17776 systemd_units: Add option to include disabled failed units in critical failed count for systemd summary

The summary check for systemd units now respects the parameter `Treat failed units that are disabled as critical`.
If set, failed disabled units are included in the critical count for failed services. Previously, disabled units were always excluded from the failed counts.

SUP-24357

Change-Id: I20563aa4a230f784e748240d9df73e7558aa49e8

Author: racicLuka

.werks/17776.md

@@ -0,0 +1,15 @@
+[//]: # (werk v2)
+# systemd_units: Add option to include disabled failed units in critical failed count for systemd summary
+
+key        | value
+---------- | ---
+date       | 2025-07-20T19:20:17+00:00
+version    | 2.4.0p9
+class      | feature
+edition    | cre
+component  | checks
+level      | 1
+compatible | yes
+
+The summary check for systemd units now respects the parameter `Treat failed units that are disabled as critical`.
+If set, failed disabled units are included in the critical count for failed services. Previously, disabled units were always excluded from the failed counts.

cmk/gui/plugins/wato/check_parameters/systemd_services_summary.py

@@ -12,7 +12,15 @@
     RulespecGroupCheckParametersApplications,
 )
 from cmk.gui.plugins.wato.utils.simple_levels import SimpleLevels
-from cmk.gui.valuespec import Age, Dictionary, ListOf, Migrate, MonitoringState, RegExp
+from cmk.gui.valuespec import (
+    Age,
+    Checkbox,
+    Dictionary,
+    ListOf,
+    Migrate,
+    MonitoringState,
+    RegExp,
+)
 
 REQUIRED_STATE_KEYS_AND_STATES = {"active": 0, "inactive": 0, "failed": 2}
 
@@ -31,6 +39,18 @@ def _parameter_valuespec_systemd_units_summary() -> Migrate:
     return Migrate(
         valuespec=Dictionary(
             elements=[
+                (
+                    "disabled_critical",
+                    Checkbox(
+                        default_value=True,
+                        label="Failed units that are disabled will be considered critical",
+                        title=_("Treat failed units that are disabled as critical"),
+                        help=_(
+                            "Failed service that are disabled are also considered for the status of the service. "
+                            "Activating this option will cause the service to be CRIT if a failed service is disabled."
+                        ),
+                    ),
+                ),
                 (
                     "states",
                     Dictionary(

cmk/plugins/collection/agent_based/systemd_units.py

@@ -768,7 +768,8 @@ def check_systemd_units_summary(
     sum_failed = sum(
         s.active_status == "failed"
         for s in units
-        if s not in services_organised["excluded"] and s not in services_organised["disabled"]
+        if s not in services_organised["excluded"]
+        and (params["disabled_critical"] is True or s not in services_organised["disabled"])
     )
 
     yield Result(
@@ -797,6 +798,7 @@ def check_systemd_units_summary(
 
 
 CHECK_DEFAULT_PARAMETERS_SUMMARY = {
+    "disabled_critical": False,
     "states": {
         "active": 0,
         "inactive": 0,

cmk/plugins/collection/checkman/systemd_units_services_summary

@@ -3,16 +3,15 @@ agents: linux
 catalog: os/ps
 license: GPLv2
 distribution: check_mk
-description:
- This check gives a summary of the linux services provided by
- the command 'systemctl --all'.
- The check reports the number of all services and for each service
- type, i.e. 'active', 'inactive', 'failed' and other ones.
- The state mapping is configurable, by default  active and inactive
- states are OK, failed and all others are CRIT.
+description: This check gives a summary of the linux services provided by
+  the command 'systemctl --all'.
+  The check reports the number of all services and for each service
+  type, i.e. 'active', 'inactive', 'failed' and other ones.
+  The state mapping is configurable, by default  active and inactive
+  states are OK, failed and all others are CRIT.
+  Note that by default, failed systemd services that are disabled do not result in CRIT.
+  This is configurable so that such services do trigger a CRIT.
 
-item:
- The fixed statement {{Summary}}
+item: The fixed statement {{Summary}}
 
-discovery:
- One summary service is created.
+discovery: One summary service is created.

