mscluster: fix cluster and network sub collectors (#1759)

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

internal/collector/mscluster/mscluster.go

@@ -8,25 +8,34 @@ import (
 	"log/slog"
 	"slices"
 	"strings"
+	"sync"
 
 	"github.com/alecthomas/kingpin/v2"
 	"github.com/prometheus-community/windows_exporter/internal/mi"
 	"github.com/prometheus/client_golang/prometheus"
 )
 
-const Name = "mscluster"
+const (
+	Name = "mscluster"
+
+	subCollectorCluster       = "cluster"
+	subCollectorNetwork       = "network"
+	subCollectorNode          = "node"
+	subCollectorResource      = "resource"
+	subCollectorResourceGroup = "resourcegroup"
+)
 
 type Config struct {
 	CollectorsEnabled []string `yaml:"collectors_enabled"`
 }
 
 var ConfigDefaults = Config{
 	CollectorsEnabled: []string{
-		"cluster",
-		"network",
-		"node",
-		"resource",
-		"resourcegroup",
+		subCollectorCluster,
+		subCollectorNetwork,
+		subCollectorNode,
+		subCollectorResource,
+		subCollectorResourceGroup,
 	},
 }
 
@@ -99,27 +108,39 @@ func (c *Collector) Build(_ *slog.Logger, miSession *mi.Session) error {
 
 	c.miSession = miSession
 
-	if slices.Contains(c.config.CollectorsEnabled, "cluster") {
-		c.buildCluster()
+	errs := make([]error, 0, 5)
+
+	if slices.Contains(c.config.CollectorsEnabled, subCollectorCluster) {
+		if err := c.buildCluster(); err != nil {
+			errs = append(errs, fmt.Errorf("failed to build cluster collector: %w", err))
+		}
 	}
 
-	if slices.Contains(c.config.CollectorsEnabled, "network") {
-		c.buildNetwork()
+	if slices.Contains(c.config.CollectorsEnabled, subCollectorNetwork) {
+		if err := c.buildNetwork(); err != nil {
+			errs = append(errs, fmt.Errorf("failed to build network collector: %w", err))
+		}
 	}
 
-	if slices.Contains(c.config.CollectorsEnabled, "node") {
-		c.buildNode()
+	if slices.Contains(c.config.CollectorsEnabled, subCollectorNode) {
+		if err := c.buildNode(); err != nil {
+			errs = append(errs, fmt.Errorf("failed to build node collector: %w", err))
+		}
 	}
 
-	if slices.Contains(c.config.CollectorsEnabled, "resource") {
-		c.buildResource()
+	if slices.Contains(c.config.CollectorsEnabled, subCollectorResource) {
+		if err := c.buildResource(); err != nil {
+			errs = append(errs, fmt.Errorf("failed to build resource collector: %w", err))
+		}
 	}
 
-	if slices.Contains(c.config.CollectorsEnabled, "resourcegroup") {
-		c.buildResourceGroup()
+	if slices.Contains(c.config.CollectorsEnabled, subCollectorResourceGroup) {
+		if err := c.buildResourceGroup(); err != nil {
+			errs = append(errs, fmt.Errorf("failed to build resource group collector: %w", err))
+		}
 	}
 
-	return nil
+	return errors.Join(errs...)
 }
 
 // Collect sends the metric values for each metric
@@ -129,40 +150,73 @@ func (c *Collector) Collect(ch chan<- prometheus.Metric) error {
 		return nil
 	}
 
-	var (
-		err       error
-		errs      []error
-		nodeNames []string
-	)
+	errCh := make(chan error, 5)
 
-	if slices.Contains(c.config.CollectorsEnabled, "cluster") {
-		if err = c.collectCluster(ch); err != nil {
-			errs = append(errs, fmt.Errorf("failed to collect cluster metrics: %w", err))
-		}
-	}
+	wg := sync.WaitGroup{}
+	wg.Add(5)
 
-	if slices.Contains(c.config.CollectorsEnabled, "network") {
-		if err = c.collectNetwork(ch); err != nil {
-			errs = append(errs, fmt.Errorf("failed to collect network metrics: %w", err))
-		}
-	}
+	go func() {
+		defer wg.Done()
 
-	if slices.Contains(c.config.CollectorsEnabled, "node") {
-		if nodeNames, err = c.collectNode(ch); err != nil {
-			errs = append(errs, fmt.Errorf("failed to collect node metrics: %w", err))
+		if slices.Contains(c.config.CollectorsEnabled, subCollectorCluster) {
+			if err := c.collectCluster(ch); err != nil {
+				errCh <- fmt.Errorf("failed to collect cluster metrics: %w", err)
+			}
 		}
-	}
+	}()
+
+	go func() {
+		defer wg.Done()
 
-	if slices.Contains(c.config.CollectorsEnabled, "resource") {
-		if err = c.collectResource(ch, nodeNames); err != nil {
-			errs = append(errs, fmt.Errorf("failed to collect resource metrics: %w", err))
+		if slices.Contains(c.config.CollectorsEnabled, subCollectorNetwork) {
+			if err := c.collectNetwork(ch); err != nil {
+				errCh <- fmt.Errorf("failed to collect network metrics: %w", err)
+			}
 		}
-	}
+	}()
+
+	go func() {
+		defer wg.Done()
 
