Merge pull request #362 from atc0005/i341-extend-metrics-for-check-da…

…tastore-plugin

Extend metrics for `check_vmware_datastore` plugin

cmd/check_vmware_datastore/main.go

@@ -11,6 +11,7 @@ import (
 	"context"
 	"errors"
 	"fmt"
+	"time"
 
 	"github.com/atc0005/go-nagios"
 	"github.com/vmware/govmomi/units"
@@ -23,6 +24,10 @@ import (
 
 func main() {
 
+	// Start the timer. We'll use this to emit the plugin runtime as a
+	// performance data metric.
+	pluginStart := time.Now()
+
 	// Set initial "state" as valid, adjust as we go.
 	var nagiosExitState = nagios.ExitState{
 		LastError:      nil,
@@ -157,11 +162,30 @@ func main() {
 	log.Debug().Msg("Successfully retrieved datastore by name")
 
 	log.Debug().Msg("Generating datastore usage summary")
-	dsUsage := vsphere.NewDatastoreUsageSummary(
+	dsUsage, dsUsageErr := vsphere.NewDatastoreUsageSummary(
+		ctx,
+		c.Client,
 		datastore,
 		cfg.DatastoreUsageCritical,
 		cfg.DatastoreUsageWarning,
 	)
+	if dsUsageErr != nil {
+		log.Error().Err(dsUsageErr).Msg(
+			"error generating datastore usage summary",
+		)
+
+		nagiosExitState.LastError = dsUsageErr
+		nagiosExitState.ServiceOutput = fmt.Sprintf(
+			"%s: Error generating summary for datastore %q",
+			nagios.StateCRITICALLabel,
+			cfg.DatastoreName,
+		)
+		nagiosExitState.ExitStatusCode = nagios.StateCRITICALExitCode
+
+		return
+	}
+
+	log.Debug().Msg("Successfully generated datastore usage summary")
 
 	log.Debug().
 		Str("datastore_name", datastore.Name).
@@ -174,27 +198,13 @@ func main() {
 		Int("datastore_warning_threshold", dsUsage.WarningThreshold).
 		Msg("Datastore usage summary")
 
-	log.Debug().Msg("Retrieving VMs for datastore")
-	dsVMs, dsVMsFetchErr := vsphere.GetVMsFromDatastore(ctx, c.Client, datastore, true)
-	if dsVMsFetchErr != nil {
-		log.Error().Err(dsFetchErr).Msg(
-			"error retrieving VirtualMachines from datastore",
-		)
-
-		nagiosExitState.LastError = dsVMsFetchErr
-		nagiosExitState.ServiceOutput = fmt.Sprintf(
-			"%s: Error retrieving VirtualMachines from datastore %q",
-			nagios.StateCRITICALLabel,
-			cfg.DatastoreName,
-		)
-		nagiosExitState.ExitStatusCode = nagios.StateCRITICALExitCode
-
-		return
-	}
-
 	log.Debug().Msg("Compiling Performance Data details")
 
 	pd := []nagios.PerformanceData{
+		{
+			Label: "time",
+			Value: fmt.Sprintf("%dms", time.Since(pluginStart).Milliseconds()),
+		},
 		{
 			Label:             "datastore_usage",
 			Value:             fmt.Sprintf("%.2f", dsUsage.StorageUsedPercent),
@@ -209,6 +219,18 @@ func main() {
 			Max:               fmt.Sprintf("%d", dsUsage.StorageTotal),
 			Min:               "0",
 		},
+		{
+			Label: "vms",
+			Value: fmt.Sprintf("%d", len(dsUsage.VMs)),
+		},
+		{
+			Label: "vms_powered_off",
+			Value: fmt.Sprintf("%d", dsUsage.VMs.NumVMsPoweredOff()),
+		},
+		{
+			Label: "vms_powered_on",
+			Value: fmt.Sprintf("%d", dsUsage.VMs.NumVMsPoweredOn()),
+		},
 	}
 
 	log.Debug().Msg("Evaluating datastore usage state")
@@ -231,7 +253,6 @@ func main() {
 
 		nagiosExitState.LongServiceOutput = vsphere.DatastoreUsageReport(
 			c.Client,
-			dsVMs,
 			dsUsage,
 		)
 
