fix: processLru migrated towards ProcessCacheTotal as separate metric

processLru migrated towards ProcessCacheTotal as separate metric

Signed-off-by: sadath-12 <sadathsadu2002@gmail.com>

pkg/metrics/eventcachemetrics/eventcachemetrics.go

@@ -61,22 +61,22 @@ func InitMetrics(registry *prometheus.Registry) {
 	registry.MustRegister(parentInfoErrors)
 }
 
-// Get a new handle on an processInfoErrors metric for an eventType
+// Get a new handle on a processInfoErrors metric for an eventType
 func ProcessInfoError(eventType string) prometheus.Counter {
 	return processInfoErrors.WithLabelValues(eventType)
 }
 
-// Get a new handle on an processInfoErrors metric for an eventType
+// Get a new handle on a podInfoErrors metric for an eventType
 func PodInfoError(eventType string) prometheus.Counter {
 	return podInfoErrors.WithLabelValues(eventType)
 }
 
-// Get a new handle on an processInfoErrors metric for an eventType
+// Get a new handle on an eventCacheErrorsTotal metric for an error
 func EventCacheError(err string) prometheus.Counter {
 	return eventCacheErrorsTotal.WithLabelValues(err)
 }
 
-// Get a new handle on the eventCacheRetriesTotal metric for an entryType
+// Get a new handle on an eventCacheRetriesTotal metric for an entryType
 func EventCacheRetries(entryType string) prometheus.Counter {
 	return eventCacheRetriesTotal.WithLabelValues(entryType)
 }

pkg/metrics/metricsconfig/initmetrics.go

@@ -19,7 +19,6 @@ import (
 	"github.com/cilium/tetragon/pkg/metrics/syscallmetrics"
 	"github.com/cilium/tetragon/pkg/metrics/watchermetrics"
 	"github.com/cilium/tetragon/pkg/observer"
-	"github.com/cilium/tetragon/pkg/process"
 	"github.com/cilium/tetragon/pkg/version"
 	grpcmetrics "github.com/grpc-ecosystem/go-grpc-middleware/providers/prometheus"
 	"github.com/prometheus/client_golang/prometheus"
@@ -46,7 +45,6 @@ func InitAllMetrics(registry *prometheus.Registry) {
 	// register BPF collectors
 	registry.MustRegister(mapmetrics.NewBPFCollector(
 		observer.NewBPFCollector(),
-		process.NewBPFCollector(),
 	))
 	registry.MustRegister(eventmetrics.NewBPFCollector())
 

pkg/process/metrics.go

@@ -4,28 +4,17 @@
 package process
 
 import (
-	"fmt"
-
-	"github.com/cilium/tetragon/pkg/metrics/mapmetrics"
+	"github.com/cilium/tetragon/pkg/metrics/consts"
 	"github.com/prometheus/client_golang/prometheus"
 )
 
-// bpfCollector implements prometheus.Collector. It collects metrics directly from BPF maps.
-type bpfCollector struct{}
-
-func NewBPFCollector() prometheus.Collector {
-	return &bpfCollector{}
-}
-
-func (c *bpfCollector) Describe(ch chan<- *prometheus.Desc) {
-	ch <- mapmetrics.MapSize.Desc()
-}
+var ProcessCacheTotal = prometheus.NewGauge(prometheus.GaugeOpts{
+	Namespace:   consts.MetricsNamespace,
+	Name:        "process_cache_size",
+	Help:        "The size of the process cache",
+	ConstLabels: nil,
+})
 
-func (c *bpfCollector) Collect(ch chan<- prometheus.Metric) {
-	if procCache != nil {
-		ch <- mapmetrics.MapSize.MustMetric(
-			float64(procCache.len()),
-			"processLru", fmt.Sprint(procCache.size),
-		)
-	}
+func InitMetrics(registry *prometheus.Registry) {
+	registry.MustRegister(ProcessCacheTotal)
 }

pkg/process/process.go

@@ -88,6 +88,7 @@ func InitCache(w watcher.K8sResourceWatcher, size int) error {
 func FreeCache() {
 	procCache.Purge()
 	procCache = nil
+	ProcessCacheTotal.Set(0)
 }
 
 // GetProcessCopy() duplicates tetragon.Process and returns it
@@ -462,6 +463,7 @@ func AddExecEvent(event *tetragonAPI.MsgExecveEventUnix) *ProcessInternal {
 	}
 
 	procCache.add(proc)
+	ProcessCacheTotal.Inc()
 	return proc
 }
 
@@ -485,6 +487,7 @@ func AddCloneEvent(event *tetragonAPI.MsgCloneEvent) error {
 
 	parent.RefInc()
 	procCache.add(proc)
+	ProcessCacheTotal.Inc()
 	return nil
 }