-	if slices.Contains(c.config.CollectorsEnabled, "resourcegroup") {
-		if err = c.collectResourceGroup(ch, nodeNames); err != nil {
-			errs = append(errs, fmt.Errorf("failed to collect resource group metrics: %w", err))
+		nodeNames := make([]string, 0)
+
+		if slices.Contains(c.config.CollectorsEnabled, subCollectorNode) {
+			var err error
+
+			nodeNames, err = c.collectNode(ch)
+			if err != nil {
+				errCh <- fmt.Errorf("failed to collect node metrics: %w", err)
+			}
 		}
+
+		go func() {
+			defer wg.Done()
+
+			if slices.Contains(c.config.CollectorsEnabled, subCollectorResource) {
+				if err := c.collectResource(ch, nodeNames); err != nil {
+					errCh <- fmt.Errorf("failed to collect resource metrics: %w", err)
+				}
+			}
+		}()
+
+		go func() {
+			defer wg.Done()
+
+			if slices.Contains(c.config.CollectorsEnabled, subCollectorResourceGroup) {
+				if err := c.collectResourceGroup(ch, nodeNames); err != nil {
+					errCh <- fmt.Errorf("failed to collect resource group metrics: %w", err)
+				}
+			}
+		}()
+	}()
+
+	wg.Wait()
+	close(errCh)
+
+	errs := make([]error, 0, 5)
+
+	for err := range errCh {
+		errs = append(errs, err)
 	}
 
 	return errors.Join(errs...)

internal/collector/mscluster/mscluster_cluster.go

@@ -7,13 +7,14 @@ import (
 
 	"github.com/prometheus-community/windows_exporter/internal/mi"
 	"github.com/prometheus-community/windows_exporter/internal/types"
-	"github.com/prometheus-community/windows_exporter/internal/utils"
 	"github.com/prometheus/client_golang/prometheus"
 )
 
 const nameCluster = Name + "_cluster"
 
 type collectorCluster struct {
+	clusterMIQuery mi.Query
+
 	clusterAddEvictDelay                           *prometheus.Desc
 	clusterAdminAccessPoint                        *prometheus.Desc
 	clusterAutoAssignNodeSite                      *prometheus.Desc
@@ -177,7 +178,14 @@ type msClusterCluster struct {
 	WitnessRestartInterval                  uint `mi:"WitnessRestartInterval"`
 }
 
-func (c *Collector) buildCluster() {
+func (c *Collector) buildCluster() error {
+	clusterMIQuery, err := mi.NewQuery("SELECT * FROM MSCluster_Cluster")
+	if err != nil {
+		return fmt.Errorf("failed to create WMI query: %w", err)
+	}
+
+	c.clusterMIQuery = clusterMIQuery
+
 	c.clusterAddEvictDelay = prometheus.NewDesc(
 		prometheus.BuildFQName(types.Namespace, nameCluster, "add_evict_delay"),
 		"Provides access to the cluster's AddEvictDelay property, which is the number a seconds that a new node is delayed after an eviction of another node.",
@@ -640,11 +648,13 @@ func (c *Collector) buildCluster() {
 		[]string{"name"},
 		nil,
 	)
+
+	return nil
 }
 
 func (c *Collector) collectCluster(ch chan<- prometheus.Metric) error {
 	var dst []msClusterCluster
-	if err := c.miSession.Query(&dst, mi.NamespaceRootMSCluster, utils.Must(mi.NewQuery("SELECT * MSCluster_Cluster"))); err != nil {
+	if err := c.miSession.Query(&dst, mi.NamespaceRootMSCluster, c.clusterMIQuery); err != nil {
 		return fmt.Errorf("WMI query failed: %w", err)
 	}
 

internal/collector/mscluster/mscluster_network.go

@@ -7,13 +7,14 @@ import (
 
 	"github.com/prometheus-community/windows_exporter/internal/mi"
 	"github.com/prometheus-community/windows_exporter/internal/types"
-	"github.com/prometheus-community/windows_exporter/internal/utils"
 	"github.com/prometheus/client_golang/prometheus"
 )
 
 const nameNetwork = Name + "_network"
 
 type collectorNetwork struct {
+	networkMIQuery mi.Query
+
 	networkCharacteristics *prometheus.Desc
 	networkFlags           *prometheus.Desc
 	networkMetric          *prometheus.Desc
@@ -33,7 +34,14 @@ type msClusterNetwork struct {
 	State           uint `mi:"State"`
 }
 
-func (c *Collector) buildNetwork() {
+func (c *Collector) buildNetwork() error {
+	networkMIQuery, err := mi.NewQuery("SELECT * FROM MSCluster_Network")
+	if err != nil {
+		return fmt.Errorf("failed to create WMI query: %w", err)
+	}
+
+	c.networkMIQuery = networkMIQuery
+
 	c.networkCharacteristics = prometheus.NewDesc(
 		prometheus.BuildFQName(types.Namespace, nameNetwork, "characteristics"),
 		"Provides the characteristics of the network.",
@@ -64,14 +72,16 @@ func (c *Collector) buildNetwork() {
 		[]string{"name"},
 		nil,
 	)
+
+	return nil
 }
 