@@ -263,7 +284,6 @@ func main() {
 
 		nagiosExitState.LongServiceOutput = vsphere.DatastoreUsageReport(
 			c.Client,
-			dsVMs,
 			dsUsage,
 		)
 
@@ -288,7 +308,6 @@ func main() {
 
 		nagiosExitState.LongServiceOutput = vsphere.DatastoreUsageReport(
 			c.Client,
-			dsVMs,
 			dsUsage,
 		)
 

internal/vsphere/datastores.go

@@ -30,7 +30,33 @@ var ErrDatastoreUsageThresholdCrossed = errors.New("datastore usage exceeds spec
 // DatastoreIDToNameIndex maps a Datastore's ID value to its name.
 type DatastoreIDToNameIndex map[string]string
 
-// DatastoreUsageSummary tracks usage details for a specific Datastore
+// DatastoreVMs provides an overview of all (visible) VirtualMachines residing
+// on a specific Datastore.
+type DatastoreVMs []DatastoreVM
+
+// DatastoreVM is a summary of details for a VirtualMachine found on a
+// specific datastore.
+type DatastoreVM struct {
+
+	// Name is the display name of the VirtualMachine.
+	Name string
+
+	// VMSize is the human readable or formatted size of the VirtualMachine.
+	VMSize string
+
+	// DatastoreUsage is the human readable or formatted percentage of the
+	// Datastore space consumed by this VirtualMachine.
+	DatastoreUsage string
+
+	// PowerState tracks the current power state for a VirtualMachine.
+	PowerState types.VirtualMachinePowerState
+
+	// DatastoreMOID is the MOID or MoRef ID for the Datastore where this
+	// VirtualMachine resides.
+	DatastoreMOID types.ManagedObjectReference
+}
+
+// DatastoreUsageSummary tracks usage details for a specific Datastore.
 type DatastoreUsageSummary struct {
 	Datastore               mo.Datastore
 	StorageRemainingPercent float64
@@ -40,21 +66,67 @@ type DatastoreUsageSummary struct {
 	StorageRemaining        int64
 	CriticalThreshold       int
 	WarningThreshold        int
+	VMs                     DatastoreVMs
+}
+
+// DatastoreVMsSummary evaluates provided Datastore and collection of
+// VirtualMachines and provides a basic human readable / formatted summary of
+// VirtualMachine details.
+func DatastoreVMsSummary(ds mo.Datastore, vms []mo.VirtualMachine) DatastoreVMs {
+
+	datastoreVMs := make(DatastoreVMs, 0, len(vms))
+
+	for _, vm := range vms {
+
+		var vmStorageUsed int64
+		for _, usage := range vm.Storage.PerDatastoreUsage {
+			if usage.Datastore == ds.Reference() {
+				vmStorageUsed += usage.Committed + usage.Uncommitted
+			}
+		}
+
+		vmPercentOfDSUsed := float64(vmStorageUsed) / float64(ds.Summary.Capacity) * 100
+		dsVM := DatastoreVM{
+			Name:           vm.Name,
+			VMSize:         units.ByteSize(vmStorageUsed).String(),
+			DatastoreUsage: fmt.Sprintf("%2.2f%%", vmPercentOfDSUsed),
+			PowerState:     vm.Runtime.PowerState,
+		}
+
+		datastoreVMs = append(datastoreVMs, dsVM)
+
+	}
+
+	return datastoreVMs
+
 }
 
