Perfmon ignore non existent counters

metricbeat/module/windows/perfmon/_meta/docs.asciidoc

@@ -6,8 +6,9 @@ performance counters.
 
 You must configure queries for the Windows performance counters that you wish
 to collect. The example below collects processor time and disk writes.
-With `format` you can set the output format for a specific counter. Possible values are
-`float` and `long`. If nothing is selected the default value is `float`.
+`ignore_non_existent_counters` ignores failures for non-existent counters without
+to interrupt the service. With `format` you can set the output format for a specific counter.
+Possible values are `float` and `long`. If nothing is selected the default value is `float`.
 With `instance_name`, you can specify the name of the instance. Use this setting when:
 - You want to use an instance name that is different from the computed name. For example, `Total` instead of `_Total`.
 - You specify a counter that has no instance. For example, `\TCPIP Performance Diagnostics\IPv4 NBLs/sec indicated without prevalidation`.
@@ -19,6 +20,7 @@ For wildcard queries this setting has no effect.
 - module: windows
   metricsets: ["perfmon"]
   period: 10s
+  perfmon.ignore_non_existent_counters: true
   perfmon.counters:
     - instance_label: "processor.name"
       instance_name: "Total"

metricbeat/module/windows/perfmon/pdh_integration_windows_test.go

@@ -85,11 +85,13 @@ func TestQuery(t *testing.T) {
 }
 
 func TestExistingCounter(t *testing.T) {
-	config := make([]CounterConfig, 1)
-	config[0].InstanceLabel = "processor.name"
-	config[0].MeasurementLabel = "processor.time.total.pct"
-	config[0].Query = processorTimeCounter
-	config[0].Format = "float"
+	config := PerfmonConfig{
+		CounterConfig: make([]CounterConfig, 1),
+	}
+	config.CounterConfig[0].InstanceLabel = "processor.name"
+	config.CounterConfig[0].MeasurementLabel = "processor.time.total.pct"
+	config.CounterConfig[0].Query = processorTimeCounter
+	config.CounterConfig[0].Format = "float"
 	handle, err := NewPerfmonReader(config)
 	if err != nil {
 		t.Fatal(err)
@@ -105,11 +107,13 @@ func TestExistingCounter(t *testing.T) {
 }
 
 func TestNonExistingCounter(t *testing.T) {
-	config := make([]CounterConfig, 1)
-	config[0].InstanceLabel = "processor.name"
-	config[0].MeasurementLabel = "processor.time.total.pct"
-	config[0].Query = "\\Processor Information(_Total)\\not existing counter"
-	config[0].Format = "float"
+	config := PerfmonConfig{
+		CounterConfig: make([]CounterConfig, 1),
+	}
+	config.CounterConfig[0].InstanceLabel = "processor.name"
+	config.CounterConfig[0].MeasurementLabel = "processor.time.total.pct"
+	config.CounterConfig[0].Query = "\\Processor Information(_Total)\\not existing counter"
+	config.CounterConfig[0].Format = "float"
 	handle, err := NewPerfmonReader(config)
 	if assert.Error(t, err) {
 		assert.EqualValues(t, PDH_CSTATUS_NO_COUNTER, errors.Cause(err))
@@ -121,12 +125,41 @@ func TestNonExistingCounter(t *testing.T) {
 	}
 }
 
+func TestIgnoreNonExistentCounter(t *testing.T) {
+	config := PerfmonConfig{
+		CounterConfig:    make([]CounterConfig, 1),
+		IgnoreNECounters: true,
+	}
+	config.CounterConfig[0].InstanceLabel = "processor.name"
+	config.CounterConfig[0].MeasurementLabel = "processor.time.total.pct"
+	config.CounterConfig[0].Query = "\\Processor Information(_Total)\\not existing counter"
+	config.CounterConfig[0].Format = "float"
+	handle, err := NewPerfmonReader(config)
+
+	values, err := handle.Read()
+
+	time.Sleep(time.Millisecond * 1000)
+
+	if assert.Error(t, err) {
+		assert.EqualValues(t, PDH_NO_DATA, errors.Cause(err))
+	}
+
+	if handle != nil {
+		err = handle.query.Close()
+		assert.NoError(t, err)
+	}
+
+	t.Log(values)
+}
+
 func TestNonExistingObject(t *testing.T) {
-	config := make([]CounterConfig, 1)
-	config[0].InstanceLabel = "processor.name"
-	config[0].MeasurementLabel = "processor.time.total.pct"
-	config[0].Query = "\\non existing object\\% Processor Performance"
-	config[0].Format = "float"
+	config := PerfmonConfig{
+		CounterConfig: make([]CounterConfig, 1),
+	}
+	config.CounterConfig[0].InstanceLabel = "processor.name"
+	config.CounterConfig[0].MeasurementLabel = "processor.time.total.pct"
+	config.CounterConfig[0].Query = "\\non existing object\\% Processor Performance"
+	config.CounterConfig[0].Format = "float"
 	handle, err := NewPerfmonReader(config)
 	if assert.Error(t, err) {
 		assert.EqualValues(t, PDH_CSTATUS_NO_OBJECT, errors.Cause(err))
@@ -253,11 +286,13 @@ func TestRawValues(t *testing.T) {
 }
 