 // Collect sends the metric values for each metric
 // to the provided prometheus metric channel.
 func (c *Collector) collectNetwork(ch chan<- prometheus.Metric) error {
 	var dst []msClusterNetwork
 
-	if err := c.miSession.Query(&dst, mi.NamespaceRootMSCluster, utils.Must(mi.NewQuery("SELECT * MSCluster_Node"))); err != nil {
+	if err := c.miSession.Query(&dst, mi.NamespaceRootMSCluster, c.networkMIQuery); err != nil {
 		return fmt.Errorf("WMI query failed: %w", err)
 	}
 

internal/collector/mscluster/mscluster_node.go

@@ -7,13 +7,14 @@ import (
 
 	"github.com/prometheus-community/windows_exporter/internal/mi"
 	"github.com/prometheus-community/windows_exporter/internal/types"
-	"github.com/prometheus-community/windows_exporter/internal/utils"
 	"github.com/prometheus/client_golang/prometheus"
 )
 
 const nameNode = Name + "_node"
 
 type collectorNode struct {
+	nodeMIQuery mi.Query
+
 	nodeBuildNumber           *prometheus.Desc
 	nodeCharacteristics       *prometheus.Desc
 	nodeDetectedCloudPlatform *prometheus.Desc
@@ -51,7 +52,14 @@ type msClusterNode struct {
 	StatusInformation     uint `mi:"StatusInformation"`
 }
 
-func (c *Collector) buildNode() {
+func (c *Collector) buildNode() error {
+	nodeMIQuery, err := mi.NewQuery("SELECT * FROM MSCluster_Node")
+	if err != nil {
+		return fmt.Errorf("failed to create WMI query: %w", err)
+	}
+
+	c.nodeMIQuery = nodeMIQuery
+
 	c.nodeBuildNumber = prometheus.NewDesc(
 		prometheus.BuildFQName(types.Namespace, nameNode, "build_number"),
 		"Provides access to the node's BuildNumber property.",
@@ -136,14 +144,16 @@ func (c *Collector) buildNode() {
 		[]string{"name"},
 		nil,
 	)
+
+	return nil
 }
 
 // Collect sends the metric values for each metric
 // to the provided prometheus Metric channel.
 func (c *Collector) collectNode(ch chan<- prometheus.Metric) ([]string, error) {
 	var dst []msClusterNode
 
-	if err := c.miSession.Query(&dst, mi.NamespaceRootMSCluster, utils.Must(mi.NewQuery("SELECT * FROM MSCluster_Node"))); err != nil {
+	if err := c.miSession.Query(&dst, mi.NamespaceRootMSCluster, c.nodeMIQuery); err != nil {
 		return nil, fmt.Errorf("WMI query failed: %w", err)
 	}
 

internal/collector/mscluster/mscluster_resource.go

@@ -7,13 +7,14 @@ import (
 
 	"github.com/prometheus-community/windows_exporter/internal/mi"
 	"github.com/prometheus-community/windows_exporter/internal/types"
-	"github.com/prometheus-community/windows_exporter/internal/utils"
 	"github.com/prometheus/client_golang/prometheus"
 )
 
 const nameResource = Name + "_resource"
 
 type collectorResource struct {
+	resourceMIQuery mi.Query
+
 	resourceCharacteristics        *prometheus.Desc
 	resourceDeadlockTimeout        *prometheus.Desc
 	resourceEmbeddedFailureAction  *prometheus.Desc
@@ -59,7 +60,14 @@ type msClusterResource struct {
 	Subclass               uint `mi:"Subclass"`
 }
 
-func (c *Collector) buildResource() {
+func (c *Collector) buildResource() error {
+	resourceMIQuery, err := mi.NewQuery("SELECT * FROM MSCluster_Resource")
+	if err != nil {
+		return fmt.Errorf("failed to create WMI query: %w", err)
+	}
+
+	c.resourceMIQuery = resourceMIQuery
+
 	c.resourceCharacteristics = prometheus.NewDesc(
 		prometheus.BuildFQName(types.Namespace, nameResource, "characteristics"),
 		"Provides the characteristics of the object.",
@@ -168,14 +176,16 @@ func (c *Collector) buildResource() {
 		[]string{"type", "owner_group", "name"},
 		nil,
 	)
+
+	return nil
 }
 
 // Collect sends the metric values for each metric
 // to the provided prometheus Metric channel.
 func (c *Collector) collectResource(ch chan<- prometheus.Metric, nodeNames []string) error {
 	var dst []msClusterResource
 
-	if err := c.miSession.Query(&dst, mi.NamespaceRootMSCluster, utils.Must(mi.NewQuery("SELECT * FROM MSCluster_Resource"))); err != nil {
+	if err := c.miSession.Query(&dst, mi.NamespaceRootMSCluster, c.resourceMIQuery); err != nil {
 		return fmt.Errorf("WMI query failed: %w", err)
 	}