[show][config] Add CLI support for configurable drop monitor feature

*   Implements commands to configure persistent drop monitor thresholds.
*   Adds support for setting monitoring window size, drop count threshold, and incident count threshold.
*   Includes `show dropcounters monitor` command to display current configuration.
*   Provides `config dropcounters monitor` command to enable/disable and configure the monitor.
*   Adds unit tests to cover the new CLI functionality.

config/main.py

@@ -7391,6 +7391,28 @@ def install(counter_name, alias, group, counter_type, desc, reasons, verbose):
 
     clicommon.run_command(command, display_cmd=verbose)
 
+#
+# 'monitor' subcommand ('config dropcounters monitor')
+#
+@dropcounters.command()
+@click.option("-s", "--status", type=str, help="Status can be set to enabled/disabled")
+@click.option("-w", "--window", type=int, help="Window size in seconds")
+@click.option("-dct", "--drop-count-threshold",  type=int, help="Minimum threshold for drop counts to be classified as an incident")
+@click.option('-ict', '--incident-count-threshold', type=int, help="Minimum number of incidents to trigger a syslog entry")
+@click.option('-v', '--verbose', is_flag=True, help="Enable verbose output")
+def monitor(status, window, drop_count_threshold, incident_count_threshold, verbose):
+    """Update configurations of drop counter monitor"""
+    command = ['dropconfig', '-c', 'config_monitor']
+    if status:
+        command += ['-s', str(status)]
+    if window:
+        command += ['-w', str(window)]
+    if drop_count_threshold:
+        command += ['-dct', str(drop_count_threshold)]
+    if incident_count_threshold:
+        command += ['-ict', str(incident_count_threshold)]
+
+    clicommon.run_command(command, display_cmd=verbose)
 
 #
 # 'delete' subcommand ('config dropcounters delete')

scripts/dropconfig

@@ -33,6 +33,7 @@ from swsscommon.swsscommon import SonicV2Connector, ConfigDBConnector
 # CONFIG_DB Tables
 DEBUG_COUNTER_CONFIG_TABLE = 'DEBUG_COUNTER'
 DROP_REASON_CONFIG_TABLE = 'DEBUG_COUNTER_DROP_REASON'
+DEBUG_DROP_MONITOR_TABLE = 'DEBUG_DROP_MONITOR'
 
 # STATE_DB Tables
 DEBUG_COUNTER_CAPABILITY_TABLE = 'DEBUG_COUNTER_CAPABILITIES'
@@ -60,6 +61,59 @@ class DropConfig(object):
         self.state_db = SonicV2Connector(use_unix_socket_path=False)
         self.state_db.connect(self.state_db.STATE_DB)
 
