container_checker on supervisor should check containers based on asic presence

files/image_config/monit/container_checker

@@ -20,9 +20,34 @@ import docker
 import sys
 
 import swsssdk
-from sonic_py_common import multi_asic, device_info
+from sonic_py_common import multi_asic, device_info, daemon_base
 from swsscommon import swsscommon
 
+def get_asic_presence_list():
+    """
+    @summary: This function will get the asic presence list. On Supervisor, the list includes only the asics
+              for inserted and detected fabric cards. For non-supervisor cards, e.g. line card, the list should
+              contain all supported asics by the card. The function gets the asic list from CHASSIS_ASIC_TABLE from
+              CHASSIS_STATE_DB. The function assumes that the first N asic ids (asic0 to asic(N-1)) in
+              CHASSIS_ASIC_TABLE belongs to the supervisor, where N is the max number of asics supported by the Chassis 
+    @return:  List of asics present
+    """
+    asics_list = []
+    if multi_asic.is_multi_asic():
+        if not device_info.is_supervisor():
+            # Supervisor has FRU Fabric cards. If not supervisor, all asics
+            # should be present. Add all asics, 0 - num_asics to the list.
+            asics_list = list(range(0,multi_asic.get_num_asics()))
+        else:
+            # Get asic list from CHASSIS_ASIC_TABLE
+            chassis_state_db = daemon_base.db_connect("CHASSIS_STATE_DB")
+            asic_table = swsscommon.Table(chassis_state_db, 'CHASSIS_ASIC_TABLE')
+            if asic_table:
+                asics_presence_list = list(asic_table.getKeys())
+                for asic in asics_presence_list:
+                    # asic is asid id: asic0, asic1.... asicN. Get the numeric value.
+                    asics_list.append(int(asic[4:]))
+    return asics_list
 
 def get_expected_running_containers():
     """
@@ -41,7 +66,15 @@ def get_expected_running_containers():
 
     expected_running_containers = set()
     always_running_containers = set()
-
+
+    # Get current asic presence list. For multi_asic system, multi instance containers
+    # should be checked only for asics present.
+    asics_id_presence = get_asic_presence_list()
+
+    # Some services, like database and bgp run all the instances irrespective of asic presence.
+    # Add those to exception list.
+    run_all_instance_list = ['database', 'bgp']
+
     for container_name in feature_table.keys():
         if feature_table[container_name]["state"] not in ["disabled", "always_disabled"]:
             if multi_asic.is_multi_asic():
@@ -50,7 +83,8 @@ def get_expected_running_containers():
                 if feature_table[container_name]["has_per_asic_scope"] == "True":
                     num_asics = multi_asic.get_num_asics()
                     for asic_id in range(num_asics):
-                        expected_running_containers.add(container_name + str(asic_id))
+                        if asic_id in asics_id_presence or container_name in run_all_instance_list:
+                            expected_running_containers.add(container_name + str(asic_id))
             else:
                 expected_running_containers.add(container_name)
         if feature_table[container_name]["state"] == 'always_enabled':
@@ -60,9 +94,11 @@ def get_expected_running_containers():
                 if feature_table[container_name]["has_per_asic_scope"] == "True":
                     num_asics = multi_asic.get_num_asics()
                     for asic_id in range(num_asics):
-                        always_running_containers.add(container_name + str(asic_id))
+                        if asic_id in asics_id_presence or container_name in run_all_instance_list:
+                            always_running_containers.add(container_name + str(asic_id))
             else:
                 always_running_containers.add(container_name)
+
     if device_info.is_supervisor():
         always_running_containers.add("database-chassis")
     return expected_running_containers, always_running_containers