Publish additional events

files/build_templates/dhcp_relay_regex.json

@@ -3,5 +3,10 @@
         "tag": "dhcp-relay-discard",
         "regex": "Discarding packet received on ([a-zA-Z0-9-_]*) interface that has no IPv4 address assigned.",
         "params": [ "ifname" ]
+    },
+    {
+        "tag": "dhcp-relay-bind-failure",
+        "regex": "Failed to bind socket to (link local|global) ipv6 address on interface ([a-zA-Z0-9]*)",
+        "params": [ "type:ret=(arg==\"link local\")and\"local\"or\"global\")", "vlan" ]
     }
 ]

files/image_config/monit/memory_checker

@@ -27,6 +27,10 @@ import re
 
 import docker
 
+from swsscommon import swsscommon
+
+EVENTS_PUBLISHER_SOURCE = "sonic-events-host"
+EVENTS_PUBLISHER_TAG = "mem-threshold-exceeded"
 
 def get_command_result(command):
     """Executes the command and return the resulting output.
@@ -54,6 +58,14 @@ def get_command_result(command):
 
     return command_stdout.strip()
 
+def publish_events(container_name, mem_usage_bytes, threshold_value):
+    events_handle = swsscommon.events_init_publisher(EVENTS_PUBLISHER_SOURCE)
+    params = swsscommon.FieldValueMap()
+    params["ctr_name"] = container_name
+    params["mem_usage"] = mem_usage_bytes
+    params["threshold"] = threshold_value
+    swsscommon.event_publish(events_handle, EVENTS_PUBLISHER_TAG, params)
+    swsscommon.events_deinit_publisher(events_handle)
 
 def check_memory_usage(container_name, threshold_value):
     """Checks the memory usage of a container and writes an alerting messages into
@@ -89,6 +101,8 @@ def check_memory_usage(container_name, threshold_value):
                   .format(container_name, mem_usage_bytes, threshold_value))
             syslog.syslog(syslog.LOG_INFO, "[{}]: Memory usage ({} Bytes) is larger than the threshold ({} Bytes)!"
                           .format(container_name, mem_usage_bytes, threshold_value))
+            # publish event
+            publish_events(container_name, mem_usage_bytes, threshold_value)
             sys.exit(3)
     else:
         syslog.syslog(syslog.LOG_ERR, "[memory_checker] Failed to retrieve memory value from '{}'"

files/scripts/supervisor-proc-exit-listener

@@ -31,6 +31,8 @@ SELECT_TIMEOUT_SECS = 1.0
 # Alerting message will be written into syslog in the following interval
 ALERTING_INTERVAL_SECS = 60
 
+EVENTS_PUBLISHER_SOURCE = "sonic-events-host"
+EVENTS_PUBLISHER_TAG = "process-exited-unexpectedly"
 
 def get_critical_group_and_process_list():
     """
@@ -106,6 +108,13 @@ def get_autorestart_state(container_name):
 
     return is_auto_restart
 
+def publish_events(process_name, container_name):
+    events_handle = swsscommon.events_init_publisher(EVENTS_PUBLISHER_SOURCE)
+    params = swsscommon.FieldValueMap()
+    params["process_name"] = process_name
+    params["ctr_name"] = container_name
+    swsscommon.event_publish(events_handle, EVENTS_PUBLISHER_TAG, params)
+    swsscommon.events_deinit_publisher(events_handle)
 
 def main(argv):
     container_name = None
@@ -145,6 +154,7 @@ def main(argv):
                         MSG_FORMAT_STR = "Process '{}' exited unexpectedly. Terminating supervisor '{}'"
                         msg = MSG_FORMAT_STR.format(payload_headers['processname'], container_name)
                         syslog.syslog(syslog.LOG_INFO, msg)
+                        publish_events(payload_headers['processname'], container_name)
                         os.kill(os.getppid(), signal.SIGTERM)
                     else:
                         process_under_alerting[process_name]["last_alerted"] = time.time()

src/system-health/health_checker/service_checker.py

@@ -12,6 +12,8 @@
 SYSLOG_IDENTIFIER = 'service_checker'
 logger = Logger(log_identifier=SYSLOG_IDENTIFIER)
 
+EVENTS_PUBLISHER_SOURCE = "sonic-events-host"
+EVENTS_PUBLISHER_TAG = "process-not-running"
 
 class ServiceChecker(HealthChecker):
     """
@@ -308,6 +310,15 @@ def _parse_supervisorctl_status(self, process_status):
             data[items[0].strip()] = items[1].strip()
         return data
 
+    def publish_events(self, container_name, critical_process_list):
+        events_handle = swsscommon.events_init_publisher(EVENTS_PUBLISHER_SOURCE)
+        params = swsscommon.FieldValueMap()
+        params["ctr_name"] = container_name
+        for process_name in critical_process_list:
+            params["process_name"] = process_name
+            swsscommon.event_publish(events_handle, EVENTS_PUBLISHER_TAG, params)
+        swsscommon.events_deinit_publisher(events_handle)
+
     def check_process_existence(self, container_name, critical_process_list, config, feature_table):
         """Check whether the process in the specified container is running or not.
 
@@ -332,6 +343,7 @@ def check_process_existence(self, container_name, critical_process_list, config,
                 if process_status is None:
                     for process_name in critical_process_list:
                         self.set_object_not_ok('Process', '{}:{}'.format(container_name, process_name), "'{}' is not running".format(process_name))
+                    publish_events(container_name, critical_process_list)
                     return
 
                 process_status = self._parse_supervisorctl_status(process_status.strip().splitlines())