+    def show_drop_monitor(self):
+        """
+           The function to show the current configuration of DEBUG_DROP_MONITOR table in CONFIG_DB
+        """
+        current_config = self.config_db.get_entry(DEBUG_DROP_MONITOR_TABLE, "CONFIG")
+        if current_config == {}:
+            print("Debug drop monitoring has not been configured. Use 'config dropcounters monitor' CLI to configure it.")
+            return
+
+        print("Current configuration of debug drop monitor")
+        print("The status: ", current_config.get("status", None))
+        print("The window size: ", current_config.get("window", None))
+        print("The drop_count_threshold: ", current_config.get("drop_count_threshold", None))
+        print("The incident_count_threshold: ", current_config.get("incident_count_threshold", None))
+
+    def configure_drop_monitor(self, status, window, drop_count_threshold, incident_count_threshold):
+        """
+            The function to configure Debug Drop Monitor feature.
+        """
+        config = self.config_db.get_entry(DEBUG_DROP_MONITOR_TABLE, "CONFIG")
+
+        if status:
+            if status not in ['enabled', 'disabled']:
+                raise InvalidArgumentError('Invalid status: {}, expected: enabled/disabled'.format(status))
+            config["status"] = status
+        elif "status" not in config.keys():
+            config["status"] = "disabled"
+
+        if window:
+            if window <= 0:
+                raise InvalidArgumentError('Invalid window size. Window size should be positive, received: {}'.format(window))
+            config["window"] = str(window)
+        elif "window" not in config.keys():
+            config["window"] = "900"
+
+        if drop_count_threshold:
+            if drop_count_threshold < 0:
+                raise InvalidArgumentError('Invalid drop count threshold. Drop count threshold should be positive, received: {}'.format(drop_count_threshold))
+            config["drop_count_threshold"] = str(drop_count_threshold)
+        elif "drop_count_threshold" not in config.keys():
+            config["drop_count_threshold"] = "100"
+
+        if incident_count_threshold:
+            if incident_count_threshold < 0:
+                raise InvalidArgumentError('Invalid incident count threshold. Incident count threshold should be positive, received: {}'.format(incident_count_threshold))
+            config["incident_count_threshold"] = str(incident_count_threshold)
+        elif "incident_count_threshold" not in config.keys():
+            config["incident_count_threshold"] = "2"
+
+        # Update the DEBUG_DROP_MONITOR table
+        self.config_db.set_entry(DEBUG_DROP_MONITOR_TABLE, ("CONFIG"), config)
+        print(f"Successfully updated the DEBUG_DROP_MONITOR config")
+
     # -c show_config
     def print_counter_config(self, group):
         """
@@ -349,6 +403,10 @@ Examples:
     parser.add_argument('-t', '--type',    type=str, help='The type of the target drop counter',                   default=None)
     parser.add_argument('-d', '--desc',    type=str, help='The description for the target drop counter',           default=None)
     parser.add_argument('-r', '--reasons', type=str, help='The list of drop reasons for the target drop counter',  default=None)
+    parser.add_argument('-s', '--status', type=str, help='To set the status of drop counter monitoring feature (enabled/disabled)',  default=None)
+    parser.add_argument('-w', '--window', type=int, help='The window for drop counter monitor in seconds',  default=None)
+    parser.add_argument('-dct', '--drop-count-threshold', type=int, help='The minimum drops needed to classify a drop as an incident by the drop count monitor',  default=None)
+    parser.add_argument('-ict', '--incident-count-threshold', type=int, help='The minimum number of incidents needed to register a syslog by drop count monitor',  default=None)
 
     args = parser.parse_args()
 
@@ -361,6 +419,13 @@ Examples:
     description = args.desc
     drop_reasons = args.reasons
 
+    # These arguments are used to control the Drop Counter Monitor feature
+    # This is an optional feature and more details on its usage can be found at: https://github.com/sonic-net/SONiC/blob/master/doc/drop_counters/drop_counters_HLD.md
+    status = args.status
+    window = args.window
+    drop_count_threshold = args.drop_count_threshold
+    incident_count_threshold = args.incident_count_threshold
+
     reasons = deserialize_reason_list(drop_reasons)
 
     dconfig = DropConfig()
@@ -398,6 +463,14 @@ Examples:
         dconfig.print_counter_config(group)
     elif command == 'show_capabilities':
         dconfig.print_device_capabilities()
+    elif command == 'config_monitor':
+        try:
+            dconfig.configure_drop_monitor(status, window, drop_count_threshold, incident_count_threshold)
+        except InvalidArgumentError as err:
+            print('Encountered error trying to configure drop monitor: {}'.format(err.message))
+            sys.exit(1)
+    elif command == 'show_monitor':
+        dconfig.show_drop_monitor()
     else:
         print("Command not recognized")
 

show/dropcounters.py

@@ -36,6 +36,14 @@ def capabilities(verbose):
 
     clicommon.run_command(cmd, display_cmd=verbose)
 
+# 'monitor' subcommand ("show dropcounters monitor")
+@dropcounters.command()
+@click.option('--verbose', is_flag=True, help="Enable verbose output")
+def monitor(verbose):
+    """Show device drop counter capabilities"""
+    cmd = ['dropconfig', '-c', 'show_monitor']
+
+    clicommon.run_command(cmd, display_cmd=verbose)
 
 # 'counts' subcommand ("show dropcounters counts")
 @dropcounters.command()

tests/config_test.py

@@ -3179,6 +3179,19 @@ def test_remove_reasons(self, mock_run_command):
         assert result.exit_code == 0
         mock_run_command.assert_called_once_with(['dropconfig', '-c', 'remove', '-n', str(counter_name), '-r', str(reasons)], display_cmd=True)
 
