metrics: Add metrics label filter configuration

cmd/tetragon/flags.go

@@ -5,6 +5,7 @@ package main
 
 import (
 	"github.com/cilium/tetragon/pkg/logger"
+	"github.com/cilium/tetragon/pkg/metrics/config"
 	"github.com/cilium/tetragon/pkg/option"
 
 	"github.com/spf13/viper"
@@ -31,13 +32,14 @@ const (
 	keyEnableCiliumAPI        = "enable-cilium-api"
 	keyEnableProcessAncestors = "enable-process-ancestors"
 
-	keyMetricsServer     = "metrics-server"
-	keyServerAddress     = "server-address"
-	keyGopsAddr          = "gops-address"
-	keyEnableProcessCred = "enable-process-cred"
-	keyEnableProcessNs   = "enable-process-ns"
-	keyTracingPolicy     = "tracing-policy"
-	keyTracingPolicyDir  = "tracing-policy-dir"
+	keyMetricsServer      = "metrics-server"
+	keyMetricsLabelFilter = "metrics-label-filter"
+	keyServerAddress      = "server-address"
+	keyGopsAddr           = "gops-address"
+	keyEnableProcessCred  = "enable-process-cred"
+	keyEnableProcessNs    = "enable-process-ns"
+	keyTracingPolicy      = "tracing-policy"
+	keyTracingPolicyDir   = "tracing-policy-dir"
 
 	keyCpuProfile = "cpuprofile"
 	keyMemProfile = "memprofile"
@@ -113,6 +115,7 @@ func readAndSetFlags() {
 	option.Config.DataCacheSize = viper.GetInt(keyDataCacheSize)
 
 	option.Config.MetricsServer = viper.GetString(keyMetricsServer)
+	option.Config.MetricsLabelFilter = config.ParseMetricsLabelFilter(viper.GetString(keyMetricsLabelFilter))
 	option.Config.ServerAddress = viper.GetString(keyServerAddress)
 
 	option.Config.ExportFilename = viper.GetString(keyExportFilename)

docs/content/en/docs/reference/helm-chart.md

@@ -92,6 +92,7 @@ To use [the values available](#values), with `helm install` or `helm upgrade`, u
 | tetragon.processCacheSize | int | `65536` |  |
 | tetragon.prometheus.address | string | `""` | The address at which to expose metrics. Set it to "" to expose on all available interfaces. |
 | tetragon.prometheus.enabled | bool | `true` | Whether to enable exposing Tetragon metrics. |
+| tetragon.prometheus.metricsLabelFilter | string | `"namespace,workload,pod,binary"` | The labels to include with supporting metrics. The possible values are "namespace", "workload", "pod" and "binary". |
 | tetragon.prometheus.port | int | `2112` | The port at which to expose metrics. |
 | tetragon.prometheus.serviceMonitor.enabled | bool | `false` | Whether to create a 'ServiceMonitor' resource targeting the 'tetragon' pods. |
 | tetragon.prometheus.serviceMonitor.labelsOverride | object | `{}` | The set of labels to place on the 'ServiceMonitor' resource. |

install/kubernetes/README.md

@@ -75,6 +75,7 @@ Helm chart for Tetragon
 | tetragon.processCacheSize | int | `65536` |  |
 | tetragon.prometheus.address | string | `""` | The address at which to expose metrics. Set it to "" to expose on all available interfaces. |
 | tetragon.prometheus.enabled | bool | `true` | Whether to enable exposing Tetragon metrics. |
+| tetragon.prometheus.metricsLabelFilter | string | `"namespace,workload,pod,binary"` | The labels to include with supporting metrics. The possible values are "namespace", "workload", "pod" and "binary". |
 | tetragon.prometheus.port | int | `2112` | The port at which to expose metrics. |
 | tetragon.prometheus.serviceMonitor.enabled | bool | `false` | Whether to create a 'ServiceMonitor' resource targeting the 'tetragon' pods. |
 | tetragon.prometheus.serviceMonitor.labelsOverride | object | `{}` | The set of labels to place on the 'ServiceMonitor' resource. |

install/kubernetes/templates/tetragon_configmap.yaml

@@ -37,6 +37,9 @@ data:
 {{- else }}
   metrics-server: ""
 {{- end }}
+{{- if .Values.tetragon.prometheus.enabled }}
+  metrics-label-filter: {{ .Values.tetragon.prometheus.metricsLabelFilter }}
+{{- end }}
 {{- if .Values.tetragon.grpc.enabled }}
   server-address: {{ .Values.tetragon.grpc.address }}
 {{- else }}

