Exclude resources by inventory path in vsphere input (#6859)

plugins/inputs/vsphere/README.md

@@ -31,6 +31,7 @@ vm_metric_exclude = [ "*" ]
   ## VMs
   ## Typical VM metrics (if omitted or empty, all metrics are collected)
   # vm_include = [ "/*/vm/**"] # Inventory path to VMs to collect (by default all are collected)
+  # vm_exclude = [] # Inventory paths to exclude
   vm_metric_include = [
     "cpu.demand.average",
     "cpu.idle.summation",
@@ -73,6 +74,7 @@ vm_metric_exclude = [ "*" ]
   ## Hosts
   ## Typical host metrics (if omitted or empty, all metrics are collected)
   # host_include = [ "/*/host/**"] # Inventory path to hosts to collect (by default all are collected)
+  # host_exclude [] # Inventory paths to exclude
   host_metric_include = [
     "cpu.coreUtilization.average",
     "cpu.costop.summation",
@@ -130,18 +132,21 @@ vm_metric_exclude = [ "*" ]
 
   ## Clusters
   # cluster_include = [ "/*/host/**"] # Inventory path to clusters to collect (by default all are collected)
+  # cluster_exclude = [] # Inventory paths to exclude
   # cluster_metric_include = [] ## if omitted or empty, all metrics are collected
   # cluster_metric_exclude = [] ## Nothing excluded by default
   # cluster_instances = false ## false by default
 
   ## Datastores
   # cluster_include = [ "/*/datastore/**"] # Inventory path to datastores to collect (by default all are collected)
+  # cluster_exclude = [] # Inventory paths to exclude
   # datastore_metric_include = [] ## if omitted or empty, all metrics are collected
   # datastore_metric_exclude = [] ## Nothing excluded by default
   # datastore_instances = false ## false by default
 
   ## Datacenters
   # datacenter_include = [ "/*/host/**"] # Inventory path to clusters to collect (by default all are collected)
+  # datacenter_exclude = [] # Inventory paths to exclude
   datacenter_metric_include = [] ## if omitted or empty, all metrics are collected
   datacenter_metric_exclude = [ "*" ] ## Datacenters are not collected by default.
   # datacenter_instances = false ## false by default

plugins/inputs/vsphere/endpoint.go

@@ -32,8 +32,6 @@ var isIPv6 = regexp.MustCompile("^(?:[A-Fa-f0-9]{0,4}:){1,7}[A-Fa-f0-9]{1,4}$")
 
 const metricLookback = 3 // Number of time periods to look back at for non-realtime metrics
 
-const rtMetricLookback = 3 // Number of time periods to look back at for realtime metrics
-
 const maxSampleConst = 10 // Absolute maximim number of samples regardless of period
 
