mscluster: restore support for Windows Server 2016-2019 (#1882)

Signed-off-by: Jan-Otto Kröpke <[email protected]>

Author: jkroepke

internal/collector/hyperv/hyperv.go

@@ -23,11 +23,10 @@ import (
 	"strings"
 	"sync"
 
+	"github.com/Microsoft/hcsshim/osversion"
 	"github.com/alecthomas/kingpin/v2"
 	"github.com/prometheus-community/windows_exporter/internal/mi"
-	"github.com/prometheus-community/windows_exporter/internal/types"
 	"github.com/prometheus/client_golang/prometheus"
-	"golang.org/x/sys/windows"
 )
 
 const (
@@ -157,19 +156,17 @@ func (c *Collector) Build(_ *slog.Logger, _ *mi.Session) error {
 		return nil
 	}
 
-	version := windows.RtlGetVersion()
-
 	subCollectors := map[string]struct {
 		build          func() error
 		collect        func(ch chan<- prometheus.Metric) error
 		close          func()
-		minBuildNumber uint32
+		minBuildNumber uint16
 	}{
 		subCollectorDataStore: {
 			build:          c.buildDataStore,
 			collect:        c.collectDataStore,
 			close:          c.perfDataCollectorDataStore.Close,
-			minBuildNumber: types.BuildNumberWindowsServer2022,
+			minBuildNumber: osversion.LTSC2022,
 		},
 		subCollectorDynamicMemoryBalancer: {
 			build:   c.buildDynamicMemoryBalancer,
@@ -243,6 +240,8 @@ func (c *Collector) Build(_ *slog.Logger, _ *mi.Session) error {
 		},
 	}
 
+	buildNumber := osversion.Build()
+
 	// Result must order, to prevent test failures.
 	sort.Strings(c.config.CollectorsEnabled)
 
@@ -253,7 +252,7 @@ func (c *Collector) Build(_ *slog.Logger, _ *mi.Session) error {
 			return fmt.Errorf("unknown collector: %s", name)
 		}
 
-		if version.BuildNumber < subCollectors[name].minBuildNumber {
+		if buildNumber < subCollectors[name].minBuildNumber {
 			errs = append(errs, fmt.Errorf("collector %s requires Windows Server 2022 or newer", name))
 
 			continue

internal/collector/mscluster/mscluster_cluster.go

@@ -18,6 +18,7 @@ package mscluster
 import (
 	"fmt"
 
+	"github.com/Microsoft/hcsshim/osversion"
 	"github.com/prometheus-community/windows_exporter/internal/mi"
 	"github.com/prometheus-community/windows_exporter/internal/types"
 	"github.com/prometheus/client_golang/prometheus"
@@ -192,7 +193,14 @@ type msClusterCluster struct {
 }
 
 func (c *Collector) buildCluster() error {
-	clusterMIQuery, err := mi.NewQuery("SELECT * FROM MSCluster_Cluster")
+	buildNumber := osversion.Build()
+
+	wmiSelect := "AddEvictDelay,AdminAccessPoint,AutoAssignNodeSite,AutoBalancerLevel,AutoBalancerMode,BackupInProgress,BlockCacheSize,ClusSvcHangTimeout,ClusSvcRegroupOpeningTimeout,ClusSvcRegroupPruningTimeout,ClusSvcRegroupStageTimeout,ClusSvcRegroupTickInMilliseconds,ClusterEnforcedAntiAffinity,ClusterFunctionalLevel,ClusterGroupWaitDelay,ClusterLogLevel,ClusterLogSize,ClusterUpgradeVersion,CrossSiteDelay,CrossSiteThreshold,CrossSubnetDelay,CrossSubnetThreshold,CsvBalancer,DatabaseReadWriteMode,DefaultNetworkRole,DisableGroupPreferredOwnerRandomization,DrainOnShutdown,DynamicQuorumEnabled,EnableSharedVolumes,FixQuorum,GracePeriodEnabled,GracePeriodTimeout,GroupDependencyTimeout,HangRecoveryAction,IgnorePersistentStateOnStartup,LogResourceControls,LowerQuorumPriorityNodeId,MessageBufferLength,MinimumNeverPreemptPriority,MinimumPreemptorPriority,NetftIPSecEnabled,PlacementOptions,PlumbAllCrossSubnetRoutes,PreventQuorum,QuarantineDuration,QuarantineThreshold,QuorumArbitrationTimeMax,QuorumArbitrationTimeMin,QuorumLogFileSize,QuorumTypeValue,RequestReplyTimeout,ResiliencyDefaultPeriod,ResiliencyLevel,ResourceDllDeadlockPeriod,RootMemoryReserved,RouteHistoryLength,S2DBusTypes,S2DCacheDesiredState,S2DCacheFlashReservePercent,S2DCachePageSizeKBytes,S2DEnabled,S2DIOLatencyThreshold,S2DOptimizations,SameSubnetDelay,SameSubnetThreshold,SecurityLevel,SharedVolumeVssWriterOperationTimeout,ShutdownTimeoutInMinutes,UseClientAccessNetworksForSharedVolumes,WitnessDatabaseWriteTimeout,WitnessDynamicWeight,WitnessRestartInterval"
+	if buildNumber >= osversion.LTSC2022 {
+		wmiSelect += ",DetectManagedEvents,SecurityLevelForStorage,MaxNumberOfNodes,DetectManagedEventsThreshold,DetectedCloudPlatform"
+	}
+
+	clusterMIQuery, err := mi.NewQuery(fmt.Sprintf("SELECT %s FROM MSCluster_Cluster", wmiSelect))
 	if err != nil {
 		return fmt.Errorf("failed to create WMI query: %w", err)
 	}
@@ -852,27 +860,6 @@ func (c *Collector) collectCluster(ch chan<- prometheus.Metric) error {
 			v.Name,
 		)
 