install/kubernetes/values.yaml

@@ -126,6 +126,9 @@ tetragon:
     address: ""
     # -- The port at which to expose metrics.
     port: 2112
+    # -- The labels to include with supporting metrics.
+    # The possible values are "namespace", "workload", "pod" and "binary".
+    metricsLabelFilter: "namespace,workload,pod,binary"
     serviceMonitor:
       # -- Whether to create a 'ServiceMonitor' resource targeting the 'tetragon' pods.
       enabled: false

pkg/metrics/config/initmetrics.go

@@ -23,6 +23,8 @@ import (
 	grpcmetrics "github.com/grpc-ecosystem/go-grpc-middleware/providers/prometheus"
 	"github.com/prometheus/client_golang/prometheus"
 	"github.com/prometheus/client_golang/prometheus/collectors"
+
+	"strings"
 )
 
 func InitAllMetrics(registry *prometheus.Registry) {
@@ -47,3 +49,11 @@ func InitAllMetrics(registry *prometheus.Registry) {
 	registry.MustRegister(grpcmetrics.NewServerMetrics())
 	version.InitMetrics(registry)
 }
+
+func ParseMetricsLabelFilter(labels string) map[string]interface{} {
+	result := make(map[string]interface{})
+	for _, label := range strings.Split(labels, ",") {
+		result[label] = nil
+	}
+	return result
+}

pkg/metrics/consts/consts.go

@@ -4,3 +4,4 @@
 package consts
 
 var MetricsNamespace = "tetragon"
+var KnownMetricLabelFilters = []string{"namespace", "workload", "pod", "binary"}

pkg/metrics/eventmetrics/eventmetrics.go

@@ -20,12 +20,12 @@ import (
 )
 
 var (
-	EventsProcessed = metrics.NewCounterVecWithPod(prometheus.CounterOpts{
+	EventsProcessed = metrics.MustNewGranularCounter(prometheus.CounterOpts{
 		Namespace:   consts.MetricsNamespace,
 		Name:        "events_total",
 		Help:        "The total number of Tetragon events",
 		ConstLabels: nil,
-	}, []string{"type", "namespace", "workload", "pod", "binary"})
+	}, []string{"type"})
 	FlagCount = prometheus.NewCounterVec(prometheus.CounterOpts{
 		Namespace:   consts.MetricsNamespace,
 		Name:        "flags_total",
@@ -39,19 +39,19 @@ var (
 		ConstLabels: nil,
 	})
 
-	policyStats = metrics.NewCounterVecWithPod(prometheus.CounterOpts{
+	policyStats = metrics.MustNewGranularCounter(prometheus.CounterOpts{
 		Namespace:   consts.MetricsNamespace,
 		Name:        "policy_events_total",
 		Help:        "Policy events calls observed.",
 		ConstLabels: nil,
-	}, []string{"policy", "hook", "namespace", "workload", "pod", "binary"})
+	}, []string{"policy", "hook"})
 )
 
 func InitMetrics(registry *prometheus.Registry) {
-	registry.MustRegister(EventsProcessed)
+	registry.MustRegister(EventsProcessed.ToProm())
 	registry.MustRegister(FlagCount)
 	registry.MustRegister(NotifyOverflowedEvents)
-	registry.MustRegister(policyStats)
+	registry.MustRegister(policyStats.ToProm())
 }
 
 func GetProcessInfo(process *tetragon.Process) (binary, pod, workload, namespace string) {
@@ -93,10 +93,10 @@ func handleProcessedEvent(pInfo *tracingpolicy.PolicyInfo, processedEvent interf
 	default:
 		eventType = "unknown"
 	}
-	EventsProcessed.WithLabelValues(eventType, namespace, workload, pod, binary).Inc()
+	EventsProcessed.ToProm().WithLabelValues(metrics.FilterMetricLabels(eventType, namespace, workload, pod, binary)...).Inc()
 	if pInfo != nil && pInfo.Name != "" {
-		policyStats.
-			WithLabelValues(pInfo.Name, pInfo.Hook, namespace, workload, pod, binary).
+		policyStats.ToProm().
+			WithLabelValues(metrics.FilterMetricLabels(pInfo.Name, pInfo.Hook, namespace, workload, pod, binary)...).
 			Inc()
 	}
 }

pkg/metrics/eventmetrics/eventmetrics_test.go

@@ -15,7 +15,7 @@ import (
 )
 
