From 9e35cbab946bf8c2d7d001dcf5a3b39b74768e8a Mon Sep 17 00:00:00 2001
From: Anna Kapuscinska <anna@isovalent.com>
Date: Tue, 20 Aug 2024 21:36:20 +0100
Subject: [PATCH] Fix tetragon_process_cache_size metric

tetragon_process_cache_size metric was increased on every add() and never
decreased. This is incorrect - fix it, so that it's increased on add() only if
there was no LRU eviction and it's decreased on remove().

Signed-off-by: Anna Kapuscinska <anna@isovalent.com>
---
 pkg/process/cache.go   | 10 ++++++++--
 pkg/process/metrics.go |  4 ++--
 pkg/process/process.go |  5 +----
 3 files changed, 11 insertions(+), 8 deletions(-)

diff --git a/pkg/process/cache.go b/pkg/process/cache.go
index 38a16f6c347..2dfd9df7b21 100644
--- a/pkg/process/cache.go
+++ b/pkg/process/cache.go
@@ -121,8 +121,9 @@ func (pc *Cache) refInc(p *ProcessInternal) {
 	atomic.AddUint32(&p.refcnt, 1)
 }
 
-func (pc *Cache) Purge() {
+func (pc *Cache) purge() {
 	pc.stopChan <- true
+	processCacheTotal.Set(0)
 }
 
 func NewCache(
@@ -159,12 +160,17 @@ func (pc *Cache) get(processID string) (*ProcessInternal, error) {
 // clone or execve events
 func (pc *Cache) add(process *ProcessInternal) bool {
 	evicted := pc.cache.Add(process.process.ExecId, process)
+	if !evicted {
+		processCacheTotal.Inc()
+	}
 	return evicted
 }
 
 func (pc *Cache) remove(process *tetragon.Process) bool {
 	present := pc.cache.Remove(process.ExecId)
-	if !present {
+	if present {
+		processCacheTotal.Dec()
+	} else {
 		errormetrics.ErrorTotalInc(errormetrics.ProcessCacheMissOnRemove)
 	}
 	return present
diff --git a/pkg/process/metrics.go b/pkg/process/metrics.go
index 0998d191b7d..c10fd545cb6 100644
--- a/pkg/process/metrics.go
+++ b/pkg/process/metrics.go
@@ -9,7 +9,7 @@ import (
 	"github.com/prometheus/client_golang/prometheus"
 )
 
-var ProcessCacheTotal = prometheus.NewGauge(prometheus.GaugeOpts{
+var processCacheTotal = prometheus.NewGauge(prometheus.GaugeOpts{
 	Namespace:   consts.MetricsNamespace,
 	Name:        "process_cache_size",
 	Help:        "The size of the process cache",
@@ -47,6 +47,6 @@ func NewCacheCollector() prometheus.Collector {
 }
 
 func RegisterMetrics(group metrics.Group) {
-	group.MustRegister(ProcessCacheTotal)
+	group.MustRegister(processCacheTotal)
 	group.MustRegister(NewCacheCollector())
 }
diff --git a/pkg/process/process.go b/pkg/process/process.go
index d11fdbe774a..d2a4d469c82 100644
--- a/pkg/process/process.go
+++ b/pkg/process/process.go
@@ -82,9 +82,8 @@ func InitCache(w watcher.K8sResourceWatcher, size int) error {
 }
 
 func FreeCache() {
-	procCache.Purge()
+	procCache.purge()
 	procCache = nil
-	ProcessCacheTotal.Set(0)
 }
 
 // GetProcessCopy() duplicates tetragon.Process and returns it
@@ -488,7 +487,6 @@ func AddExecEvent(event *tetragonAPI.MsgExecveEventUnix) *ProcessInternal {
 	}
 
 	procCache.add(proc)
-	ProcessCacheTotal.Inc()
 	return proc
 }
 
@@ -512,7 +510,6 @@ func AddCloneEvent(event *tetragonAPI.MsgCloneEvent) error {
 
 	parent.RefInc()
 	procCache.add(proc)
-	ProcessCacheTotal.Inc()
 	return nil
 }