systemdunits: dimension per unit state

modules/systemdunits/README.md

@@ -7,11 +7,13 @@ sidebar_label: "Systemd units"
 
 # Systemd units state monitoring with Netdata
 
-[`Systemd`](https://www.freedesktop.org/wiki/Software/systemd/) is a suite of basic building blocks for a Linux system.
+[Systemd](https://www.freedesktop.org/wiki/Software/systemd/) is a suite of basic building blocks for a Linux system.
 
-This module monitors `Systemd` units state.
+This module monitors Systemd units state.
 
-- Works only on linux systems.
+## Requirements
+
+- Works only on Linux systems.
 - Disabled by default. Should be explicitly enabled in the `go.d.conf`:
 
 ```yaml
@@ -20,31 +22,26 @@ modules:
   systemdunits: yes
 ```
 
-## Charts
-
-It produces the following charts:
-
-- Service Unit State in `state`
-- Socket Unit State in `state`
-- Target Unit State in `state`
-- Path Unit State in `state`
-- Device Unit State in `state`
-- Mount Unit State in `state`
-- Automount Unit State in `state`
-- Swap Unit State in `state`
-- Timer Unit State in `state`
-- Scope Unit State in `state`
-- Slice Unit State in `state`
-
-## Unit states
-
-| Code | Name           | Meaning                                                                                                                              |
-|------|----------------|--------------------------------------------------------------------------------------------------------------------------------------|
-| 1    | `active`       | started, bound, plugged in, ..., depending on the unit type                                                                          |
-| 2    | `inactive`     | stopped, unbound, unplugged, ..., depending on the unit type                                                                         |
-| 3    | `activating`   | in the process of being activated                                                                                                    |
-| 4    | `deactivating` | in the process of being deactivated                                                                                                  |
-| 5    | `failed`       | the service failed in some way (process returned error code on exit, or crashed, an operation timed out, or after too many restarts) |
+## Metrics
+
+The unit types and states description can be found in
+the [official documentation](https://www.freedesktop.org/software/systemd/man/systemd.html#Concepts).
+
+All metrics have "systemd." prefix.
+
+| Metric               | Scope |                     Dimensions                     | Units |
+|----------------------|:-----:|:--------------------------------------------------:|:-----:|
+| service_unit_state   | unit  | active, inactive, activating, deactivating, failed | state |
+| socket_unit_state    | unit  | active, inactive, activating, deactivating, failed | state |
+| target_unit_state    | unit  | active, inactive, activating, deactivating, failed | state |
+| path_unit_state      | unit  | active, inactive, activating, deactivating, failed | state |
+| device_unit_state    | unit  | active, inactive, activating, deactivating, failed | state |
+| mount_unit_state     | unit  | active, inactive, activating, deactivating, failed | state |
+| automount_unit_state | unit  | active, inactive, activating, deactivating, failed | state |
+| swap_unit_state      | unit  | active, inactive, activating, deactivating, failed | state |
+| timer_unit_state     | unit  | active, inactive, activating, deactivating, failed | state |
+| scope_unit_state     | unit  | active, inactive, activating, deactivating, failed | state |
+| slice_unit_state     | unit  | active, inactive, activating, deactivating, failed | state |
 
 ## Configuration
 

modules/systemdunits/charts.go

@@ -6,86 +6,71 @@
 package systemdunits
 
 import (
+	"fmt"
+
 	"github.com/netdata/go.d.plugin/agent/module"
+
+	"golang.org/x/text/cases"
+	"golang.org/x/text/language"
 )
 