 func TestHandleProcessedEvent(t *testing.T) {
-	assert.NoError(t, testutil.CollectAndCompare(EventsProcessed, strings.NewReader("")))
+	assert.NoError(t, testutil.CollectAndCompare(EventsProcessed.ToProm(), strings.NewReader("")))
 	handleProcessedEvent(nil, nil)
 	// empty process
 	handleProcessedEvent(nil, &tetragon.GetEventsResponse{Event: &tetragon.GetEventsResponse_ProcessKprobe{ProcessKprobe: &tetragon.ProcessKprobe{}}})
@@ -79,7 +79,7 @@ tetragon_events_total{binary="binary_c",namespace="namespace_c",pod="pod_c",type
 tetragon_events_total{binary="binary_e",namespace="",pod="",type="PROCESS_EXIT",workload=""} 1
 tetragon_events_total{binary="binary_e",namespace="namespace_e",pod="pod_e",type="PROCESS_EXIT",workload="workload_e"} 1
 `)
-	assert.NoError(t, testutil.CollectAndCompare(EventsProcessed, expected))
+	assert.NoError(t, testutil.CollectAndCompare(EventsProcessed.ToProm(), expected))
 }
 
 func TestHandleOriginalEvent(t *testing.T) {

pkg/metrics/metrics.go

@@ -4,11 +4,16 @@
 package metrics
 
 import (
+	"fmt"
 	"net/http"
 	"sync"
 	"time"
 
+	"golang.org/x/exp/slices"
+
 	"github.com/cilium/tetragon/pkg/logger"
+	"github.com/cilium/tetragon/pkg/metrics/consts"
+	"github.com/cilium/tetragon/pkg/option"
 	"github.com/cilium/tetragon/pkg/podhooks"
 	"github.com/prometheus/client_golang/prometheus"
 	"github.com/prometheus/client_golang/prometheus/promhttp"
@@ -27,6 +32,33 @@ var (
 	deleteDelay         = 1 * time.Minute
 )
 
+type GranularCounter struct {
+	counter     *prometheus.CounterVec
+	CounterOpts prometheus.CounterOpts
+	labels      []string
+	register    sync.Once
+}
+
+func MustNewGranularCounter(opts prometheus.CounterOpts, labels []string) *GranularCounter {
+	for _, label := range labels {
+		if slices.Contains(consts.KnownMetricLabelFilters, label) {
+			panic(fmt.Sprintf("labels passed to GranularCounter can't contain any of the following: %v. These labels are added by Tetragon.", consts.KnownMetricLabelFilters))
+		}
+	}
+	return &GranularCounter{
+		CounterOpts: opts,
+		labels:      append(labels, consts.KnownMetricLabelFilters...),
+	}
+}
+
+func (m *GranularCounter) ToProm() *prometheus.CounterVec {
+	m.register.Do(func() {
+		m.labels = FilterMetricLabels(m.labels...)
+		m.counter = NewCounterVecWithPod(m.CounterOpts, m.labels)
+	})
+	return m.counter
+}
+
 // NewCounterVecWithPod is a wrapper around prometheus.NewCounterVec that also registers the metric
 // to be cleaned up when a pod is deleted. It should be used only to register metrics that have
 // "pod" and "namespace" labels.
@@ -142,3 +174,20 @@ func EnableMetrics(address string) {
 	http.Handle("/metrics", promhttp.HandlerFor(reg, promhttp.HandlerOpts{Registry: reg}))
 	http.ListenAndServe(address, nil)
 }
+
+// The FilterMetricLabels func takes in string arguments and returns a slice of those strings omitting the labels it is not configured for.
+// IMPORTANT! The filtered metric labels must be passed last and in the exact order of consts.KnownMetricLabelFilters.
+func FilterMetricLabels(labels ...string) []string {
+	offset := len(labels) - len(consts.KnownMetricLabelFilters)
+	if offset < 0 {
+		logger.GetLogger().WithField("labels", labels).Debug("Not enough labels provided to metrics.FilterMetricLabels.")
+		return labels
+	}
+	result := labels[:offset]
+	for i, label := range consts.KnownMetricLabelFilters {
+		if _, ok := option.Config.MetricsLabelFilter[label]; ok {
+			result = append(result, labels[offset+i])
+		}
+	}
+	return result
+}

pkg/metrics/metrics_test.go

@@ -18,12 +18,41 @@ import (
 	"github.com/cilium/tetragon/pkg/metrics"
 	"github.com/cilium/tetragon/pkg/metrics/config"
 	"github.com/cilium/tetragon/pkg/metrics/eventmetrics"
+	"github.com/cilium/tetragon/pkg/option"
 )
 
 var sampleMsgGenericTracepointUnix = tracing.MsgGenericTracepointUnix{
 	PolicyName: "fake-policy",
 }
 
