When the socket has data to send, then the TCP_KEEPALIVE timer doesn'…

…t get

activated and it takes much longer time before the socket is closed.
TCP_USER_TIMEOUT also needs to be set for the keepalive to function
when there's traffic on the tcp socket. This will cause the socket
to close when the remote end is not reachable.

Change-Id: I14c45d05992c604691f2a4c1e7ddb4c741f8fc99
Closes-Bug: #1461761

library/cpp/sandesh_session.cc

@@ -287,6 +287,7 @@ SandeshSession::SandeshSession(TcpServer *client, Socket *socket,
     keepalive_idle_time_(kSessionKeepaliveIdleTime),
     keepalive_interval_(kSessionKeepaliveInterval),
     keepalive_probes_(kSessionKeepaliveProbes),
+    tcp_user_timeout_(kSessionTcpUserTimeout),
     reader_task_id_(reader_task_id) {
     if (Sandesh::role() == Sandesh::SandeshRole::Collector) {
         send_buffer_queue_.reset(new Sandesh::SandeshBufferQueue(writer_task_id,
@@ -340,7 +341,7 @@ boost::system::error_code SandeshSession::SetSocketOptions() {
         return ec;
     }
     return SetSocketKeepaliveOptions(keepalive_idle_time_, keepalive_interval_,
-            keepalive_probes_);
+            keepalive_probes_, tcp_user_timeout_);
 }
 
 void SandeshSession::OnRead(Buffer buffer) {

library/cpp/sandesh_session.h

@@ -219,9 +219,10 @@ class SandeshSession : public TcpSession {
     friend class SandeshSessionTest;
 
     // 60 seconds - 45s + (3*5)s
-    static const int kSessionKeepaliveIdleTime = 45; // in seconds
+    static const int kSessionKeepaliveIdleTime = 15; // in seconds
     static const int kSessionKeepaliveInterval = 3; // in seconds
     static const int kSessionKeepaliveProbes = 5; // count
+    static const int kSessionTcpUserTimeout = 30000; // ms
 
     bool SendMsg(Sandesh *sandesh);
     bool SendBuffer(boost::shared_ptr<TMemoryBuffer> sbuffer);
@@ -238,6 +239,7 @@ class SandeshSession : public TcpSession {
     int keepalive_idle_time_;
     int keepalive_interval_;
     int keepalive_probes_;
+    int tcp_user_timeout_;
     int reader_task_id_;
 
     // Session statistics

library/python/pysandesh/sandesh_session.py

@@ -228,9 +228,11 @@ def _send(self, send_buf):
 
 
 class SandeshSession(TcpSession):
-    _KEEPALIVE_IDLE_TIME = 45  # in secs
+    _KEEPALIVE_IDLE_TIME = 15  # in secs
     _KEEPALIVE_INTERVAL = 3  # in secs
     _KEEPALIVE_PROBES = 5
+    _TCP_USER_TIMEOUT_OPT = 18
+    _TCP_USER_TIMEOUT_VAL = 30000 # ms
 
     def __init__(self, sandesh_instance, server, event_handler,
                  sandesh_msg_handler):
@@ -299,6 +301,12 @@ def _set_socket_options(self):
         if hasattr(socket, 'TCP_KEEPCNT'):
             self._socket.setsockopt(
                 socket.IPPROTO_TCP, socket.TCP_KEEPCNT, self._KEEPALIVE_PROBES)
+        try:
+            self._socket.setsockopt(socket.IPPROTO_TCP,
+                self._TCP_USER_TIMEOUT_OPT, self._TCP_USER_TIMEOUT_VAL)
+        except:
+            self._logger.error('setsockopt failed: option %d, value %d' %
+                (self._TCP_USER_TIMEOUT_OPT, self._TCP_USER_TIMEOUT_VAL))
     # end _set_socket_options
 
     # Private functions