logon: BREAKING: replace wmi query by Win32 API calls and expose deta…

…iled logon. (click PR for more information) Signed-off-by: Jan-Otto Kröpke <[email protected]>
prometheus-community · Oct 12, 2024 · d10074a · d10074a
1 parent 22fdb33
commit d10074a
Show file tree

Hide file tree

Showing 6 changed files with 304 additions and 181 deletions.
diff --git a/README.md b/README.md
@@ -35,6 +35,7 @@ Name     | Description | Enabled by default
 [netframework](docs/collector.netframework.md) | .NET Framework metrics |
 [net](docs/collector.net.md) | Network interface I/O | &#10003;
 [os](docs/collector.os.md) | OS metrics (memory, processes, users) | &#10003;
+[perfdata](docs/collector.perfdata.md) | Custom perfdata metrics |
 [physical_disk](docs/collector.physical_disk.md) | physical disk metrics | &#10003;
 [printer](docs/collector.printer.md) | Printer metrics |
 [process](docs/collector.process.md) | Per-process metrics |

diff --git a/docs/collector.logon.md b/docs/collector.logon.md
@@ -2,35 +2,60 @@
 
 The logon collector exposes metrics detailing the active user logon sessions.
 
-|||
--|-
-Metric name prefix  | `logon`
-Classes             | [`Win32_LogonSession`](https://docs.microsoft.com/en-us/windows/win32/cimwin32prov/win32-logonsession)
-Enabled by default? | No
-
-> :warning: **On some deployments, this collector seems to have some memory/timeout issues**: See [#583](https://github.com/prometheus-community/windows_exporter/issues/583)
+|                     |           |
+|---------------------|-----------|
+| Metric name prefix  | `logon`   |
+| Source              | Win32 API |
+| Enabled by default? | No        |
 
 ## Flags
 
 None
 
 ## Metrics
 
-Name | Description | Type | Labels
------|-------------|------|-------
-`windows_logon_logon_type` | Number of active user logon sessions | gauge | status
+| Name                                      | Description                                | Type  | Labels                             |
+|-------------------------------------------|--------------------------------------------|-------|------------------------------------|
+| `windows_logon_session_timestamp_seconds` | timestamp of the logon session in seconds. | gauge | `domain`, `id`, `type`, `username` |
 
 ### Example metric
 Query the total number of interactive logon sessions
 ```
-windows_logon_logon_type{status="interactive"}
+# HELP windows_logon_session_timestamp_seconds timestamp of the logon session in seconds
+# TYPE windows_logon_session_timestamp_seconds gauge
+windows_logon_session_timestamp_seconds{domain="",id="103590",type="System",username=""} 1.728759837e+09
+windows_logon_session_timestamp_seconds{domain="Font Driver Host",id="104592",type="Interactive",username="UMFD-0"} 1.728759837e+09
+windows_logon_session_timestamp_seconds{domain="Font Driver Host",id="124850",type="Interactive",username="UMFD-1"} 1.728759838e+09
+windows_logon_session_timestamp_seconds{domain="JOK-PC",id="521539",type="Interactive",username="Jan"} 1.728759839e+09
+windows_logon_session_timestamp_seconds{domain="JOK-PC",id="521983",type="Interactive",username="Jan"} 1.728759839e+09
+windows_logon_session_timestamp_seconds{domain="NT-AUTORITÄT",id="997",type="Service",username="Lokaler Dienst"} 1.728759838e+09
+windows_logon_session_timestamp_seconds{domain="WORKGROUP",id="996",type="Service",username="JOK-PC$"} 1.728759838e+09
+windows_logon_session_timestamp_seconds{domain="WORKGROUP",id="999",type="System",username="JOK-PC$"} 1.728759837e+09
+windows_logon_session_timestamp_seconds{domain="Window Manager",id="148473",type="Interactive",username="DWM-1"} 1.728759838e+09
+windows_logon_session_timestamp_seconds{domain="Window Manager",id="148582",type="Interactive",username="DWM-1"} 1.728759838e+09
 ```
 