 const maxMetadataSamples = 100 // Number of resources to sample for metric metadata
@@ -67,6 +65,7 @@ type resourceKind struct {
 	objects          objectMap
 	filters          filter.Filter
 	paths            []string
+	excludePaths     []string
 	collectInstances bool
 	getObjects       func(context.Context, *Endpoint, *ResourceFilter) (objectMap, error)
 	include          []string
@@ -132,6 +131,7 @@ func NewEndpoint(ctx context.Context, parent *VSphere, url *url.URL) (*Endpoint,
 			objects:          make(objectMap),
 			filters:          newFilterOrPanic(parent.DatacenterMetricInclude, parent.DatacenterMetricExclude),
 			paths:            parent.DatacenterInclude,
+			excludePaths:     parent.DatacenterExclude,
 			simple:           isSimple(parent.DatacenterMetricInclude, parent.DatacenterMetricExclude),
 			include:          parent.DatacenterMetricInclude,
 			collectInstances: parent.DatacenterInstances,
@@ -149,6 +149,7 @@ func NewEndpoint(ctx context.Context, parent *VSphere, url *url.URL) (*Endpoint,
 			objects:          make(objectMap),
 			filters:          newFilterOrPanic(parent.ClusterMetricInclude, parent.ClusterMetricExclude),
 			paths:            parent.ClusterInclude,
+			excludePaths:     parent.ClusterExclude,
 			simple:           isSimple(parent.ClusterMetricInclude, parent.ClusterMetricExclude),
 			include:          parent.ClusterMetricInclude,
 			collectInstances: parent.ClusterInstances,
@@ -166,6 +167,7 @@ func NewEndpoint(ctx context.Context, parent *VSphere, url *url.URL) (*Endpoint,
 			objects:          make(objectMap),
 			filters:          newFilterOrPanic(parent.HostMetricInclude, parent.HostMetricExclude),
 			paths:            parent.HostInclude,
+			excludePaths:     parent.HostExclude,
 			simple:           isSimple(parent.HostMetricInclude, parent.HostMetricExclude),
 			include:          parent.HostMetricInclude,
 			collectInstances: parent.HostInstances,
@@ -183,6 +185,7 @@ func NewEndpoint(ctx context.Context, parent *VSphere, url *url.URL) (*Endpoint,
 			objects:          make(objectMap),
 			filters:          newFilterOrPanic(parent.VMMetricInclude, parent.VMMetricExclude),
 			paths:            parent.VMInclude,
+			excludePaths:     parent.VMExclude,
 			simple:           isSimple(parent.VMMetricInclude, parent.VMMetricExclude),
 			include:          parent.VMMetricInclude,
 			collectInstances: parent.VMInstances,
@@ -199,6 +202,7 @@ func NewEndpoint(ctx context.Context, parent *VSphere, url *url.URL) (*Endpoint,
 			objects:          make(objectMap),
 			filters:          newFilterOrPanic(parent.DatastoreMetricInclude, parent.DatastoreMetricExclude),
 			paths:            parent.DatastoreInclude,
+			excludePaths:     parent.DatastoreExclude,
 			simple:           isSimple(parent.DatastoreMetricInclude, parent.DatastoreMetricExclude),
 			include:          parent.DatastoreMetricInclude,
 			collectInstances: parent.DatastoreInstances,
@@ -329,11 +333,13 @@ func (e *Endpoint) getDatacenterName(ctx context.Context, client *Client, cache
 	path := make([]string, 0)
 	returnVal := ""
 	here := r
-	for {
+	done := false
+	for !done {
+		done = func() bool {
 			if name, ok := cache[here.Reference().String()]; ok {
 				// Populate cache for the entire chain of objects leading here.
 				returnVal = name
-			break
+				return true
 			}
 			path = append(path, here.Reference().String())
 			o := object.NewCommon(client.Client.Client, r)
@@ -343,18 +349,20 @@ func (e *Endpoint) getDatacenterName(ctx context.Context, client *Client, cache
 			err := o.Properties(ctx1, here, []string{"parent", "name"}, &result)
 			if err != nil {
 				e.Parent.Log.Warnf("Error while resolving parent. Assuming no parent exists. Error: %s", err.Error())
-			break
+				return true
 			}
 			if result.Reference().Type == "Datacenter" {
 				// Populate cache for the entire chain of objects leading here.
 				returnVal = result.Name
-			break
+				return true
 			}
 			if result.Parent == nil {
 				e.Parent.Log.Debugf("No parent found for %s (ascending from %s)", here.Reference(), r.Reference())
-			break
+				return true
 			}
 			here = result.Parent.Reference()
+			return false
+		}()
 	}
 	for _, s := range path {
 		cache[s] = returnVal
@@ -389,13 +397,16 @@ func (e *Endpoint) discover(ctx context.Context) error {
 	// Populate resource objects, and endpoint instance info.
 	newObjects := make(map[string]objectMap)
 	for k, res := range e.resourceKinds {
+		err := func() error {
 			e.Parent.Log.Debugf("Discovering resources for %s", res.name)
 			// Need to do this for all resource types even if they are not enabled
 			if res.enabled || k != "vm" {
 				rf := ResourceFilter{
 					finder:       &Finder{client},
 					resType:      res.vcName,
-				paths:   res.paths}
+					paths:        res.paths,
+					excludePaths: res.excludePaths,
+				}
 
