go.d systemdunits add "skip_transient" (#17725)

8 files changed, 109 insertions, 2 deletions

diff --git a/src/go/collectors/go.d.plugin/modules/systemdunits/client.go b/src/go/collectors/go.d.plugin/modules/systemdunits/client.go
index b1152865f4..e6363d1329 100644
--- a/src/go/collectors/go.d.plugin/modules/systemdunits/client.go
+++ b/src/go/collectors/go.d.plugin/modules/systemdunits/client.go
@@ -17,6 +17,7 @@ type systemdClient interface {
 type systemdConnection interface {
 	Close()
 	GetManagerProperty(string) (string, error)
+	GetUnitPropertyContext(ctx context.Context, unit string, propertyName string) (*dbus.Property, error)
 	ListUnitsContext(ctx context.Context) ([]dbus.UnitStatus, error)
 	ListUnitsByPatternsContext(ctx context.Context, states []string, patterns []string) ([]dbus.UnitStatus, error)
 	ListUnitFilesByPatternsContext(ctx context.Context, states []string, patterns []string) ([]dbus.UnitFile, error)
diff --git a/src/go/collectors/go.d.plugin/modules/systemdunits/collect_units.go b/src/go/collectors/go.d.plugin/modules/systemdunits/collect_units.go
index 4cfc89d604..0cf97af03f 100644
--- a/src/go/collectors/go.d.plugin/modules/systemdunits/collect_units.go
+++ b/src/go/collectors/go.d.plugin/modules/systemdunits/collect_units.go
@@ -14,6 +14,8 @@ import (
 	"github.com/coreos/go-systemd/v22/dbus"
 )
 
+const transientProperty = "Transient"
+
 const (
 	// https://www.freedesktop.org/software/systemd/man/systemd.html
 	unitStateActive       = "active"
@@ -55,6 +57,20 @@ func (s *SystemdUnits) collectUnits(mx map[string]int64, conn systemdConnection)
 
 		seen[unit.Name] = true
 
+		if s.SkipTransient {
+			if _, ok := s.unitTransient[unit.Name]; !ok {
+				prop, err := s.getUnitTransientProperty(conn, unit.Name)
+				if err != nil {
+					return err
+				}
+				prop = strings.Trim(prop, "\"")
+				s.unitTransient[unit.Name] = prop == "true"
+			}
+			if s.unitTransient[unit.Name] {
+				continue
+			}
+		}
+
 		if !s.seenUnits[unit.Name] {
 			s.seenUnits[unit.Name] = true
 			s.addUnitCharts(name, typ)
@@ -75,6 +91,12 @@ func (s *SystemdUnits) collectUnits(mx map[string]int64, conn systemdConnection)
 		}
 	}
 
+	for k := range s.unitTransient {
+		if !seen[k] {
+			delete(s.unitTransient, k)
+		}
+	}
+
 	return nil
 }
 
@@ -130,6 +152,20 @@ func (s *SystemdUnits) getLoadedUnitsByPatterns(conn systemdConnection) ([]dbus.
 	return loaded, nil
 }
 
