Add heartbeat interval parameter.

orchagent/main.cpp

@@ -62,6 +62,9 @@ extern bool gIsNatSupported;
 #define SWSS_RECORD_ENABLE (0x1 << 1)
 #define RESPONSE_PUBLISHER_RECORD_ENABLE (0x1 << 2)
 
+/* orchagent heart beat message interval */
+#define HEART_BEAT_INTERVAL_MSECS_DEFAULT 10 * 1000
+
 string gMySwitchType = "";
 int32_t gVoqMySwitchId = -1;
 int32_t gVoqMaxCores = 0;
@@ -73,7 +76,7 @@ uint32_t create_switch_timeout = 0;
 
 void usage()
 {
-    cout << "usage: orchagent [-h] [-r record_type] [-d record_location] [-f swss_rec_filename] [-j sairedis_rec_filename] [-b batch_size] [-m MAC] [-i INST_ID] [-s] [-z mode] [-k bulk_size] [-q zmq_server_address] [-c mode] [-t create_switch_timeout] [-v VRF]" << endl;
+    cout << "usage: orchagent [-h] [-r record_type] [-d record_location] [-f swss_rec_filename] [-j sairedis_rec_filename] [-b batch_size] [-m MAC] [-i INST_ID] [-s] [-z mode] [-k bulk_size] [-q zmq_server_address] [-c mode] [-t create_switch_timeout] [-v VRF] [-I heart_beat_interval]" << endl;
     cout << "    -h: display this message" << endl;
     cout << "    -r record_type: record orchagent logs with type (default 3)" << endl;
     cout << "                    Bit 0: sairedis.rec, Bit 1: swss.rec, Bit 2: responsepublisher.rec. For example:" << endl;
@@ -95,6 +98,7 @@ void usage()
     cout << "    -c counter mode (traditional|asic_db), default: asic_db" << endl;
     cout << "    -t Override create switch timeout, in sec" << endl;
     cout << "    -v vrf: VRF name (default empty)" << endl;
+    cout << "    -I heart_beat_interval: Heart beat interval in millisecond (default 10)" << endl;
 }
 
 void sighup_handler(int signo)
@@ -349,8 +353,9 @@ int main(int argc, char **argv)
     bool   enable_zmq = false;
     string responsepublisher_rec_filename = Recorder::RESPPUB_FNAME;
     int record_type = 3; // Only swss and sairedis recordings enabled by default.
+    long heartBeatInterval = HEART_BEAT_INTERVAL_MSECS_DEFAULT;
 
-    while ((opt = getopt(argc, argv, "b:m:r:f:j:d:i:hsz:k:q:c:t:v:")) != -1)
+    while ((opt = getopt(argc, argv, "b:m:r:f:j:d:i:hsz:k:q:c:t:v:I:")) != -1)
     {
         switch (opt)
         {
@@ -450,6 +455,22 @@ int main(int argc, char **argv)
                 vrf = optarg;
             }
             break;
+        case 'I':
+            if (optarg)
+            {
+                auto interval = atoi(optarg);
+                if (interval >= 0)
+                {
+                    heartBeatInterval = interval;
+                    SWSS_LOG_NOTICE("Setting heartbeat interval as %ld", heartBeatInterval);
+                }
+                else
+                {
+                    heartBeatInterval = HEART_BEAT_INTERVAL_MSECS_DEFAULT;
+                    SWSS_LOG_ERROR("Invalid input for heartbeat interval: %d. use default interval: %ld", interval, heartBeatInterval);
+                }
+            }
+            break;
         default: /* '?' */
             exit(EXIT_FAILURE);
         }
@@ -815,7 +836,7 @@ int main(int argc, char **argv)
         syncd_apply_view();
     }
 
-    orchDaemon->start();
+    orchDaemon->start(heartBeatInterval);
 
     return 0;
 }

orchagent/orchdaemon.cpp

@@ -23,9 +23,6 @@ using namespace swss;
 #define APP_FABRIC_MONITOR_PORT_TABLE_NAME      "FABRIC_PORT_TABLE"
 #define APP_FABRIC_MONITOR_DATA_TABLE_NAME      "FABRIC_MONITOR_TABLE"
 
-/* orchagent heart beat message interval */
-#define HEART_BEAT_INTERVAL_MSECS 10 * 1000
-
 extern sai_switch_api_t*           sai_switch_api;
 extern sai_object_id_t             gSwitchId;
 extern string                      gMySwitchType;
@@ -827,7 +824,7 @@ void OrchDaemon::logRotate() {
 }
 
 
-void OrchDaemon::start()
+void OrchDaemon::start(long heartBeatInterval)
 {
     SWSS_LOG_ENTER();
 
@@ -848,7 +845,7 @@ void OrchDaemon::start()
         ret = m_select->select(&s, SELECT_TIMEOUT);
 
         auto tend = std::chrono::high_resolution_clock::now();
-        heartBeat(tend);
+        heartBeat(tend, heartBeatInterval);
 
         auto diff = std::chrono::duration_cast<std::chrono::milliseconds>(tend - tstart);
 