 				ctx1, cancel1 := context.WithTimeout(ctx, e.Parent.Timeout.Duration)
 				defer cancel1()
@@ -427,14 +438,19 @@ func (e *Endpoint) discover(ctx context.Context) error {
 				SendInternalCounterWithTags("discovered_objects", e.URL.Host, map[string]string{"type": res.name}, int64(len(objects)))
 				numRes += int64(len(objects))
 			}
+			return nil
+		}()
+		if err != nil {
+			return err
+		}
 	}
 
 	// Build lun2ds map
 	dss := newObjects["datastore"]
 	l2d := make(map[string]string)
 	for _, ds := range dss {
-		url := ds.altID
+		lunId := ds.altID
-		m := isolateLUN.FindStringSubmatch(url)
+		m := isolateLUN.FindStringSubmatch(lunId)
 		if m != nil {
 			l2d[m[1]] = ds.name
 		}
@@ -583,6 +599,8 @@ func getClusters(ctx context.Context, e *Endpoint, filter *ResourceFilter) (obje
 	cache := make(map[string]*types.ManagedObjectReference)
 	m := make(objectMap, len(resources))
 	for _, r := range resources {
+		// Wrap in a function to make defer work correctly.
+		err := func() error {
 			// We're not interested in the immediate parent (a folder), but the data center.
 			p, ok := cache[r.Parent.Value]
 			if !ok {
@@ -590,7 +608,7 @@ func getClusters(ctx context.Context, e *Endpoint, filter *ResourceFilter) (obje
 				defer cancel2()
 				client, err := e.clientFactory.GetClient(ctx2)
 				if err != nil {
-				return nil, err
+					return err
 				}
 				o := object.NewFolder(client.Client.Client, *r.Parent)
 				var folder mo.Folder
@@ -612,10 +630,16 @@ func getClusters(ctx context.Context, e *Endpoint, filter *ResourceFilter) (obje
 				parentRef:    p,
 				customValues: e.loadCustomAttributes(&r.ManagedEntity),
 			}
+			return nil
+		}()
+		if err != nil {
+			return nil, err
+		}
 	}
 	return m, nil
 }
 
+//noinspection GoUnusedParameter
 func getHosts(ctx context.Context, e *Endpoint, filter *ResourceFilter) (objectMap, error) {
 	var resources []mo.HostSystem
 	err := filter.FindAll(ctx, &resources)
@@ -717,18 +741,18 @@ func getDatastores(ctx context.Context, e *Endpoint, filter *ResourceFilter) (ob
 	}
 	m := make(objectMap)
 	for _, r := range resources {
-		url := ""
+		lunId := ""
 		if r.Info != nil {
 			info := r.Info.GetDatastoreInfo()
 			if info != nil {
-				url = info.Url
+				lunId = info.Url
 			}
 		}
 		m[r.ExtensibleManagedObject.Reference().Value] = &objectRef{
 			name:         r.Name,
 			ref:          r.ExtensibleManagedObject.Reference(),
 			parentRef:    r.Parent,
-			altID:        url,
+			altID:        lunId,
 			customValues: e.loadCustomAttributes(&r.ManagedEntity),
 		}
 	}
@@ -814,7 +838,7 @@ func submitChunkJob(ctx context.Context, te *ThrottledExecutor, job func([]types
 	})
 }
 