 func TestWildcardQuery(t *testing.T) {
-	config := make([]CounterConfig, 1)
-	config[0].InstanceLabel = "processor.name"
-	config[0].MeasurementLabel = "processor.time.pct"
-	config[0].Query = `\Processor Information(*)\% Processor Time`
-	config[0].Format = "float"
+	config := PerfmonConfig{
+		CounterConfig: make([]CounterConfig, 1),
+	}
+	config.CounterConfig[0].InstanceLabel = "processor.name"
+	config.CounterConfig[0].MeasurementLabel = "processor.time.pct"
+	config.CounterConfig[0].Query = `\Processor Information(*)\% Processor Time`
+	config.CounterConfig[0].Format = "float"
 	handle, err := NewPerfmonReader(config)
 	if err != nil {
 		t.Fatal(err)

metricbeat/module/windows/perfmon/pdh_windows.go

@@ -10,6 +10,8 @@ import (
 	"unicode/utf16"
 	"unsafe"
 
+	"github.com/elastic/beats/libbeat/logp"
+
 	"github.com/joeshaw/multierror"
 	"github.com/pkg/errors"
 	"golang.org/x/sys/windows"
@@ -222,7 +224,7 @@ func (q *Query) AddCounter(counterPath string, format Format, instanceName strin
 
 	h, err := PdhAddCounter(q.handle, counterPath, 0)
 	if err != nil {
-		return errors.Wrapf(err, `failed to add counter (path="%v")`, counterPath)
+		return err
 	}
 
 	wildcard := wildcardRegexp.MatchString(counterPath)
@@ -314,7 +316,7 @@ type PerfmonReader struct {
 	executed      bool // Indicates if the query has been executed.
 }
 
-func NewPerfmonReader(config []CounterConfig) (*PerfmonReader, error) {
+func NewPerfmonReader(config PerfmonConfig) (*PerfmonReader, error) {
 	query, err := NewQuery("")
 	if err != nil {
 		return nil, err
@@ -326,7 +328,7 @@ func NewPerfmonReader(config []CounterConfig) (*PerfmonReader, error) {
 		measurement:   map[string]string{},
 	}
 
-	for _, counter := range config {
+	for _, counter := range config.CounterConfig {
 		var format Format
 		switch counter.Format {
 		case "float":
@@ -335,8 +337,14 @@ func NewPerfmonReader(config []CounterConfig) (*PerfmonReader, error) {
 			format = LongFormat
 		}
 		if err := query.AddCounter(counter.Query, format, counter.InstanceName); err != nil {
+			if config.IgnoreNECounters {
+				if err == PDH_CSTATUS_NO_COUNTER {
+					logp.Info(`ignore non existent counter (path="%v")`, counter.Query)
+					continue
+				}
+			}
 			query.Close()
-			return nil, err
+			return nil, errors.Wrapf(err, `failed to add counter (path="%v")`, counter.Query)
 		}
 
 		r.instanceLabel[counter.Query] = counter.InstanceLabel

metricbeat/module/windows/perfmon/perfmon.go

@@ -21,6 +21,11 @@ type CounterConfig struct {
 	Format           string `config:"format"`
 }
 
+type PerfmonConfig struct {
+	IgnoreNECounters bool            `config:"perfmon.ignore_non_existent_counters"`
+	CounterConfig    []CounterConfig `config:"perfmon.counters" validate:"required"`
+}
+
 func init() {
 	if err := mb.Registry.AddMetricSet("windows", "perfmon", New); err != nil {
 		panic(err)
@@ -36,9 +41,7 @@ type MetricSet struct {
 func New(base mb.BaseMetricSet) (mb.MetricSet, error) {
 	cfgwarn.Beta("The perfmon metricset is beta")
 
-	config := struct {
-		CounterConfig []CounterConfig `config:"perfmon.counters" validate:"required"`
-	}{}
+	config := PerfmonConfig{}
 
 	if err := base.Module().UnpackConfig(&config); err != nil {
 		return nil, err
@@ -57,7 +60,7 @@ func New(base mb.BaseMetricSet) (mb.MetricSet, error) {
 
 	}
 
-	reader, err := NewPerfmonReader(config.CounterConfig)
+	reader, err := NewPerfmonReader(config)
 	if err != nil {
 		return nil, errors.Wrap(err, "initialization failed")
 	}