 // NewDatastoreUsageSummary receives a Datastore and generates summary
 // information used to determine if usage levels have crossed user-specified
 // thresholds.
-func NewDatastoreUsageSummary(ds mo.Datastore, criticalThreshold int, warningThreshold int) DatastoreUsageSummary {
+// func NewDatastoreUsageSummary(ds mo.Datastore, dsVMs []mo.VirtualMachine, criticalThreshold int, warningThreshold int) DatastoreUsageSummary {
+func NewDatastoreUsageSummary(
+	ctx context.Context,
+	c *vim25.Client,
+	ds mo.Datastore,
+	criticalThreshold int,
+	warningThreshold int,
+) (DatastoreUsageSummary, error) {
 
 	storageRemainingPercentage := float64(ds.Summary.FreeSpace) / float64(ds.Summary.Capacity) * 100
 	storageUsedPercentage := 100 - storageRemainingPercentage
 	storageRemaining := ds.Summary.FreeSpace
 	storageTotal := ds.Summary.Capacity
 	storageUsed := storageTotal - storageRemaining
 
+	dsVMs, err := GetVMsFromDatastore(ctx, c, ds, true)
+	if err != nil {
+		return DatastoreUsageSummary{}, err
+	}
+
 	dsUsage := DatastoreUsageSummary{
 		Datastore:               ds,
+		VMs:                     DatastoreVMsSummary(ds, dsVMs),
 		StorageRemainingPercent: storageRemainingPercentage,
 		StorageUsedPercent:      storageUsedPercentage,
 		StorageTotal:            storageTotal,
@@ -64,7 +136,7 @@ func NewDatastoreUsageSummary(ds mo.Datastore, criticalThreshold int, warningThr
 		WarningThreshold:        warningThreshold,
 	}
 
-	return dsUsage
+	return dsUsage, nil
 
 }
 
@@ -81,6 +153,26 @@ func (dus DatastoreUsageSummary) IsCriticalState() bool {
 	return dus.StorageUsedPercent >= float64(dus.CriticalThreshold)
 }
 
+// NumVMsPoweredOn indicates how many VirtualMachines on a specific Datastore
+// are powered on.
+func (dsVMs DatastoreVMs) NumVMsPoweredOn() int {
+
+	var numOn int
+	for _, vm := range dsVMs {
+		if vm.PowerState == types.VirtualMachinePowerStatePoweredOn {
+			numOn++
+		}
+	}
+
+	return numOn
+}
+
+// NumVMsPoweredOff indicates how many VirtualMachines on a specific Datastore
+// are powered off OR suspended.
+func (dsVMs DatastoreVMs) NumVMsPoweredOff() int {
+	return len(dsVMs) - dsVMs.NumVMsPoweredOn()
+}
+
 // GetDatastores accepts a context, a connected client and a boolean value
 // indicating whether a subset of properties per Datastore are retrieved. A
 // collection of Datastores with requested properties is returned. If
@@ -265,7 +357,6 @@ func DatastoreUsageOneLineCheckSummary(
 // the web UI or in the body of many notifications.
 func DatastoreUsageReport(
 	c *vim25.Client,
-	dsVMs []mo.VirtualMachine,
 	dsUsageSummary DatastoreUsageSummary,
 ) string {
 
@@ -306,22 +397,13 @@ func DatastoreUsageReport(
 		nagios.CheckOutputEOL,
 	)
 
-	for _, vm := range dsVMs {
-
-		var vmStorageUsed int64
-		for _, usage := range vm.Storage.PerDatastoreUsage {
-			if usage.Datastore == dsUsageSummary.Datastore.Reference() {
-				vmStorageUsed += usage.Committed + usage.Uncommitted
-			}
-		}
-
-		vmPercentOfDSUsed := float64(vmStorageUsed) / float64(dsUsageSummary.StorageTotal) * 100
+	for _, vm := range dsUsageSummary.VMs {
 		fmt.Fprintf(
 			&report,
-			"* %s [Size: %v, Datastore Usage: %2.2f%%]%s",
+			"* %s [Size: %s, Datastore Usage: %s]%s",
 			vm.Name,
-			units.ByteSize(vmStorageUsed),
-			vmPercentOfDSUsed,
+			vm.VMSize,
+			vm.DatastoreUsage,
 			nagios.CheckOutputEOL,
 		)
 	}