-func (e *Endpoint) chunkify(ctx context.Context, res *resourceKind, now time.Time, latest time.Time, acc telegraf.Accumulator, job func([]types.PerfQuerySpec)) {
+func (e *Endpoint) chunkify(ctx context.Context, res *resourceKind, now time.Time, latest time.Time, job func([]types.PerfQuerySpec)) {
 	te := NewThrottledExecutor(e.Parent.CollectConcurrency)
 	maxMetrics := e.Parent.MaxQueryMetrics
 	if maxMetrics < 1 {
@@ -831,7 +855,7 @@ func (e *Endpoint) chunkify(ctx context.Context, res *resourceKind, now time.Tim
 	metrics := 0
 	total := 0
 	nRes := 0
-	for _, object := range res.objects {
+	for _, resource := range res.objects {
 		mr := len(res.metrics)
 		for mr > 0 {
 			mc := mr
@@ -841,14 +865,14 @@ func (e *Endpoint) chunkify(ctx context.Context, res *resourceKind, now time.Tim
 			}
 			fm := len(res.metrics) - mr
 			pq := types.PerfQuerySpec{
-				Entity:     object.ref,
+				Entity:     resource.ref,
 				MaxSample:  maxSampleConst,
 				MetricId:   res.metrics[fm : fm+mc],
 				IntervalId: res.sampling,
 				Format:     "normal",
 			}
 
-			start, ok := e.hwMarks.Get(object.ref.Value)
+			start, ok := e.hwMarks.Get(resource.ref.Value)
 			if !ok {
 				// Look back 3 sampling periods by default
 				start = latest.Add(time.Duration(-res.sampling) * time.Second * (metricLookback - 1))
@@ -917,7 +941,7 @@ func (e *Endpoint) collectResource(ctx context.Context, resourceType string, acc
 	// Estimate the interval at which we're invoked. Use local time (not server time)
 	// since this is about how we got invoked locally.
 	localNow := time.Now()
-	estInterval := time.Duration(time.Minute)
+	estInterval := time.Minute
 	if !res.lastColl.IsZero() {
 		s := time.Duration(res.sampling) * time.Second
 		rawInterval := localNow.Sub(res.lastColl)
@@ -957,13 +981,14 @@ func (e *Endpoint) collectResource(ctx context.Context, resourceType string, acc
 	latestSample := time.Time{}
 
 	// Divide workload into chunks and process them concurrently
-	e.chunkify(ctx, res, now, latest, acc,
+	e.chunkify(ctx, res, now, latest,
 		func(chunk []types.PerfQuerySpec) {
-			n, localLatest, err := e.collectChunk(ctx, chunk, res, acc, now, estInterval)
+			n, localLatest, err := e.collectChunk(ctx, chunk, res, acc, estInterval)
-			e.Parent.Log.Debugf("CollectChunk for %s returned %d metrics", resourceType, n)
 			if err != nil {
 				acc.AddError(errors.New("while collecting " + res.name + ": " + err.Error()))
+				return
 			}
+			e.Parent.Log.Debugf("CollectChunk for %s returned %d metrics", resourceType, n)
 			atomic.AddInt64(&count, int64(n))
 			tsMux.Lock()
 			defer tsMux.Unlock()
@@ -1004,7 +1029,7 @@ func alignSamples(info []types.PerfSampleInfo, values []int64, interval time.Dur
 		if roundedTs == lastBucket {
 			bi++
 			p := len(rValues) - 1
-			rValues[p] = ((bi-1)/bi)*float64(rValues[p]) + v/bi
+			rValues[p] = ((bi-1)/bi)*rValues[p] + v/bi
 		} else {
 			rValues = append(rValues, v)
 			roundedInfo := types.PerfSampleInfo{
@@ -1019,7 +1044,7 @@ func alignSamples(info []types.PerfSampleInfo, values []int64, interval time.Dur
 	return rInfo, rValues
 }
 