tests/unit/cmk/plugins/collection/agent_based/test_systemd_units.py

@@ -1008,6 +1008,7 @@ def test_discover_systemd_units_services_summary(
                 "else": 2,
                 "states": {"active": 0, "failed": 2, "inactive": 0},
                 "states_default": 2,
+                "disabled_critical": False,
             },
             SECTION,
             [
@@ -1029,6 +1030,7 @@ def test_discover_systemd_units_services_summary(
                 "else": 2,
                 "states": {"active": 0, "failed": 2, "inactive": 0},
                 "states_default": 2,
+                "disabled_critical": False,
             },
             SECTION,
             [
@@ -1044,6 +1046,7 @@ def test_discover_systemd_units_services_summary(
                 "else": 2,
                 "states": {"active": 0, "failed": 2, "inactive": 0},
                 "states_default": 2,
+                "disabled_critical": False,
             },
             SECTION,
             [
@@ -1060,6 +1063,7 @@ def test_discover_systemd_units_services_summary(
                 "else": 2,
                 "states": {"active": 0, "failed": 2, "inactive": 0},
                 "states_default": 2,
+                "disabled_critical": False,
             },
             SECTION,
             [
@@ -1076,6 +1080,7 @@ def test_discover_systemd_units_services_summary(
                 "else": 2,
                 "states": {"active": 0, "failed": 2, "inactive": 0},
                 "states_default": 2,
+                "disabled_critical": False,
             },
             Section(services={}, sockets={}),
             [
@@ -1103,6 +1108,7 @@ def test_check_systemd_units_services(
                 "else": 2,
                 "states": {"active": 2, "failed": 1, "inactive": 0},
                 "states_default": 2,
+                "disabled_critical": False,
             },
             Section(
                 sockets={
@@ -1139,6 +1145,7 @@ def test_check_systemd_units_sockets(
                 "else": 2,
                 "states": {"active": 0, "failed": 2, "inactive": 0},
                 "states_default": 2,
+                "disabled_critical": False,
                 "ignored": [],
                 "activating_levels": (30, 60),
                 "deactivating_levels": (30, 60),
@@ -1158,6 +1165,7 @@ def test_check_systemd_units_sockets(
                 "else": 2,
                 "states": {"active": 0, "failed": 1, "inactive": 0},
                 "states_default": 2,
+                "disabled_critical": False,
                 "ignored": [],
                 "activating_levels": (30, 60),
                 "deactivating_levels": (30, 60),
@@ -1206,6 +1214,7 @@ def test_check_systemd_units_sockets(
                 "activating_levels": (30, 60),
                 "deactivating_levels": (30, 60),
                 "reloading_levels": (30, 60),
+                "disabled_critical": False,
             },
             Section(
                 sockets={},
@@ -1245,6 +1254,7 @@ def test_check_systemd_units_sockets(
                 "activating_levels": (30, 60),
                 "deactivating_levels": (30, 60),
                 "reloading_levels": (30, 60),
+                "disabled_critical": False,
             },
             Section(
                 sockets={},
@@ -1274,6 +1284,7 @@ def test_check_systemd_units_sockets(
                 "activating_levels": (30, 60),
                 "deactivating_levels": (30, 60),
                 "reloading_levels": (30, 60),
+                "disabled_critical": False,
             },
             Section(
                 sockets={},
@@ -1303,6 +1314,7 @@ def test_check_systemd_units_sockets(
                 "activating_levels": (30, 60),
                 "deactivating_levels": (30, 60),
                 "reloading_levels": (30, 60),
+                "disabled_critical": False,
             },
             Section(
                 sockets={},
@@ -1329,6 +1341,7 @@ def test_check_systemd_units_sockets(
                 "else": 2,
                 "states": {"active": 0, "failed": 2, "inactive": 0},
                 "states_default": 2,
+                "disabled_critical": False,
                 "ignored": [],
                 "activating_levels": (30, 60),
                 "deactivating_levels": (30, 60),
@@ -1360,6 +1373,7 @@ def test_check_systemd_units_sockets(
                 "else": 2,
                 "states": {"active": 0, "failed": 2, "inactive": 0},
                 "states_default": 2,
+                "disabled_critical": False,
                 "ignored": [
                     "systemd-timesyncd.service",
                     "systemd-ask-password-console",
@@ -1395,6 +1409,7 @@ def test_check_systemd_units_sockets(
                 "else": 2,
                 "states": {"active": 0, "failed": 2, "inactive": 0},
                 "states_default": 2,
+                "disabled_critical": False,
                 "ignored": [
                     "systemd.",
                 ],
@@ -1440,6 +1455,7 @@ def test_check_systemd_units_sockets(
                 "else": 2,
                 "states": {"active": 0, "failed": 2, "inactive": 0},
                 "states_default": 2,
+                "disabled_critical": False,
                 "ignored": [],
                 "activating_levels": (30, 60),
                 "deactivating_levels": (30, 60),
@@ -1465,11 +1481,43 @@ def test_check_systemd_units_sockets(
             ],
             id="One failed, but disabled",
         ),
+        pytest.param(
+            {
+                "disabled_critical": True,
+                "else": 2,
+                "states": {"active": 0, "failed": 2, "inactive": 0},
+                "states_default": 2,
+                "ignored": [],
+                "activating_levels": (30, 60),
+                "deactivating_levels": (30, 60),
+                "reloading_levels": (30, 60),
+            },
+            Section(
+                sockets={},
+                services={
+                    "systemd-timesyncd.service": UnitEntry(
+                        name="systemd-timesyncd.service",
+                        loaded_status="loaded",
+                        active_status="failed",
+                        current_state="failed",
+                        description="Import ZFS pool fgprs\\x2dpbs02\\x2dpool1\\x2d100",
+                        enabled_status="disabled",
+                    ),
+                },
+            ),
+            [
+                Result(state=State.OK, summary="Total: 1"),
+                Result(state=State.OK, summary="Disabled: 1"),
+                Result(state=State.CRIT, summary="Failed: 1"),
+            ],
+            id="One disabled, but critical state configured in params -> CRIT",
+        ),
         pytest.param(
             {
                 "else": 2,
                 "states": {"active": 0, "failed": 0, "inactive": 0},
                 "states_default": 2,
+                "disabled_critical": False,
                 "ignored": [],
                 "activating_levels": (30, 60),
                 "deactivating_levels": (30, 60),