+func (s *SystemdUnits) getUnitTransientProperty(conn systemdConnection, unit string) (string, error) {
+	ctx, cancel := context.WithTimeout(context.Background(), s.Timeout.Duration())
+	defer cancel()
+
+	s.Debugf("calling function 'GetUnitProperty' for unit '%s'", unit)
+
+	prop, err := conn.GetUnitPropertyContext(ctx, unit, transientProperty)
+	if err != nil {
+		return "", fmt.Errorf("error on GetUnitProperty: %v", err)
+	}
+
+	return prop.Value.String(), nil
+}
+
 func extractUnitNameType(name string) (string, string, bool) {
 	idx := strings.LastIndexByte(name, '.')
 	if idx <= 0 {
diff --git a/src/go/collectors/go.d.plugin/modules/systemdunits/config_schema.json b/src/go/collectors/go.d.plugin/modules/systemdunits/config_schema.json
index c13cabc6ec..016e984ced 100644
--- a/src/go/collectors/go.d.plugin/modules/systemdunits/config_schema.json
+++ b/src/go/collectors/go.d.plugin/modules/systemdunits/config_schema.json
@@ -18,6 +18,12 @@
         "minimum": 0.5,
         "default": 2
       },
+      "skip_transient": {
+        "title": "Skip transient units",
+        "description": "If set, skip data collection for systemd transient units.",
+        "type": "boolean",
+        "default": false
+      },
       "include": {
         "title": "Include",
         "description": "Configuration for monitoring specific systemd units. Include systemd units whose names match any of the specified [patterns](https://golang.org/pkg/path/filepath/#Match).",
@@ -86,6 +92,7 @@
           "fields": [
             "update_every",
             "timeout",
+            "skip_transient",
             "include"
           ]
         },
@@ -102,6 +109,9 @@
     "timeout": {
       "ui:help": "Accepts decimals for precise control (e.g., type 1.5 for 1.5 seconds)."
     },
+    "skip_transient": {
+      "ui:help": "A systemd transient unit is a temporary unit created on-the-fly, typically used for ad-hoc tasks or testing purposes. They are created using the `systemd-run` command, which allows you to specify unit properties directly on the command line."
+    },
     "include": {
       "ui:listFlavour": "list"
     },
diff --git a/src/go/collectors/go.d.plugin/modules/systemdunits/metadata.yaml b/src/go/collectors/go.d.plugin/modules/systemdunits/metadata.yaml
index 6ca804fb30..791e58400f 100644
--- a/src/go/collectors/go.d.plugin/modules/systemdunits/metadata.yaml
+++ b/src/go/collectors/go.d.plugin/modules/systemdunits/metadata.yaml
@@ -77,6 +77,10 @@ modules:
                   - pattern1
                   - pattern2
                 ```
+            - name: skip_transient
+              description: If set, skip data collection for systemd transient units.
+              default_value: "false"
+              required: false
             - name: collect_unit_files
               description: If set to true, collect the state of installed unit files. Enabling this may increase system overhead.
               default_value: "false"
diff --git a/src/go/collectors/go.d.plugin/modules/systemdunits/systemdunits.go b/src/go/collectors/go.d.plugin/modules/systemdunits/systemdunits.go
index c39a5fc5bb..c5bebfdc71 100644
--- a/src/go/collectors/go.d.plugin/modules/systemdunits/systemdunits.go
+++ b/src/go/collectors/go.d.plugin/modules/systemdunits/systemdunits.go
@@ -36,6 +36,7 @@ func New() *SystemdUnits {
 		Config: Config{
 			Timeout:               web.Duration(time.Second * 2),
 			Include:               []string{"*.service"},
+			SkipTransient:         false,
 			CollectUnitFiles:      false,
 			IncludeUnitFiles:      []string{"*.service"},
 			CollectUnitFilesEvery: web.Duration(time.Minute * 5),
@@ -43,6 +44,7 @@ func New() *SystemdUnits {
 		charts:        &module.Charts{},
 		client:        newSystemdDBusClient(),
 		seenUnits:     make(map[string]bool),
+		unitTransient: make(map[string]bool),
 		seenUnitFiles: make(map[string]bool),
 	}
 }
@@ -51,6 +53,7 @@ type Config struct {
 	UpdateEvery           int          `yaml:"update_every,omitempty" json:"update_every"`
 	Timeout               web.Duration `yaml:"timeout,omitempty" json:"timeout"`
 	Include               []string     `yaml:"include,omitempty" json:"include"`
+	SkipTransient         bool         `yaml:"skip_transient" json:"skip_transient"`
 	CollectUnitFiles      bool         `yaml:"collect_unit_files,omitempty" json:"collect_unit_files"`
 	IncludeUnitFiles      []string     `yaml:"include_unit_files,omitempty" json:"include_unit_files"`
 	CollectUnitFilesEvery web.Duration `yaml:"collect_unit_files_every,omitempty" json:"collect_unit_files_every"`
@@ -65,8 +68,9 @@ type SystemdUnits struct {
 
 	systemdVersion int
 
-	seenUnits map[string]bool
-	unitSr    matcher.Matcher
+	seenUnits     map[string]bool
+	unitTransient map[string]bool
+	unitSr        matcher.Matcher
 
 	lastListUnitFilesTime time.Time
 	cachedUnitFiles       []dbus.UnitFile
diff --git a/src/go/collectors/go.d.plugin/modules/systemdunits/systemdunits_test.go b/src/go/collectors/go.d.plugin/modules/systemdunits/systemdunits_test.go
index fa9326b03f..89c0a92dda 100644
--- a/src/go/collectors/go.d.plugin/modules/systemdunits/systemdunits_test.go
+++ b/src/go/collectors/go.d.plugin/modules/systemdunits/systemdunits_test.go
@@ -12,6 +12,7 @@ import (
 	"os"
 	"path/filepath"
 	"slices"
+	"strings"
 	"testing"
 