-func (e *Endpoint) collectChunk(ctx context.Context, pqs []types.PerfQuerySpec, res *resourceKind, acc telegraf.Accumulator, now time.Time, interval time.Duration) (int, time.Time, error) {
+func (e *Endpoint) collectChunk(ctx context.Context, pqs []types.PerfQuerySpec, res *resourceKind, acc telegraf.Accumulator, interval time.Duration) (int, time.Time, error) {
 	e.Parent.Log.Debugf("Query for %s has %d QuerySpecs", res.name, len(pqs))
 	latestSample := time.Time{}
 	count := 0
@@ -1100,7 +1125,7 @@ func (e *Endpoint) collectChunk(ctx context.Context, pqs []types.PerfQuerySpec,
 				}
 				v := alignedValues[idx]
 				if info.UnitInfo.GetElementDescription().Key == "percent" {
-					bucket.fields[fn] = float64(v) / 100.0
+					bucket.fields[fn] = v / 100.0
 				} else {
 					if e.Parent.UseIntSamples {
 						bucket.fields[fn] = int64(round(v))

plugins/inputs/vsphere/finder.go

@@ -28,31 +28,51 @@ type ResourceFilter struct {
 	finder       *Finder
 	resType      string
 	paths        []string
+	excludePaths []string
 }
 
 // FindAll returns the union of resources found given the supplied resource type and paths.
-func (f *Finder) FindAll(ctx context.Context, resType string, paths []string, dst interface{}) error {
+func (f *Finder) FindAll(ctx context.Context, resType string, paths, excludePaths []string, dst interface{}) error {
+	objs := make(map[string]types.ObjectContent)
 	for _, p := range paths {
-		if err := f.Find(ctx, resType, p, dst); err != nil {
+		if err := f.find(ctx, resType, p, objs); err != nil {
 			return err
 		}
 	}
-	return nil
+	if len(excludePaths) > 0 {
+		excludes := make(map[string]types.ObjectContent)
+		for _, p := range excludePaths {
+			if err := f.find(ctx, resType, p, excludes); err != nil {
+				return err
+			}
+		}
+		for k := range excludes {
+			delete(objs, k)
+		}
+	}
+	return objectContentToTypedArray(objs, dst)
 }
 
 // Find returns the resources matching the specified path.
 func (f *Finder) Find(ctx context.Context, resType, path string, dst interface{}) error {
+	objs := make(map[string]types.ObjectContent)
+	err := f.find(ctx, resType, path, objs)
+	if err != nil {
+		return err
+	}
+	return objectContentToTypedArray(objs, dst)
+}
+
+func (f *Finder) find(ctx context.Context, resType, path string, objs map[string]types.ObjectContent) error {
 	p := strings.Split(path, "/")
 	flt := make([]property.Filter, len(p)-1)
 	for i := 1; i < len(p); i++ {
 		flt[i-1] = property.Filter{"name": p[i]}
 	}
-	objs := make(map[string]types.ObjectContent)
 	err := f.descend(ctx, f.client.Client.ServiceContent.RootFolder, resType, flt, 0, objs)
 	if err != nil {
 		return err
 	}
-	objectContentToTypedArray(objs, dst)
 	f.client.log.Debugf("Find(%s, %s) returned %d objects", resType, path, len(objs))
 	return nil
 }
@@ -94,6 +114,9 @@ func (f *Finder) descend(ctx context.Context, root types.ManagedObjectReference,
 			// Special case: The last token is a recursive wildcard, so we can grab everything
 			// recursively in a single call.
 			v2, err := m.CreateContainerView(ctx, root, []string{resType}, true)
+			if err != nil {
+				return err
+			}
 			defer v2.Destroy(ctx)
 			err = v2.Retrieve(ctx, []string{resType}, fields, &content)
 			if err != nil {
@@ -204,7 +227,7 @@ func objectContentToTypedArray(objs map[string]types.ObjectContent, dst interfac
 // FindAll finds all resources matching the paths that were specified upon creation of
 // the ResourceFilter.
 func (r *ResourceFilter) FindAll(ctx context.Context, dst interface{}) error {
-	return r.finder.FindAll(ctx, r.resType, r.paths, dst)
+	return r.finder.FindAll(ctx, r.resType, r.paths, r.excludePaths, dst)
 }
 