+### Possible values for `type`
+
+- System
+- Interactive
+- Network
+- Batch
+- Service
+- Proxy
+- Unlock
+- NetworkCleartext
+- NewCredentials
+- RemoteInteractive
+- CachedInteractive
+- CachedRemoteInteractive
+- CachedUnlock
+
 ## Useful queries
 Query the total number of local and remote (I.E. Terminal Services) interactive sessions.
 ```
-windows_logon_logon_type{status=~"interactive|remote_interactive"}
+count(windows_logon_logon_type{type=~"Interactive|RemoteInteractive"}) by (type)
 ```
 
 ## Alerting examples
-_This collector does not yet have alerting examples, we would appreciate your help adding them!_
+_This collector doesn’t yet have alerting examples, we would appreciate your help adding them!_
diff --git a/internal/collector/logon/logon.go b/internal/collector/logon/logon.go
@@ -3,10 +3,11 @@
 package logon
 
 import (
-	"errors"
+	"fmt"
 	"log/slog"
 
 	"github.com/alecthomas/kingpin/v2"
+	"github.com/prometheus-community/windows_exporter/internal/headers/secur32"
 	"github.com/prometheus-community/windows_exporter/internal/types"
 	"github.com/prometheus/client_golang/prometheus"
 	"github.com/yusufpapurcu/wmi"
@@ -20,10 +21,9 @@ var ConfigDefaults = Config{}
 
 // A Collector is a Prometheus Collector for WMI metrics.
 type Collector struct {
-	config    Config
-	wmiClient *wmi.Client
+	config Config
 
-	logonType *prometheus.Desc
+	sessionInfo *prometheus.Desc
 }
 
 func New(config *Config) *Collector {
@@ -54,16 +54,11 @@ func (c *Collector) Close(_ *slog.Logger) error {
 	return nil
 }
 
-func (c *Collector) Build(_ *slog.Logger, wmiClient *wmi.Client) error {
-	if wmiClient == nil || wmiClient.SWbemServicesClient == nil {
-		return errors.New("wmiClient or SWbemServicesClient is nil")
-	}
-
-	c.wmiClient = wmiClient
-	c.logonType = prometheus.NewDesc(
-		prometheus.BuildFQName(types.Namespace, Name, "logon_type"),
-		"Number of active logon sessions (LogonSession.LogonType)",
-		[]string{"status"},
+func (c *Collector) Build(_ *slog.Logger, _ *wmi.Client) error {
+	c.sessionInfo = prometheus.NewDesc(
+		prometheus.BuildFQName(types.Namespace, Name, "session_timestamp_seconds"),
+		"timestamp of the logon session in seconds.",
+		[]string{"id", "username", "domain", "type"},
 		nil,
 	)
 
@@ -72,171 +67,28 @@ func (c *Collector) Build(_ *slog.Logger, wmiClient *wmi.Client) error {
 
 // Collect sends the metric values for each metric
 // to the provided prometheus Metric channel.
-func (c *Collector) Collect(_ *types.ScrapeContext, logger *slog.Logger, ch chan<- prometheus.Metric) error {
-	logger = logger.With(slog.String("collector", Name))
+func (c *Collector) Collect(_ *types.ScrapeContext, _ *slog.Logger, ch chan<- prometheus.Metric) error {
 	if err := c.collect(ch); err != nil {
-		logger.Error("failed collecting user metrics",
-			slog.Any("err", err),
-		)
-
 		return err
 	}
 
 	return nil
 }
 
-// Win32_LogonSession docs:
-// - https://docs.microsoft.com/en-us/windows/win32/cimwin32prov/win32-logonsession
-type Win32_LogonSession struct {
-	LogonType uint32
-}
-
 func (c *Collector) collect(ch chan<- prometheus.Metric) error {
-	var dst []Win32_LogonSession
-	if err := c.wmiClient.Query("SELECT * FROM Win32_LogonSession", &dst); err != nil {
-		return err
-	}
-
-	if len(dst) == 0 {
-		return errors.New("WMI query returned empty result set")
+	logonSessions, err := secur32.GetLogonSessions()
+	if err != nil {
+		return fmt.Errorf("failed to get logon sessions: %w", err)
 	}
 
