[system-health] Improve code structure of system health CLIs (sonic-net#2453)

Author: Junchao-Mellanox

show/system_health.py

@@ -5,144 +5,82 @@
 from tabulate import tabulate
 import utilities_common.cli as clicommon
 
-#
-# 'system-health' command ("show system-health")
-#
-@click.group(name='system-health', cls=clicommon.AliasedGroup)
-def system_health():
-    """Show system-health information"""
-    return
 
-@system_health.command()
-def summary():
-    """Show system-health summary information"""
-    # Mock the redis for unit test purposes #
-    try:
-        if os.environ["UTILITIES_UNIT_TESTING"] == "1":
-            modules_path = os.path.join(os.path.dirname(__file__), "..")
-            sys.path.insert(0, modules_path)
-            from tests.system_health_test import MockerManager
-            from tests.system_health_test import MockerChassis
-            HealthCheckerManager = MockerManager
-            Chassis = MockerChassis
-    except Exception:
-        # Normal run... #
+def get_system_health_status():
+    if os.environ["UTILITIES_UNIT_TESTING"] == "1":
+        modules_path = os.path.join(os.path.dirname(__file__), "..")
+        sys.path.insert(0, modules_path)
+        from tests.system_health_test import MockerManager
+        from tests.system_health_test import MockerChassis
+        HealthCheckerManager = MockerManager
+        Chassis = MockerChassis
+    else:
         if os.geteuid():
             click.echo("Root privileges are required for this operation")
-            return
+            exit(1)
         from health_checker.manager import HealthCheckerManager
         from sonic_platform.chassis import Chassis
 
+
     manager = HealthCheckerManager()
     if not manager.config.config_file_exists():
         click.echo("System health configuration file not found, exit...")
-        return
+        exit(1)
+
     chassis = Chassis()
     stat = manager.check(chassis)
     chassis.initizalize_system_led()
-    led = chassis.get_status_led()
-    click.echo("System status summary\n\n  System status LED  " + led)
-    services_list = []
-    fs_list = []
-    device_list =[]
-    for category, elements in stat.items():
-        for element in elements:
-            if elements[element]['status'] != "OK":
-                if 'Running' in elements[element]['message']:
-                    services_list.append(element)
-                elif 'Accessible' in elements[element]['message']:
-                    fs_list.append(element)
-                else:
-                    device_list.append(elements[element]['message'])
-    if len(services_list) or len(fs_list):
-        click.echo("  Services:\n    Status: Not OK")
-    else:
-        click.echo("  Services:\n    Status: OK")
-    if len(services_list):
-        services_list_string = str(services_list)
-        click.echo("    Not Running: " + services_list_string.replace("[", "").replace(']', ""))
-    if len(fs_list):
-        fs_list_string = str(fs_list)
-        click.echo("    Not Accessible: " + fs_list_string.replace("[", "").replace(']', ""))
-    if len(device_list):
-        click.echo("  Hardware:\n    Status: Not OK")
-        click.echo("    Reasons: " + device_list.pop())
-        while len(device_list):
-            click.echo("\t     " + device_list.pop())
-    else:
-        click.echo("  Hardware:\n    Status: OK")
 
-@system_health.command()
-def detail():
-    """Show system-health detail information"""
-    # Mock the redis for unit test purposes #
-    try:
-        if os.environ["UTILITIES_UNIT_TESTING"] == "1":
-            modules_path = os.path.join(os.path.dirname(__file__), "..")
-            sys.path.insert(0, modules_path)
-            from tests.system_health_test import MockerManager
-            from tests.system_health_test import MockerChassis
-            HealthCheckerManager = MockerManager
-            Chassis = MockerChassis
-    except Exception:
-        # Normal run... #
-        if os.geteuid():
-            click.echo("Root privileges are required for this operation")
-            return
-        from health_checker.manager import HealthCheckerManager
-        from sonic_platform.chassis import Chassis
+    return manager, chassis, stat
 
-    manager = HealthCheckerManager()
-    if not manager.config.config_file_exists():
-        click.echo("System health configuration file not found, exit...")
-        return
-    chassis = Chassis()
-    stat = manager.check(chassis)
-    #summary output
-    chassis.initizalize_system_led()
-    led = chassis.get_status_led()
+def display_system_health_summary(stat, led):
     click.echo("System status summary\n\n  System status LED  " + led)
     services_list = []
     fs_list = []
     device_list =[]
     for category, elements in stat.items():
         for element in elements:
             if elements[element]['status'] != "OK":
-                if 'Running' in elements[element]['message']:
-                    services_list.append(element)
-                elif 'Accessible' in elements[element]['message']:
-                    fs_list.append(element)
+                if category == 'Services':
+                    if 'Accessible' in elements[element]['message']:
+                        fs_list.append(element)
+                    else:
+                        services_list.append(element)
                 else:
                     device_list.append(elements[element]['message'])
-    if len(services_list) or len(fs_list):
+    if services_list or fs_list:
         click.echo("  Services:\n    Status: Not OK")
     else:
         click.echo("  Services:\n    Status: OK")
-    if len(services_list):
-        services_list_string = str(services_list)
-        click.echo("    Not Running: " + services_list_string.replace("[", "").replace(']', ""))
-    if len(fs_list):
-        fs_list_string = str(fs_list)
-        click.echo("    Not Accessible: " + fs_list_string.replace("[", "").replace(']', ""))
-    if len(device_list):
+    if services_list:
+        click.echo("    Not Running: " + ', '.join(services_list))
+    if fs_list:
+        click.echo("    Not Accessible: " + ', '.join(fs_list))
+    if device_list:
         click.echo("  Hardware:\n    Status: Not OK")
-        click.echo("    Reasons: " + device_list.pop())
-        while len(device_list):
-            click.echo("\t     " + device_list.pop())
+        device_list.reverse()
+        click.echo("    Reasons: " + device_list[0])
+        if len(device_list) > 1:
+            click.echo('\n'.join(("\t     " + x) for x in device_list[1:]))
     else:
         click.echo("  Hardware:\n    Status: OK")
 