 func matchName(f property.Filter, props []types.DynamicProperty) bool {

plugins/inputs/vsphere/vsphere.go

@@ -22,22 +22,27 @@ type VSphere struct {
 	DatacenterMetricInclude []string
 	DatacenterMetricExclude []string
 	DatacenterInclude       []string
+	DatacenterExclude       []string
 	ClusterInstances        bool
 	ClusterMetricInclude    []string
 	ClusterMetricExclude    []string
 	ClusterInclude          []string
+	ClusterExclude          []string
 	HostInstances           bool
 	HostMetricInclude       []string
 	HostMetricExclude       []string
 	HostInclude             []string
+	HostExclude             []string
 	VMInstances             bool     `toml:"vm_instances"`
 	VMMetricInclude         []string `toml:"vm_metric_include"`
 	VMMetricExclude         []string `toml:"vm_metric_exclude"`
 	VMInclude               []string `toml:"vm_include"`
+	VMExclude               []string `toml:"vm_exclude"`
 	DatastoreInstances      bool
 	DatastoreMetricInclude  []string
 	DatastoreMetricExclude  []string
 	DatastoreInclude        []string
+	DatastoreExclude        []string
 	Separator               string
 	CustomAttributeInclude  []string
 	CustomAttributeExclude  []string

plugins/inputs/vsphere/vsphere_test.go

@@ -377,10 +377,59 @@ func TestFinder(t *testing.T) {
 	testLookupVM(ctx, t, &f, "/*/host/**/*DC*/*/*DC*", 4, "")
 
 	vm = []mo.VirtualMachine{}
-	err = f.FindAll(ctx, "VirtualMachine", []string{"/DC0/vm/DC0_H0*", "/DC0/vm/DC0_C0*"}, &vm)
+	err = f.FindAll(ctx, "VirtualMachine", []string{"/DC0/vm/DC0_H0*", "/DC0/vm/DC0_C0*"}, []string{}, &vm)
 	require.NoError(t, err)
 	require.Equal(t, 4, len(vm))
 
+	rf := ResourceFilter{
+		finder:       &f,
+		paths:        []string{"/DC0/vm/DC0_H0*", "/DC0/vm/DC0_C0*"},
+		excludePaths: []string{"/DC0/vm/DC0_H0_VM0"},
+		resType:      "VirtualMachine",
+	}
+	vm = []mo.VirtualMachine{}
+	require.NoError(t, rf.FindAll(ctx, &vm))
+	require.Equal(t, 3, len(vm))
+
+	rf = ResourceFilter{
+		finder:       &f,
+		paths:        []string{"/DC0/vm/DC0_H0*", "/DC0/vm/DC0_C0*"},
+		excludePaths: []string{"/**"},
+		resType:      "VirtualMachine",
+	}
+	vm = []mo.VirtualMachine{}
+	require.NoError(t, rf.FindAll(ctx, &vm))
+	require.Equal(t, 0, len(vm))
+
+	rf = ResourceFilter{
+		finder:       &f,
+		paths:        []string{"/**"},
+		excludePaths: []string{"/**"},
+		resType:      "VirtualMachine",
+	}
+	vm = []mo.VirtualMachine{}
+	require.NoError(t, rf.FindAll(ctx, &vm))
+	require.Equal(t, 0, len(vm))
+
+	rf = ResourceFilter{
+		finder:       &f,
+		paths:        []string{"/**"},
+		excludePaths: []string{"/this won't match anything"},
+		resType:      "VirtualMachine",
+	}
+	vm = []mo.VirtualMachine{}
+	require.NoError(t, rf.FindAll(ctx, &vm))
+	require.Equal(t, 8, len(vm))
+
+	rf = ResourceFilter{
+		finder:       &f,
+		paths:        []string{"/**"},
+		excludePaths: []string{"/**/*VM0"},
+		resType:      "VirtualMachine",
+	}
+	vm = []mo.VirtualMachine{}
+	require.NoError(t, rf.FindAll(ctx, &vm))
+	require.Equal(t, 4, len(vm))
 }
 
 func TestFolders(t *testing.T) {