+    @patch('utilities_common.cli.run_command')
+    def test_configure_drop_monitor(self, mock_run_command):
+        status = 'enabled'
+        window = '600'
+        drop_count_threshold = '5'
+        incident_count_threshold = '3'
+        runner = CliRunner()
+        result = runner.invoke(config.config.commands['dropcounters'].commands['monitor'], ['-s', status, '-w', window, '-dct', drop_count_threshold, '-ict', incident_count_threshold, '-v'])
+        print(result.exit_code)
+        print(result.output)
+        assert result.exit_code == 0
+        mock_run_command.assert_called_once_with(['dropconfig', '-c', 'config_monitor', '-s', status, '-w', window, '-dct', drop_count_threshold, '-ict', incident_count_threshold], display_cmd=True)
+
     def teardown(self):
         print("TEARDOWN")
 

tests/dropconfig_test.py

@@ -46,6 +46,38 @@ def test_remove_error(self, mock_print):
         mock_print.assert_called_once_with('Encountered error trying to remove reasons: No counter name provided')
         assert e.value.code == 1
 
+    @patch('builtins.print')
+    @patch('sys.argv', ['dropconfig', '-c', 'config_monitor', '-s', 'off'])
+    def test_config_monitor_status_error(self, mock_print):
+        with pytest.raises(SystemExit) as e:
+            dropconfig.main()
+        mock_print.assert_called_once_with('Encountered error trying to configure drop monitor: Invalid status: off, expected: enabled/disabled')
+        assert e.value.code == 1
+
+    @patch('builtins.print')
+    @patch('sys.argv', ['dropconfig', '-c', 'config_monitor', '-w', '-1'])
+    def test_config_monitor_window_error(self, mock_print):
+        with pytest.raises(SystemExit) as e:
+            dropconfig.main()
+        mock_print.assert_called_once_with('Encountered error trying to configure drop monitor: Invalid window size. Window size should be positive, received: -1')
+        assert e.value.code == 1
+
+    @patch('builtins.print')
+    @patch('sys.argv', ['dropconfig', '-c', 'config_monitor', '-dct', '-1'])
+    def test_config_monitor_dct_error(self, mock_print):
+        with pytest.raises(SystemExit) as e:
+            dropconfig.main()
+        mock_print.assert_called_once_with('Encountered error trying to configure drop monitor: Invalid drop count threshold. Drop count threshold should be positive, received: -1')
+        assert e.value.code == 1
+
+    @patch('builtins.print')
+    @patch('sys.argv', ['dropconfig', '-c', 'config_monitor', '-ict', '-1'])
+    def test_config_monitor_ict_error(self, mock_print):
+        with pytest.raises(SystemExit) as e:
+            dropconfig.main()
+        mock_print.assert_called_once_with('Encountered error trying to configure drop monitor: Invalid incident count threshold. Incident count threshold should be positive, received: -1')
+        assert e.value.code == 1
+
     def teardown(self):
         print('TEARDOWN')
 

tests/drops_group_test.py

@@ -98,6 +98,13 @@
 sonic_drops_test               0                    0
 """
 
+expected_drop_monitor_config = """\
+Current configuration of debug drop monitor
+The status:  disabled
+The window size:  900
+The drop_count_threshold:  100
+The incident_count_threshold:  2
+"""
 
 def remove_tmp_dropstat_file():
     # remove the tmp portstat
@@ -163,6 +170,12 @@ def test_show_counts_with_clear(self):
         print(result.output)
         assert result.output == expected_counts_with_clear
 
+    def test_show_drop_monitor_config(self):
+        runner = CliRunner()
+        result = runner.invoke(show.cli.commands["dropcounters"].commands["monitor"], [])
+        print(result.output)
+        assert result.output == expected_drop_monitor_config
+
     @classmethod
     def teardown_class(cls):
         print("TEARDOWN")

tests/mock_tables/config_db.json

@@ -2837,5 +2837,11 @@
         "hello_time": "2",
         "max_age": "20",
         "priority": "32768"
+     },
+     "DEBUG_DROP_MONITOR|CONFIG": {
+        "status": "disabled",
+        "window": 900,
+        "drop_count_threshold": 100,
+        "incident_count_threshold": 2
      }
 }