@@ -925,7 +922,7 @@ void OrchDaemon::start()
                     flush();
 
                     SWSS_LOG_WARN("Orchagent is frozen for warm restart!");
-                    freezeAndHeartBeat(UINT_MAX);
+                    freezeAndHeartBeat(UINT_MAX, heartBeatInterval);
                 }
             }
         }
@@ -1089,25 +1086,31 @@ void OrchDaemon::addOrchList(Orch *o)
     m_orchList.push_back(o);
 }
 
-void OrchDaemon::heartBeat(std::chrono::time_point<std::chrono::high_resolution_clock> tcurrent)
+void OrchDaemon::heartBeat(std::chrono::time_point<std::chrono::high_resolution_clock> tcurrent, long interval)
 {
+    if (interval == 0)
+    {
+        // disable heart beat feature when interval is 0
+        return;
+    }
+
     // output heart beat message to SYSLOG
     auto diff = std::chrono::duration_cast<std::chrono::milliseconds>(tcurrent - m_lastHeartBeat);
-    if (diff.count() >= HEART_BEAT_INTERVAL_MSECS)
+    if (diff.count() >= interval)
     {
         m_lastHeartBeat = tcurrent;
         // output heart beat message to supervisord with 'PROCESS_COMMUNICATION_STDOUT' event: http://supervisord.org/events.html
         cout << "<!--XSUPERVISOR:BEGIN-->heartbeat<!--XSUPERVISOR:END-->" << endl;
     }
 }
 
-void OrchDaemon::freezeAndHeartBeat(unsigned int duration)
+void OrchDaemon::freezeAndHeartBeat(unsigned int duration, long interval)
 {
     while (duration > 0)
     {
         // Send heartbeat message to prevent Orchagent stuck alert.
         auto tend = std::chrono::high_resolution_clock::now();
-        heartBeat(tend);
+        heartBeat(tend, interval);
 
         duration--;
         sleep(1);

orchagent/orchdaemon.h

@@ -63,7 +63,7 @@ class OrchDaemon
     ~OrchDaemon();
 
     virtual bool init();
-    void start();
+    void start(long heartBeatInterval);
     bool warmRestoreAndSyncUp();
     void getTaskToSync(vector<string> &ts);
     bool warmRestoreValidation();
@@ -102,9 +102,9 @@ class OrchDaemon
 
     void flush();
 
-    void heartBeat(std::chrono::time_point<std::chrono::high_resolution_clock> tcurrent);
+    void heartBeat(std::chrono::time_point<std::chrono::high_resolution_clock> tcurrent, long interval);
 
-    void freezeAndHeartBeat(unsigned int duration);
+    void freezeAndHeartBeat(unsigned int duration, long interval);
 };
 
 class FabricOrchDaemon : public OrchDaemon

tests/test_zmq.py

@@ -120,3 +120,37 @@ def test_vrf(self, dvs):
         dvs.runcmd("cp /usr/bin/orchagent.sh_vrf_ut_backup /usr/bin/orchagent.sh")
         dvs.stop_swss()
         dvs.start_swss()
+
+    def test_heartbeat(self, dvs):
+        # Improve test code coverage, change orchagent to disable heartbeat
+        dvs.runcmd("cp /usr/bin/orchagent.sh /usr/bin/orchagent.sh_hb_ut_backup")
+        dvs.runcmd("sed -i.bak 's/\/usr\/bin\/orchagent /\/usr\/bin\/orchagent -I 0 /g' /usr/bin/orchagent.sh")
+        dvs.stop_swss()
+        dvs.start_swss()
+
+        # wait orchagent start
+        time.sleep(3)
+        process_statue = dvs.runcmd("ps -ef")
+        zmq_logger.debug("Process status: {}".format(process_statue))
+
+        # revert change
+        dvs.runcmd("cp /usr/bin/orchagent.sh_hb_ut_backup /usr/bin/orchagent.sh")
+        dvs.stop_swss()
+        dvs.start_swss()
+
+    def test_usage(self, dvs):
+        # Improve test code coverage, change orchagent to display usage
+        dvs.runcmd("cp /usr/bin/orchagent.sh /usr/bin/orchagent.sh_usage_ut_backup")
+        dvs.runcmd("sed -i.bak 's/\/usr\/bin\/orchagent /\/usr\/bin\/orchagent -h /g' /usr/bin/orchagent.sh")
+        dvs.stop_swss()
+        dvs.start_swss()
+
+        # wait orchagent start
+        time.sleep(3)
+        process_statue = dvs.runcmd("ps -ef")
+        zmq_logger.debug("Process status: {}".format(process_statue))
+
+        # revert change
+        dvs.runcmd("cp /usr/bin/orchagent.sh_usage_ut_backup /usr/bin/orchagent.sh")
+        dvs.stop_swss()
+        dvs.start_swss()