 	"github.com/netdata/netdata/go/go.d.plugin/agent/module"
@@ -384,6 +385,37 @@ func TestSystemdUnits_Collect(t *testing.T) {
 				"unit_var-lib-nfs-rpc_pipefs_mount_state_inactive":                 1,
 			},
 		},
+		"success v230+ on collecting all unit type with skip transient": {
+			prepare: func() *SystemdUnits {
+				systemd := New()
+				systemd.Include = []string{"*"}
+				systemd.SkipTransient = true
+				systemd.client = prepareOKClient(230)
+				return systemd
+			},
+			wantCollected: map[string]int64{
+				"unit_systemd-ask-password-wall_service_state_activating":   0,
+				"unit_systemd-ask-password-wall_service_state_active":       0,
+				"unit_systemd-ask-password-wall_service_state_deactivating": 0,
+				"unit_systemd-ask-password-wall_service_state_failed":       0,
+				"unit_systemd-ask-password-wall_service_state_inactive":     1,
+				"unit_systemd-fsck-root_service_state_activating":           0,
+				"unit_systemd-fsck-root_service_state_active":               0,
+				"unit_systemd-fsck-root_service_state_deactivating":         0,
+				"unit_systemd-fsck-root_service_state_failed":               0,
+				"unit_systemd-fsck-root_service_state_inactive":             1,
+				"unit_user-runtime-dir@1000_service_state_activating":       0,
+				"unit_user-runtime-dir@1000_service_state_active":           1,
+				"unit_user-runtime-dir@1000_service_state_deactivating":     0,
+				"unit_user-runtime-dir@1000_service_state_failed":           0,
+				"unit_user-runtime-dir@1000_service_state_inactive":         0,
+				"unit_user@1000_service_state_activating":                   0,
+				"unit_user@1000_service_state_active":                       1,
+				"unit_user@1000_service_state_deactivating":                 0,
+				"unit_user@1000_service_state_failed":                       0,
+				"unit_user@1000_service_state_inactive":                     0,
+			},
+		},
 		"success v230- on collecting all unit types": {
 			prepare: func() *SystemdUnits {
 				systemd := New()
@@ -946,6 +978,24 @@ func (m *mockConn) GetManagerProperty(prop string) (string, error) {
 	return fmt.Sprintf("%d.6-1-manjaro", m.version), nil
 }
 
+func (m *mockConn) GetUnitPropertyContext(_ context.Context, unit string, propertyName string) (*dbus.Property, error) {
+	if propertyName != transientProperty {
+		return nil, fmt.Errorf("'GetUnitProperty' unkown property name: %s", propertyName)
+	}
+
+	var prop dbus.Property
+
+	if strings.HasSuffix(unit, ".service") {
+		prop = dbus.PropDescription("false")
+	} else {
+		prop = dbus.PropDescription("true")
+	}
+
+	prop.Name = propertyName
+
+	return &prop, nil
+}
+
 func (m *mockConn) ListUnitsContext(_ context.Context) ([]dbus.UnitStatus, error) {
 	if m.errOnListUnits {
 		return nil, errors.New("'ListUnits' call error")
diff --git a/src/go/collectors/go.d.plugin/modules/systemdunits/testdata/config.json b/src/go/collectors/go.d.plugin/modules/systemdunits/testdata/config.json
index ac1f53fb6b..1ab5b47ead 100644
--- a/src/go/collectors/go.d.plugin/modules/systemdunits/testdata/config.json
+++ b/src/go/collectors/go.d.plugin/modules/systemdunits/testdata/config.json
@@ -4,6 +4,7 @@
   "include": [
     "ok"
   ],
+  "skip_transient": true,
   "collect_unit_files": true,
   "collect_unit_files_every": 123.123,
   "include_unit_files": [
diff --git a/src/go/collectors/go.d.plugin/modules/systemdunits/testdata/config.yaml b/src/go/collectors/go.d.plugin/modules/systemdunits/testdata/config.yaml
index dfed52c3f8..d1894aea14 100644
--- a/src/go/collectors/go.d.plugin/modules/systemdunits/testdata/config.yaml
+++ b/src/go/collectors/go.d.plugin/modules/systemdunits/testdata/config.yaml
@@ -2,6 +2,7 @@ update_every: 123
 timeout: 123.123
 include:
   - ok
+skip_transient: true
 collect_unit_files: true
 collect_unit_files_every: 123.123
 include_unit_files: