WIP docs for bpf metrics

cmd/tetra/metrics/print.go

@@ -57,6 +57,6 @@ func New() *cobra.Command {
 }
 
 func initMetrics(_ string, reg *prometheus.Registry, _ *slog.Logger) error {
-	metricsconfig.InitAllMetrics(reg)
+	metricsconfig.InitAllMetricsForDocs(reg)
 	return nil
 }

pkg/metrics/bpfmetric.go

@@ -3,7 +3,9 @@
 
 package metrics
 
-import "github.com/prometheus/client_golang/prometheus"
+import (
+	"github.com/prometheus/client_golang/prometheus"
+)
 
 // BPFMetric represents a metric read directly from a BPF map.
 // It's intended to be used in custom collectors. The interface doesn't provide
@@ -16,20 +18,46 @@ type BPFMetric interface {
 
 type bpfCounter struct {
 	desc *prometheus.Desc
+	prometheus.CounterOpts
+	VariableLabels []string
 }
 
 func NewBPFCounter(desc *prometheus.Desc) BPFMetric {
 	return &bpfCounter{desc: desc}
 }
 
+func NewBPFCounterFromOpts(opts prometheus.CounterOpts, labelNames []string) *bpfCounter {
+	return &bpfCounter{
+		CounterOpts:    opts,
+		VariableLabels: labelNames,
+	}
+}
+
 func (c *bpfCounter) Desc() *prometheus.Desc {
+	if c.desc == nil {
+		c.desc = prometheus.NewDesc(
+			prometheus.BuildFQName(c.Namespace, c.Subsystem, c.Name),
+			c.Help,
+			c.VariableLabels,
+			c.ConstLabels,
+		)
+	}
 	return c.desc
 }
 
 func (c *bpfCounter) MustMetric(value float64, labelValues ...string) prometheus.Metric {
 	return prometheus.MustNewConstMetric(c.desc, prometheus.CounterValue, value, labelValues...)
 }
 
+func (c *bpfCounter) ToProm() *prometheus.CounterVec {
+	return prometheus.NewCounterVec(prometheus.CounterOpts{
+		Namespace:   c.Namespace,
+		Name:        c.Name,
+		Help:        c.Help,
+		ConstLabels: c.ConstLabels,
+	}, c.VariableLabels)
+}
+
 type bpfGauge struct {
 	desc *prometheus.Desc
 }

pkg/metrics/consts/consts.go

@@ -6,3 +6,11 @@ package consts
 const MetricsNamespace = "tetragon"
 
 var KnownMetricLabelFilters = []string{"namespace", "workload", "pod", "binary"}
+
+var (
+	EmptyPolicyHookLabels = []string{"", ""}
+	EmptyProcessLabels    = []string{"", "", "", ""}
+
+	ExamplePolicyHookLabels = []string{"example-tracingpolicy", "example-kprobe"}
+	ExampleProcessLabels    = []string{"example-namespace", "example-workload", "example-pod", "example-binary"}
+)

pkg/metrics/eventmetrics/eventmetrics.go

@@ -4,8 +4,12 @@
 package eventmetrics
 
 import (
+	"fmt"
+	"slices"
+
 	"github.com/cilium/tetragon/api/v1/tetragon"
 	"github.com/cilium/tetragon/api/v1/tetragon/codegen/helpers"
+	"github.com/cilium/tetragon/pkg/api/ops"
 	"github.com/cilium/tetragon/pkg/api/processapi"
 	"github.com/cilium/tetragon/pkg/filters"
 	"github.com/cilium/tetragon/pkg/logger"
@@ -26,11 +30,17 @@ var (
 		Help:        "The total number of Tetragon events",
 		ConstLabels: nil,
 	}, []string{"type"})
-	MissedEvents = metrics.NewBPFCounter(prometheus.NewDesc(
-		prometheus.BuildFQName(consts.MetricsNamespace, "", "missed_events_total"),
-		"The total number of Tetragon events per type that are failed to sent from the kernel.",
-		[]string{"msg_op"}, nil,
-	))
+	// MissedEvents = metrics.NewBPFCounter(prometheus.NewDesc(
+	// 	prometheus.BuildFQName(consts.MetricsNamespace, "", "missed_events_total"),
+	// 	"The total number of Tetragon events per type that are failed to sent from the kernel.",
+	// 	[]string{"msg_op"}, nil,
+	// ))
+	MissedEvents = metrics.NewBPFCounterFromOpts(prometheus.CounterOpts{
+		Namespace:   consts.MetricsNamespace,
+		Name:        "flagsmissed_events_total_total",
+		Help:        "The total number of Tetragon events per type that are failed to sent from the kernel.",
+		ConstLabels: nil,
+	}, []string{"msg_op"})
 	FlagCount = prometheus.NewCounterVec(prometheus.CounterOpts{
 		Namespace:   consts.MetricsNamespace,
 		Name:        "flags_total",
@@ -63,11 +73,31 @@ func InitMetrics(registry *prometheus.Registry) {
 	registerMetrics(registry)
 
 	// Initialize metrics with labels
-	// TODO:
-	// events_total: list all possible event types, add example process metadata for docs
-	// missed_events_total: list all reasonable opcodes for docs
-	// flags_total: list all possible flags
-	// policy_events_total: add example policy and hook for docs
+	for ev := range tetragon.EventType_name {
+		if tetragon.EventType(ev) != tetragon.EventType_UNDEF && tetragon.EventType(ev) != tetragon.EventType_TEST {
+			EventsProcessed.WithLabelValues(slices.Concat([]string{tetragon.EventType(ev).String()}, consts.EmptyProcessLabels)...).Add(0)
+		}
+	}
+	for _, v := range exec.FlagStrings {
+		FlagCount.WithLabelValues(v).Add(0)
+	}
+	policyStats.WithLabelValues(slices.Concat(consts.EmptyPolicyHookLabels, consts.EmptyProcessLabels)...).Add(0)
+}
+
+func InitMetricsForDocs(registry *prometheus.Registry) {
+	InitMetrics(registry)
+	for ev := range tetragon.EventType_name {
+		if tetragon.EventType(ev) != tetragon.EventType_UNDEF && tetragon.EventType(ev) != tetragon.EventType_TEST {
+			EventsProcessed.WithLabelValues(slices.Concat([]string{tetragon.EventType(ev).String()}, consts.ExampleProcessLabels)...).Add(0)
+		}
+	}
+	registry.MustRegister(MissedEvents.ToProm())
+	for opcode := range ops.OpCodeStrings {
+		if opcode != ops.MsgOpUndef && opcode != ops.MsgOpTest {
+			MissedEvents.ToProm().WithLabelValues(fmt.Sprint(int32(opcode))).Add(0)
+		}
+	}
+	policyStats.WithLabelValues(slices.Concat(consts.ExamplePolicyHookLabels, consts.ExampleProcessLabels)...).Add(0)
 }
 
 func GetProcessInfo(process *tetragon.Process) (binary, pod, workload, namespace string) {

pkg/metrics/metricsconfig/initmetrics.go

@@ -60,3 +60,36 @@ func InitAllMetrics(registry *prometheus.Registry) {
 	registry.MustRegister(grpcmetrics.NewServerMetrics())
 	version.InitMetrics(registry)
 }
+
+func InitAllMetricsForDocs(registry *prometheus.Registry) {
+	errormetrics.InitMetrics(registry)
+	eventcachemetrics.InitMetrics(registry)
+	eventmetrics.InitMetricsForDocs(registry)
+	kprobemetrics.InitMetrics(registry)
+	mapmetrics.InitMetrics(registry)
+	opcodemetrics.InitMetrics(registry)
+	policyfiltermetrics.InitMetrics(registry)
+	processexecmetrics.InitMetrics(registry)
+	ringbufmetrics.InitMetrics(registry)
+	ringbufqueuemetrics.InitMetrics(registry)
+	syscallmetrics.InitMetrics(registry)
+	watchermetrics.InitMetrics(registry)
+	observer.InitMetrics(registry)
+	tracing.InitMetrics(registry)
+	ratelimitmetrics.InitMetrics(registry)
+	policystatemetrics.InitMetrics(registry)
+
+	// register BPF collectors
+	registry.MustRegister(mapmetrics.NewBPFCollector(
+		eventcache.NewBPFCollector(),
+		observer.NewBPFCollector(),
+		process.NewBPFCollector(),
+	))
+	// registry.MustRegister(eventmetrics.NewBPFCollector())
+
+	// register common third-party collectors
+	registry.MustRegister(collectors.NewGoCollector())
+	registry.MustRegister(collectors.NewProcessCollector(collectors.ProcessCollectorOpts{}))
+	registry.MustRegister(grpcmetrics.NewServerMetrics())
+	version.InitMetrics(registry)
+}