;
+  
+  resetKeepAlive();
 
   // CORS response handling
   if (!_origin.empty()) {
@@ -220,12 +221,14 @@ void HttpCommTask::addResponse(GeneralResponse& baseResponse, RequestStatistics*
 bool HttpCommTask::processRead(double startTime) {
   TRI_ASSERT(_peer->runningInThisThread());
 
-  cancelKeepAlive();
   TRI_ASSERT(_readBuffer.c_str() != nullptr);
 
   if (_requestPending) {
     return false;
   }
+  
+  // starts and extends the keep-alive timeout
+  resetKeepAlive(); // will extend the Keep-Alive timeout
 
   bool handleRequest = false;
 
@@ -603,6 +606,7 @@ bool HttpCommTask::processRead(double startTime) {
 
 void HttpCommTask::processRequest(std::unique_ptr<HttpRequest> request) {
   TRI_ASSERT(_peer->runningInThisThread());
+  cancelKeepAlive(); // timeout will be restarted
 
   {
     LOG_TOPIC("6e770", DEBUG, Logger::REQUESTS)

arangod/GeneralServer/SocketTask.cpp

@@ -230,31 +230,36 @@ void SocketTask::addToReadBuffer(char const* data, std::size_t len) {
 
 // does not need lock
 void SocketTask::resetKeepAlive() {
-  if (_useKeepAliveTimer) {
-    asio_ns::error_code err;
-    _keepAliveTimer->expires_from_now(_keepAliveTimeout, err);
-    if (err) {
-      closeStream();
-      return;
-    }
-
-    _keepAliveTimerActive.store(true, std::memory_order_relaxed);
-    _keepAliveTimer->async_wait([self = shared_from_this()](asio_ns::error_code const& error) {
-      if (!error) {  // error will be true if timer was canceled
-        LOG_TOPIC("5c1e0", ERR, Logger::COMMUNICATION)
-            << "keep alive timout - closing stream!";
-        self->closeStream();
-      }
-    });
+  if (!_useKeepAliveTimer) {
+    return;
   }
+  
+  // expires_from_now cancels pending operations
+  asio_ns::error_code err;
+  _keepAliveTimer->expires_from_now(_keepAliveTimeout, err);
+  if (err) {
+    closeStream();
+    return;
+  }
+  _keepAliveTimerActive.store(true);
+  
+  std::weak_ptr<SocketTask> self = shared_from_this();
+  _keepAliveTimer->async_wait([self](const asio_ns::error_code& ec) {
+    if (!ec) {  // error will be true if timer was canceled
+      LOG_TOPIC("f0948", INFO, Logger::COMMUNICATION)
+      << "keep alive timout - closing stream!";
+      if (auto s = self.lock()) {
+        s->closeStream();
+      }
+    }
+  });
 }
 
 // caller must hold the _lock
 void SocketTask::cancelKeepAlive() {
-  if (_useKeepAliveTimer && _keepAliveTimerActive.load(std::memory_order_relaxed)) {
+  if (_keepAliveTimerActive.exchange(false)) {
     asio_ns::error_code err;
     _keepAliveTimer->cancel(err);
-    _keepAliveTimerActive.store(false, std::memory_order_relaxed);
   }
 }
 

arangod/GeneralServer/SocketTask.h

@@ -147,7 +147,7 @@ class SocketTask : public std::enable_shared_from_this<SocketTask> {
   // caller must run in _peer->strand()
   void closeStreamNoLock();
 
-  // starts the keep alive time, no need to run on strand
+  // starts the keep alive time
   void resetKeepAlive();
 
   // cancels the keep alive timer