-// systemd unit types: https://www.freedesktop.org/software/systemd/man/systemd.html
-var charts = module.Charts{
-	{
-		ID:    "service_unit_state",
-		Title: "Service Unit State (1: active, 2: inactive, 3: activating, 4: deactivating, 5: failed)",
-		Units: "state",
-		Fam:   "service",
-		Ctx:   "systemd.service_units_state",
-	},
-	{
-		ID:    "socket_unit_state",
-		Title: "Socket Unit State (1: active, 2: inactive, 3: activating, 4: deactivating, 5: failed)",
-		Units: "state",
-		Fam:   "socket",
-		Ctx:   "systemd.socket_unit_state",
-	},
-	{
-		ID:    "target_unit_state",
-		Title: "Target Unit State (1: active, 2: inactive, 3: activating, 4: deactivating, 5: failed)",
-		Units: "state",
-		Fam:   "target",
-		Ctx:   "systemd.target_unit_state",
-	},
-	{
-		ID:    "path_unit_state",
-		Title: "Path Unit State (1: active, 2: inactive, 3: activating, 4: deactivating, 5: failed)",
-		Units: "state",
-		Fam:   "path",
-		Ctx:   "systemd.path_unit_state",
-	},
-	{
-		ID:    "device_unit_state",
-		Title: "Device Unit State (1: active, 2: inactive, 3: activating, 4: deactivating, 5: failed)",
-		Units: "state",
-		Fam:   "device",
-		Ctx:   "systemd.device_unit_state",
-	},
-	{
-		ID:    "mount_unit_state",
-		Title: "Mount Unit State (1: active, 2: inactive, 3: activating, 4: deactivating, 5: failed)",
-		Units: "state",
-		Fam:   "mount",
-		Ctx:   "systemd.mount_unit_state",
-	},
-	{
-		ID:    "automount_unit_state",
-		Title: "Automount Unit State (1: active, 2: inactive, 3: activating, 4: deactivating, 5: failed)",
-		Units: "state",
-		Fam:   "automount",
-		Ctx:   "systemd.automount_unit_state",
-	},
-	{
-		ID:    "swap_unit_state",
-		Title: "Swap Unit State (1: active, 2: inactive, 3: activating, 4: deactivating, 5: failed)",
-		Units: "state",
-		Fam:   "swap",
-		Ctx:   "systemd.swap_unit_state",
-	},
-	{
-		ID:    "timer_unit_state",
-		Title: "Timer Unit State (1: active, 2: inactive, 3: activating, 4: deactivating, 5: failed)",
-		Units: "state",
-		Fam:   "timer",
-		Ctx:   "systemd.timer_unit_state",
-	},
-	{
-		ID:    "scope_unit_state",
-		Title: "Scope Unit State (1: active, 2: inactive, 3: activating, 4: deactivating, 5: failed)",
-		Units: "state",
-		Fam:   "scope",
-		Ctx:   "systemd.scope_unit_state",
-	},
-	{
-		ID:    "slice_unit_state",
-		Title: "Slice Unit State (1: active, 2: inactive, 3: activating, 4: deactivating, 5: failed)",
-		Units: "state",
-		Fam:   "slice",
-		Ctx:   "systemd.slice_unit_state",
-	},
+const (
+	prioServiceUnitState = module.Priority + iota
+	prioSocketUnitState
+	prioTargetUnitState
+	prioPathUnitState
+	prioDeviceUnitState
+	prioMountUnitState
+	prioAutomountUnitState
+	prioSwapUnitState
+	prioTimerUnitState
+	prioScopeUnitState
+	prioSliceUnitState
+)
+
+var prioMap = map[string]int{
+	unitTypeService:   prioServiceUnitState,
+	unitTypeSocket:    prioSocketUnitState,
+	unitTypeTarget:    prioTargetUnitState,
+	unitTypePath:      prioPathUnitState,
+	unitTypeDevice:    prioDeviceUnitState,
+	unitTypeMount:     prioMountUnitState,
+	unitTypeAutomount: prioAutomountUnitState,
+	unitTypeSwap:      prioSwapUnitState,
+	unitTypeTimer:     prioTimerUnitState,
+	unitTypeScope:     prioScopeUnitState,
+	unitTypeSlice:     prioSliceUnitState,
+}
+
+func newTypedUnitStateChartTmpl(name, typ string) *module.Chart {
+	chart := module.Chart{
+		ID:       fmt.Sprintf("unit_%s_%s_state", name, typ),
+		Title:    fmt.Sprintf("%s Unit State", cases.Title(language.English, cases.Compact).String(typ)),
+		Units:    "state",
+		Fam:      fmt.Sprintf("%s units", typ),
+		Ctx:      fmt.Sprintf("systemd.%s_unit_state", typ),
+		Priority: prioMap[typ],
+		Labels: []module.Label{
+			{Key: "unit_name", Value: name},
+		},
+		Dims: module.Dims{
+			{Name: unitStateActive},
+			{Name: unitStateInactive},
+			{Name: unitStateActivating},
+			{Name: unitStateDeactivating},
+			{Name: unitStateFailed},
+		},
+	}
+	for _, d := range chart.Dims {
+		d.ID = fmt.Sprintf("unit_%s_%s_state_%s", name, typ, d.Name)
+	}
+	return &chart
+}
+
+func (s *SystemdUnits) addUnitToCharts(name, typ string) {
+	chart := newTypedUnitStateChartTmpl(name, typ)
+
+	if err := s.Charts().Add(chart); err != nil {
+		s.Warning(err)
+	}
 }

modules/systemdunits/collect.go

@@ -12,11 +12,41 @@ import (
 	"strconv"
 	"strings"
 
-	"github.com/netdata/go.d.plugin/agent/module"
-
 	"github.com/coreos/go-systemd/v22/dbus"
 )
 
