[CNM] Fix counters for failed connections telemetry (#34446)

Co-authored-by: hmahmood <[email protected]>
DataDog · Feb 26, 2025 · bd44a13 · bd44a13
1 parent 980a9e9
commit bd44a13
Show file tree

Hide file tree

Showing 2 changed files with 31 additions and 20 deletions.
diff --git a/pkg/network/tracer/connection/ebpf_tracer.go b/pkg/network/tracer/connection/ebpf_tracer.go
@@ -126,10 +126,12 @@ var EbpfTracerTelemetry = struct {
 type ebpfTracer struct {
 	m *ddebpf.Manager
 
-	conns          *maps.GenericMap[netebpf.ConnTuple, netebpf.ConnStats]
-	tcpStats       *maps.GenericMap[netebpf.ConnTuple, netebpf.TCPStats]
-	tcpRetransmits *maps.GenericMap[netebpf.ConnTuple, uint32]
-	config         *config.Config
+	conns                   *maps.GenericMap[netebpf.ConnTuple, netebpf.ConnStats]
+	tcpStats                *maps.GenericMap[netebpf.ConnTuple, netebpf.TCPStats]
+	tcpRetransmits          *maps.GenericMap[netebpf.ConnTuple, uint32]
+	ebpfTelemetryMap        *maps.GenericMap[uint32, netebpf.Telemetry]
+	tcpFailuresTelemetryMap *maps.GenericMap[int32, uint64]
+	config                  *config.Config
 
 	// tcp_close events
 	closeConsumer *tcpCloseConsumer
@@ -147,6 +149,8 @@ type ebpfTracer struct {
 	ebpfTracerType TracerType
 
 	ch *cookieHasher
+
+	lastTCPFailureTelemetry map[int32]uint64
 }
 
 // NewTracer creates a new tracer
@@ -199,8 +203,9 @@ func newEbpfTracer(config *config.Config, _ telemetryComponent.Component) (Trace
 	}
 
 	tr := &ebpfTracer{
-		removeTuple: &netebpf.ConnTuple{},
-		ch:          newCookieHasher(),
+		removeTuple:             &netebpf.ConnTuple{},
+		ch:                      newCookieHasher(),
+		lastTCPFailureTelemetry: make(map[int32]uint64),
 	}
 
 	connCloseEventHandler, err := initClosedConnEventHandler(config, tr.closedPerfCallback, connPool, extractor)
@@ -271,6 +276,16 @@ func newEbpfTracer(config *config.Config, _ telemetryComponent.Component) (Trace
 		return nil, fmt.Errorf("error retrieving the bpf %s map: %s", probes.TCPRetransmitsMap, err)
 	}
 
+	tr.ebpfTelemetryMap, err = maps.GetMap[uint32, netebpf.Telemetry](m.Manager, probes.TelemetryMap)
+	if err != nil {
+		log.Warnf("error retrieving telemetry map: %s", err)
+	}
+
+	tr.tcpFailuresTelemetryMap, err = maps.GetMap[int32, uint64](m.Manager, probes.TCPFailureTelemetry)
+	if err != nil {
+		log.Warnf("error retrieving tcp failure telemetry map: %s", err)
+	}
+
 	return tr, nil
 }
 
@@ -534,15 +549,13 @@ func (t *ebpfTracer) Remove(conn *network.ConnectionStats) error {
 }
 
 func (t *ebpfTracer) getEBPFTelemetry() *netebpf.Telemetry {
-	var zero uint32
-	mp, err := maps.GetMap[uint32, netebpf.Telemetry](t.m.Manager, probes.TelemetryMap)
-	if err != nil {
-		log.Warnf("error retrieving telemetry map: %s", err)
+	if t.ebpfTelemetryMap == nil {
 		return nil
 	}
 
+	var zero uint32
 	tm := &netebpf.Telemetry{}
-	if err := mp.Lookup(&zero, tm); err != nil {
+	if err := t.ebpfTelemetryMap.Lookup(&zero, tm); err != nil {
 		// This can happen if we haven't initialized the telemetry object yet
 		// so let's just use a trace log
 		if log.ShouldLog(log.TraceLvl) {
@@ -554,12 +567,11 @@ func (t *ebpfTracer) getEBPFTelemetry() *netebpf.Telemetry {
 }
 
 func (t *ebpfTracer) getTCPFailureTelemetry() map[int32]uint64 {
-	mp, err := maps.GetMap[int32, uint64](t.m.Manager, probes.TCPFailureTelemetry)
-	if err != nil {
-		log.Warnf("error retrieving tcp failure telemetry map: %s", err)
+	if t.tcpFailuresTelemetryMap == nil {
 		return nil
 	}
-	it := mp.IterateWithBatchSize(100)
+
+	it := t.tcpFailuresTelemetryMap.IterateWithBatchSize(100)
 	var key int32
 	var val uint64
 	result := make(map[int32]uint64)
@@ -569,7 +581,9 @@ func (t *ebpfTracer) getTCPFailureTelemetry() map[int32]uint64 {
 			log.Warnf("error retrieving tcp failure telemetry map: %s", err)
 			return nil
 		}
-		result[key] = val
+
+		result[key] = val - t.lastTCPFailureTelemetry[key]
+		t.lastTCPFailureTelemetry[key] = val
 	}
 	return result
 }

diff --git a/pkg/network/tracer/connection/ebpf_tracer_test.go b/pkg/network/tracer/connection/ebpf_tracer_test.go
@@ -13,16 +13,13 @@ import (
 
 	"github.com/stretchr/testify/require"
 
-	"github.com/DataDog/datadog-agent/pkg/ebpf/maps"
 	"github.com/DataDog/datadog-agent/pkg/network/config"
-	"github.com/DataDog/datadog-agent/pkg/network/ebpf/probes"
 )
 
 func TestFailedConnectionTelemetryMapLoads(t *testing.T) {
 	tr, err := newEbpfTracer(config.New(), nil)
 	require.NoError(t, err, "could not load tracer")
 	t.Cleanup(tr.Stop)
 
-	_, err = maps.GetMap[int32, uint64](tr.(*ebpfTracer).m.Manager, probes.TCPFailureTelemetry)
-	require.NoError(t, err, "error loading tcp failure telemetry map")
+	require.NotNil(t, tr.(*ebpfTracer).tcpFailuresTelemetryMap, "error loading tcp failure telemetry map")
 }