-	// Init counters
-	system := 0
-	interactive := 0
-	network := 0
-	batch := 0
-	service := 0
-	proxy := 0
-	unlock := 0
-	networkcleartext := 0
-	newcredentials := 0
-	remoteinteractive := 0
-	cachedinteractive := 0
-	cachedremoteinteractive := 0
-	cachedunlock := 0
-
-	for _, entry := range dst {
-		switch entry.LogonType {
-		case 0:
-			system++
-		case 2:
-			interactive++
-		case 3:
-			network++
-		case 4:
-			batch++
-		case 5:
-			service++
-		case 6:
-			proxy++
-		case 7:
-			unlock++
-		case 8:
-			networkcleartext++
-		case 9:
-			newcredentials++
-		case 10:
-			remoteinteractive++
-		case 11:
-			cachedinteractive++
-		case 12:
-			cachedremoteinteractive++
-		case 13:
-			cachedunlock++
-		}
+	for _, session := range logonSessions {
+		ch <- prometheus.MustNewConstMetric(
+			c.sessionInfo,
+			prometheus.GaugeValue,
+			float64(session.LogonTime.Unix()),
+			session.LogonId.String(), session.UserName, session.LogonDomain, session.LogonType.String(),
+		)
 	}
 
-	ch <- prometheus.MustNewConstMetric(
-		c.logonType,
-		prometheus.GaugeValue,
-		float64(system),
-		"system",
-	)
-
-	ch <- prometheus.MustNewConstMetric(
-		c.logonType,
-		prometheus.GaugeValue,
-		float64(interactive),
-		"interactive",
-	)
-
-	ch <- prometheus.MustNewConstMetric(
-		c.logonType,
-		prometheus.GaugeValue,
-		float64(network),
-		"network",
-	)
-
-	ch <- prometheus.MustNewConstMetric(
-		c.logonType,
-		prometheus.GaugeValue,
-		float64(batch),
-		"batch",
-	)
-
-	ch <- prometheus.MustNewConstMetric(
-		c.logonType,
-		prometheus.GaugeValue,
-		float64(service),
-		"service",
-	)
-
-	ch <- prometheus.MustNewConstMetric(
-		c.logonType,
-		prometheus.GaugeValue,
-		float64(proxy),
-		"proxy",
-	)
-
-	ch <- prometheus.MustNewConstMetric(
-		c.logonType,
-		prometheus.GaugeValue,
-		float64(unlock),
-		"unlock",
-	)
-
-	ch <- prometheus.MustNewConstMetric(
-		c.logonType,
-		prometheus.GaugeValue,
-		float64(networkcleartext),
-		"network_clear_text",
-	)
-
-	ch <- prometheus.MustNewConstMetric(
-		c.logonType,
-		prometheus.GaugeValue,
-		float64(newcredentials),
-		"new_credentials",
-	)
-
-	ch <- prometheus.MustNewConstMetric(
-		c.logonType,
-		prometheus.GaugeValue,
-		float64(remoteinteractive),
-		"remote_interactive",
-	)
-
-	ch <- prometheus.MustNewConstMetric(
-		c.logonType,
-		prometheus.GaugeValue,
-		float64(cachedinteractive),
-		"cached_interactive",
-	)
-
-	ch <- prometheus.MustNewConstMetric(
-		c.logonType,
-		prometheus.GaugeValue,
-		float64(remoteinteractive),
-		"cached_remote_interactive",
-	)
-
-	ch <- prometheus.MustNewConstMetric(
-		c.logonType,
-		prometheus.GaugeValue,
-		float64(cachedunlock),
-		"cached_unlock",
-	)
-
 	return nil
 }