+func TestFilterMetricLabels(t *testing.T) {
+	option.Config.MetricsLabelFilter = map[string]interface{}{
+		"namespace": nil,
+		"workload":  nil,
+		"pod":       nil,
+		"binary":    nil,
+	}
+	assert.Equal(t, []string{"type", "namespace", "workspace", "pod", "binary"}, metrics.FilterMetricLabels("type", "namespace", "workspace", "pod", "binary"))
+	assert.Equal(t, []string{"syscall", "namespace", "workspace", "pod", "binary"}, metrics.FilterMetricLabels("syscall", "namespace", "workspace", "pod", "binary"))
+	assert.Equal(t, []string{"namespace", "workspace", "pod", "binary"}, metrics.FilterMetricLabels("namespace", "workspace", "pod", "binary"))
+
+	option.Config.MetricsLabelFilter = map[string]interface{}{
+		"namespace": nil,
+		"workload":  nil,
+	}
+	assert.Equal(t, []string{"type", "namespace", "workspace"}, metrics.FilterMetricLabels("type", "namespace", "workspace", "pod", "binary"))
+	assert.Equal(t, []string{"syscall", "namespace", "workspace"}, metrics.FilterMetricLabels("syscall", "namespace", "workspace", "pod", "binary"))
+	assert.Equal(t, []string{"namespace", "workspace"}, metrics.FilterMetricLabels("namespace", "workspace", "pod", "binary"))
+
+	option.Config.MetricsLabelFilter = map[string]interface{}{
+		"namespace": nil,
+		"workload":  nil,
+		"pod":       nil,
+		"binary":    nil,
+	}
+	assert.Equal(t, []string{"type", "syscall"}, metrics.FilterMetricLabels("type", "syscall"))
+}
+
 func TestPodDelete(t *testing.T) {
 	reg := metrics.GetRegistry()
 	config.InitAllMetrics(reg)

pkg/metrics/syscallmetrics/syscallmetrics.go

@@ -12,16 +12,16 @@ import (
 )
 
 var (
-	syscallStats = metrics.NewCounterVecWithPod(prometheus.CounterOpts{
+	syscallStats = metrics.MustNewGranularCounter(prometheus.CounterOpts{
 		Namespace:   consts.MetricsNamespace,
 		Name:        "syscalls_total",
 		Help:        "System calls observed.",
 		ConstLabels: nil,
-	}, []string{"syscall", "namespace", "workload", "pod", "binary"})
+	}, []string{"syscall"})
 )
 
 func InitMetrics(registry *prometheus.Registry) {
-	registry.MustRegister(syscallStats)
+	registry.MustRegister(syscallStats.ToProm())
 }
 
 func Handle(event interface{}) {
@@ -46,7 +46,9 @@ func Handle(event interface{}) {
 	}
 
 	if syscall != "" {
-		syscallStats.WithLabelValues(syscall, namespace, workload, pod, binary).Inc()
+		syscallStats.ToProm().
+			WithLabelValues(metrics.FilterMetricLabels(syscall, namespace, workload, pod, binary)...).
+			Inc()
 	}
 }
 

pkg/option/config.go

@@ -11,6 +11,7 @@ import (
 	"time"
 
 	"github.com/cilium/tetragon/pkg/logger"
+	"github.com/cilium/tetragon/pkg/metrics/consts"
 	"github.com/spf13/viper"
 )
 
@@ -46,10 +47,11 @@ type config struct {
 	ProcessCacheSize int
 	DataCacheSize    int
 
-	MetricsServer    string
-	ServerAddress    string
-	TracingPolicy    string
-	TracingPolicyDir string
+	MetricsServer      string
+	MetricsLabelFilter map[string]interface{}
+	ServerAddress      string
+	TracingPolicy      string
+	TracingPolicyDir   string
 
 	ExportFilename             string
 	ExportFileMaxSizeMB        int
@@ -93,6 +95,15 @@ var (
 
 		// LogOpts contains logger parameters
 		LogOpts: make(map[string]string),
+
+		// Default to logging metrics with the greatest granularity.
+		MetricsLabelFilter: func() map[string]interface{} {
+			result := make(map[string]interface{})
+			for _, label := range consts.KnownMetricLabelFilters {
+				result[label] = nil
+			}
+			return result
+		}(),
 	}
 )