-		ch <- prometheus.MustNewConstMetric(
-			c.clusterDetectedCloudPlatform,
-			prometheus.GaugeValue,
-			float64(v.DetectedCloudPlatform),
-			v.Name,
-		)
-
-		ch <- prometheus.MustNewConstMetric(
-			c.clusterDetectManagedEvents,
-			prometheus.GaugeValue,
-			float64(v.DetectManagedEvents),
-			v.Name,
-		)
-
-		ch <- prometheus.MustNewConstMetric(
-			c.clusterDetectManagedEventsThreshold,
-			prometheus.GaugeValue,
-			float64(v.DetectManagedEventsThreshold),
-			v.Name,
-		)
-
 		ch <- prometheus.MustNewConstMetric(
 			c.clusterDisableGroupPreferredOwnerRandomization,
 			prometheus.GaugeValue,
@@ -957,13 +944,6 @@ func (c *Collector) collectCluster(ch chan<- prometheus.Metric) error {
 			v.Name,
 		)
 
-		ch <- prometheus.MustNewConstMetric(
-			c.clusterMaxNumberOfNodes,
-			prometheus.GaugeValue,
-			float64(v.MaxNumberOfNodes),
-			v.Name,
-		)
-
 		ch <- prometheus.MustNewConstMetric(
 			c.clusterMessageBufferLength,
 			prometheus.GaugeValue,
@@ -1167,13 +1147,6 @@ func (c *Collector) collectCluster(ch chan<- prometheus.Metric) error {
 			v.Name,
 		)
 
-		ch <- prometheus.MustNewConstMetric(
-			c.clusterSecurityLevelForStorage,
-			prometheus.GaugeValue,
-			float64(v.SecurityLevelForStorage),
-			v.Name,
-		)
-
 		ch <- prometheus.MustNewConstMetric(
 			c.clusterSharedVolumeVssWriterOperationTimeout,
 			prometheus.GaugeValue,
@@ -1215,6 +1188,43 @@ func (c *Collector) collectCluster(ch chan<- prometheus.Metric) error {
 			float64(v.WitnessRestartInterval),
 			v.Name,
 		)
+
+		if osversion.Build() >= osversion.LTSC2022 {
+			ch <- prometheus.MustNewConstMetric(
+				c.clusterDetectManagedEvents,
+				prometheus.GaugeValue,
+				float64(v.DetectManagedEvents),
+				v.Name,
+			)
+
+			ch <- prometheus.MustNewConstMetric(
+				c.clusterDetectManagedEventsThreshold,
+				prometheus.GaugeValue,
+				float64(v.DetectManagedEventsThreshold),
+				v.Name,
+			)
+
+			ch <- prometheus.MustNewConstMetric(
+				c.clusterSecurityLevelForStorage,
+				prometheus.GaugeValue,
+				float64(v.SecurityLevelForStorage),
+				v.Name,
+			)
+
+			ch <- prometheus.MustNewConstMetric(
+				c.clusterMaxNumberOfNodes,
+				prometheus.GaugeValue,
+				float64(v.MaxNumberOfNodes),
+				v.Name,
+			)
+
+			ch <- prometheus.MustNewConstMetric(
+				c.clusterDetectedCloudPlatform,
+				prometheus.GaugeValue,
+				float64(v.DetectedCloudPlatform),
+				v.Name,
+			)
+		}
 	}
 
