Add Orchagent heartbeat config

Author: liuh-80

dockers/docker-orchagent/orchagent.sh

@@ -111,4 +111,10 @@ if [[ x"${MGMT_VRF_ENABLED}" == x"true" ]]; then
     ORCHAGENT_ARGS+=" -v mgmt"
 fi
 
+# Add heartbeat interval when enabled
+HEARTBEAT_INTERVAL=`sonic-db-cli CONFIG_DB hget  "HEARTBEAT|orchagent" "heartbeat_interval"`
+if [ ! -z "$HEARTBEAT_INTERVAL" ] && [ $HEARTBEAT_INTERVAL != "null" ]; then
+    ORCHAGENT_ARGS+=" -I $HEARTBEAT_INTERVAL"
+fi
+
 exec /usr/bin/orchagent ${ORCHAGENT_ARGS}

files/scripts/supervisor-proc-exit-listener

@@ -31,6 +31,9 @@ CRITICAL_PROCESSES_FILE = '/etc/supervisor/critical_processes'
 # The FEATURE table in config db contains auto-restart field
 FEATURE_TABLE_NAME = 'FEATURE'
 
+# The HEARTBEAT table in config db contains heart beat config
+HEARTBEAT_TABLE_NAME = 'HEARTBEAT'
+
 # Value of parameter 'timeout' in select(...) method
 SELECT_TIMEOUT_SECS = 1.0
 
@@ -40,6 +43,8 @@ ALERTING_INTERVAL_SECS = 60
 EVENTS_PUBLISHER_SOURCE = "sonic-events-host"
 EVENTS_PUBLISHER_TAG = "process-exited-unexpectedly"
 
+heartbeat_alert_interval_mapping = defaultdict(dict)
+
 def get_group_and_process_list(process_file):
     """
     @summary: Read the critical processes/group names.
@@ -114,6 +119,21 @@ def get_autorestart_state(container_name, use_unix_socket_path):
 
     return is_auto_restart
 
+def load_heartbeat_alert_interval(use_unix_socket_path):
+    config_db = swsscommon.ConfigDBConnector(use_unix_socket_path=use_unix_socket_path)
+    config_db.connect()
+    heartbeat_table = config_db.get_table(HEARTBEAT_TABLE_NAME)
+    if heartbeat_table:
+        heartbeat_table_keys = heartbeat_table.keys()
+        for process in heartbeat_table_keys:
+            heartbeat_alert_interval_mapping[process] = heartbeat_table[process].get('alert_interval') / 1000
+
+def get_heartbeat_alert_interval(process):
+    if process in heartbeat_alert_interval_mapping:
+        return heartbeat_alert_interval_mapping[process]
+
+    return ALERTING_INTERVAL_SECS
+
 def publish_events(events_handle, process_name, container_name):
     params = swsscommon.FieldValueMap()
     params["process_name"] = process_name
@@ -136,6 +156,8 @@ def main(argv):
 
     critical_group_list, critical_process_list = get_group_and_process_list(CRITICAL_PROCESSES_FILE)
 
+    load_heartbeat_alert_interval(use_unix_socket_path)
+
     # WATCH_PROCESSES_FILE is optional
     watch_process_list = []
     if os.path.exists(WATCH_PROCESSES_FILE):
@@ -211,7 +233,7 @@ def main(argv):
         for process in process_heart_beat_info.keys():
             epoch_time = time.time()
             elapsed_secs = epoch_time - process_heart_beat_info[process]["last_heart_beat"]
-            if elapsed_secs >= ALERTING_INTERVAL_SECS:
+            if elapsed_secs >= get_heartbeat_alert_interval(process):
                 elapsed_mins = elapsed_secs // 60
                 generate_alerting_message(process, "stuck", elapsed_mins, syslog.LOG_WARNING)
 

src/sonic-yang-models/yang-models/sonic-heartbeat.yang

@@ -0,0 +1,53 @@
+module sonic-heartbeat {
+
+    yang-version 1.1;
+
+    namespace "http://github.com/sonic-net/sonic-heartbeat";
+    prefix heartbeat;
+
+    import ietf-inet-types {
+        prefix inet;
+    }
+
+    organization
+        "SONiC";
+
+    contact
+        "SONiC";
+
+    description "HEARTBEAT YANG Module for SONiC OS";
+
+    revision 2025-01-09 {
+        description "First Revision";
+    }
+
+    container sonic-heartbeat {
+
+        container HEARTBEAT {
+            description "HEARTBEAT config TABLE part of config_db.json";
+
+            list HEARTBEAT_LIST {
+                key "name";
+
+                leaf name {
+                    description "process name in HEARTBEAT table";
+                    type string {
+                        length 1..32;
+                    }
+                }
+
+                leaf heartbeat_interval {
+                    description "Heartbeat interval in millisecond";
+                    type uint32;
+                    default "100000";
+                }
+
+                leaf alert_interval {
+                    description "Alert interval in millisecond";
+                    type uint32;
+                    default "60000";
+                }
+            }
+        }
+    }
+}