+const (
+	// https://www.freedesktop.org/software/systemd/man/systemd.html
+	unitStateActive       = "active"
+	unitStateInactive     = "inactive"
+	unitStateActivating   = "activating"
+	unitStateDeactivating = "deactivating"
+	unitStateFailed       = "failed"
+
+	// https://www.freedesktop.org/software/systemd/man/systemd.html
+	unitTypeService   = "service"
+	unitTypeSocket    = "socket"
+	unitTypeTarget    = "target"
+	unitTypePath      = "path"
+	unitTypeDevice    = "device"
+	unitTypeMount     = "mount"
+	unitTypeAutomount = "automount"
+	unitTypeSwap      = "swap"
+	unitTypeTimer     = "timer"
+	unitTypeScope     = "scope"
+	unitTypeSlice     = "slice"
+)
+
+var (
+	unitStates = []string{
+		unitStateActive,
+		unitStateActivating,
+		unitStateFailed,
+		unitStateInactive,
+		unitStateDeactivating,
+	}
+)
+
 func (s *SystemdUnits) collect() (map[string]int64, error) {
 	conn, err := s.getConnection()
 	if err != nil {
@@ -48,19 +78,28 @@ func (s *SystemdUnits) collect() (map[string]int64, error) {
 		return nil, nil
 	}
 
-	collected := make(map[string]int64)
-	s.collectUnitsStates(collected, units)
-	return collected, nil
+	mx := make(map[string]int64)
+	s.collectUnitsStates(mx, units)
+
+	return mx, nil
 }
 
-func (s *SystemdUnits) collectUnitsStates(collected map[string]int64, units []dbus.UnitStatus) {
+func (s *SystemdUnits) collectUnitsStates(mx map[string]int64, units []dbus.UnitStatus) {
 	for _, unit := range units {
-		name := cleanUnitName(unit.Name)
-		if !s.collectedUnits[name] {
-			s.collectedUnits[name] = true
-			s.addUnitToCharts(name)
+		name, typ := extractUnitNameType(cleanUnitName(unit.Name))
+		if name == "" || typ == "" {
+			continue
+		}
+
+		if !s.units[unit.Name] {
+			s.units[unit.Name] = true
+			s.addUnitToCharts(name, typ)
 		}
-		collected[name] = convertUnitState(unit.ActiveState)
+
+		for _, s := range unitStates {
+			mx[fmt.Sprintf("unit_%s_%s_state_%s", name, typ, s)] = 0
+		}
+		mx[fmt.Sprintf("unit_%s_%s_state_%s", name, typ, unit.ActiveState)] = 1
 	}
 }
 
@@ -124,8 +163,8 @@ func (s *SystemdUnits) getLoadedUnits(conn systemdConnection) ([]dbus.UnitStatus
 			loaded = append(loaded, unit)
 		}
 	}
-
 	s.Debugf("got total/loaded %d/%d units", len(units), len(loaded))
+
 	return loaded, nil
 }
 
@@ -134,11 +173,8 @@ func (s *SystemdUnits) getLoadedUnitsByPatterns(conn systemdConnection) ([]dbus.
 	defer cancel()
 
 	s.Debugf("calling function 'ListUnitsByPatterns'")
-	units, err := conn.ListUnitsByPatternsContext(
-		ctx,
-		[]string{"active", "activating", "failed", "inactive", "deactivating"},
-		s.Include,
-	)
+
+	units, err := conn.ListUnitsByPatternsContext(ctx, unitStates, s.Include)
 	if err != nil {
 		return nil, fmt.Errorf("error on ListUnitsByPatterns: %v", err)
 	}
@@ -149,60 +185,17 @@ func (s *SystemdUnits) getLoadedUnitsByPatterns(conn systemdConnection) ([]dbus.
 			loaded = append(loaded, unit)
 		}
 	}
-
 	s.Debugf("got total/loaded %d/%d units", len(units), len(loaded))
-	return loaded, nil
-}
-
-func (s *SystemdUnits) addUnitToCharts(name string) {
-	typ := extractUnitType(name)
-	if typ == "" {
-		s.Warningf("add dimension (unit '%s'): can't extract unit type", name)
-		return
-	}
-
-	id := fmt.Sprintf("%s_unit_state", typ)
-	chart := s.Charts().Get(id)
-	if chart == nil {
-		s.Warningf("add dimension (unit '%s'): can't find '%s' chart", name, id)
-		return
-	}
 
-	dim := &module.Dim{
-		ID:   name,
-		Name: name[:len(name)-len(typ)-1], // name.type => name
-	}
-	if err := chart.AddDim(dim); err != nil {
-		s.Warningf("add dimension (unit '%s'): %v", name, err)
-	}
-	chart.MarkNotCreated()
+	return loaded, nil
 }
 
-func extractUnitType(name string) string {
-	// name.type => type
+func extractUnitNameType(name string) (string, string) {
 	idx := strings.LastIndexByte(name, '.')
 	if idx <= 0 {
-		return ""
-	}
-	return name[idx+1:]
-}
-
-func convertUnitState(state string) int64 {
-	// https://www.freedesktop.org/software/systemd/man/systemd.html
-	switch state {
-	case "active":
-		return 1
-	case "inactive":
-		return 2
-	case "activating":
-		return 3
-	case "deactivating":
-		return 4
-	case "failed":
-		return 5
-	default:
-		return -1
+		return "", ""
 	}
+	return name[:idx], name[idx+1:]
 }
 
 func cleanUnitName(name string) string {