 	return nil

internal/collector/mscluster/mscluster_network.go

@@ -48,7 +48,7 @@ type msClusterNetwork struct {
 }
 
 func (c *Collector) buildNetwork() error {
-	networkMIQuery, err := mi.NewQuery("SELECT * FROM MSCluster_Network")
+	networkMIQuery, err := mi.NewQuery("SELECT Characteristics,Flags,Metric,Role,State FROM MSCluster_Network")
 	if err != nil {
 		return fmt.Errorf("failed to create WMI query: %w", err)
 	}

internal/collector/mscluster/mscluster_node.go

@@ -18,6 +18,7 @@ package mscluster
 import (
 	"fmt"
 
+	"github.com/Microsoft/hcsshim/osversion"
 	"github.com/prometheus-community/windows_exporter/internal/mi"
 	"github.com/prometheus-community/windows_exporter/internal/types"
 	"github.com/prometheus/client_golang/prometheus"
@@ -66,7 +67,14 @@ type msClusterNode struct {
 }
 
 func (c *Collector) buildNode() error {
-	nodeMIQuery, err := mi.NewQuery("SELECT * FROM MSCluster_Node")
+	buildNumber := osversion.Build()
+
+	wmiSelect := "BuildNumber,Characteristics,DynamicWeight,Flags,MajorVersion,MinorVersion,NeedsPreventQuorum,NodeDrainStatus,NodeHighestVersion,NodeLowestVersion,NodeWeight,State,StatusInformation"
+	if buildNumber >= osversion.LTSC2022 {
+		wmiSelect += ",DetectedCloudPlatform"
+	}
+
+	nodeMIQuery, err := mi.NewQuery(fmt.Sprintf("SELECT %s FROM MSCluster_Node", wmiSelect))
 	if err != nil {
 		return fmt.Errorf("failed to create WMI query: %w", err)
 	}

internal/collector/mscluster/mscluster_resource.go

@@ -74,7 +74,7 @@ type msClusterResource struct {
 }
 
 func (c *Collector) buildResource() error {
-	resourceMIQuery, err := mi.NewQuery("SELECT * FROM MSCluster_Resource")
+	resourceMIQuery, err := mi.NewQuery("SELECT Name,Type,OwnerGroup,OwnerNode,Characteristics,DeadlockTimeout,EmbeddedFailureAction,Flags,IsAlivePollInterval,LooksAlivePollInterval,MonitorProcessId,PendingTimeout,ResourceClass,RestartAction,RestartDelay,RestartPeriod,RestartThreshold,RetryPeriodOnFailure,State,Subclass FROM MSCluster_Resource")
 	if err != nil {
 		return fmt.Errorf("failed to create WMI query: %w", err)
 	}

internal/collector/mscluster/mscluster_resourcegroup.go

@@ -66,7 +66,7 @@ type msClusterResourceGroup struct {
 }
 
 func (c *Collector) buildResourceGroup() error {
-	resourceGroupMIQuery, err := mi.NewQuery("SELECT * FROM MSCluster_ResourceGroup")
+	resourceGroupMIQuery, err := mi.NewQuery("SELECT AutoFailbackType,Characteristics,ColdStartSetting,DefaultOwner,FailbackWindowEnd,FailbackWindowStart,FailoverPeriod,FailoverThreshold,Flags,GroupType,OwnerNode,Priority,ResiliencyPeriod,State FROM MSCluster_ResourceGroup")
 	if err != nil {
 		return fmt.Errorf("failed to create WMI query: %w", err)
 	}

internal/mi/callbacks.go

@@ -138,6 +138,10 @@ func (o *OperationUnmarshalCallbacks) InstanceResult(
 
 		element, err := instance.GetElement(miTag)
 		if err != nil {
+			if errors.Is(err, MI_RESULT_NO_SUCH_PROPERTY) {
+				continue
+			}
+
 			o.errCh <- fmt.Errorf("failed to get element %s: %w", miTag, err)
 
 			return 0

internal/types/const.go

@@ -16,10 +16,5 @@
 package types
 
 const (
-	BuildNumberWindowsServer2025 uint32 = 26100
-	BuildNumberWindowsServer2022 uint32 = 20348
-	BuildNumberWindowsServer2019 uint32 = 17763
-	BuildNumberWindowsServer2016 uint32 = 14393
-
 	Namespace = "windows"
 )