+def display_monitor_list(stat):
     click.echo('\nSystem services and devices monitor list\n')
     header = ['Name', 'Status', 'Type']
     table = []
-    for category, elements in stat.items():
+    for elements in stat.values():
         for element in sorted(elements.items(), key=lambda x: x[1]['status']):
             entry = []
             entry.append(element[0])
             entry.append(element[1]['status'])
             entry.append(element[1]['type'])
             table.append(entry)
     click.echo(tabulate(table, header))
+
+
+def display_ignore_list(manager):
+    header = ['Name', 'Status', 'Type']
     click.echo('\nSystem services and devices ignore list\n')
     table = []
     if manager.config.ignore_services:
@@ -161,43 +99,35 @@ def detail():
             table.append(entry)
     click.echo(tabulate(table, header))
 
+#
+# 'system-health' command ("show system-health")
+#
+@click.group(name='system-health', cls=clicommon.AliasedGroup)
+def system_health():
+    """Show system-health information"""
+    return
+
+@system_health.command()
+def summary():
+    """Show system-health summary information"""
+    _, chassis, stat = get_system_health_status()
+    display_system_health_summary(stat, chassis.get_status_led())
+
+
+@system_health.command()
+def detail():
+    """Show system-health detail information"""
+    manager, chassis, stat = get_system_health_status()
+    display_system_health_summary(stat, chassis.get_status_led())
+    display_monitor_list(stat)
+    display_ignore_list(manager)
+
+
 @system_health.command()
 def monitor_list():
     """Show system-health monitored services and devices name list"""
-    # Mock the redis for unit test purposes #
-    try:
-        if os.environ["UTILITIES_UNIT_TESTING"] == "1":
-            modules_path = os.path.join(os.path.dirname(__file__), "..")
-            sys.path.insert(0, modules_path)
-            from tests.system_health_test import MockerManager
-            from tests.system_health_test import MockerChassis
-            HealthCheckerManager = MockerManager
-            Chassis = MockerChassis
-    except Exception:
-        # Normal run... #
-        if os.geteuid():
-            click.echo("Root privileges are required for this operation")
-            return
-        from health_checker.manager import HealthCheckerManager
-        from sonic_platform.chassis import Chassis
-
-    manager = HealthCheckerManager()
-    if not manager.config.config_file_exists():
-        click.echo("System health configuration file not found, exit...")
-        return
-    chassis = Chassis()
-    stat = manager.check(chassis)
-    click.echo('\nSystem services and devices monitor list\n')
-    header = ['Name', 'Status', 'Type']
-    table = []
-    for category, elements in stat.items():
-        for element in sorted(elements.items(), key=lambda x: x[1]['status']):
-            entry = []
-            entry.append(element[0])
-            entry.append(element[1]['status'])
-            entry.append(element[1]['type'])
-            table.append(entry)
-    click.echo(tabulate(table, header))
+    _, _, stat = get_system_health_status()
+    display_monitor_list(stat)
 
 
 @system_health.group('sysready-status',invoke_without_command=True)

tests/system_health_test.py

@@ -84,7 +84,7 @@ def test_health_summary(self):
   System status LED  red
   Services:
     Status: Not OK
-    Not Running: 'telemetry', 'snmp_subagent'
+    Not Running: telemetry, snmp_subagent
   Hardware:
     Status: OK
 """
@@ -171,7 +171,7 @@ def test_health_detail(self):
   System status LED  red
   Services:
     Status: Not OK
-    Not Running: 'telemetry', 'sflowmgrd'
+    Not Running: telemetry, sflowmgrd
   Hardware:
     Status: Not OK
     Reasons: Failed to get voltage minimum threshold data for PSU 1
@@ -243,7 +243,7 @@ def test_health_detail(self):
   System status LED  red
   Services:
     Status: Not OK
-    Not Running: 'telemetry', 'sflowmgrd'
+    Not Running: telemetry, sflowmgrd
   Hardware:
     Status: OK