Optimize pod cache (envoyproxy#6936)

* Optimize pod cache

Signed-off-by: jukie <10012479+Jukie@users.noreply.github.com>

* release note

Signed-off-by: jukie <10012479+Jukie@users.noreply.github.com>

* Remove retry

Signed-off-by: jukie <10012479+Jukie@users.noreply.github.com>

* cleanup

Signed-off-by: jukie <10012479+Jukie@users.noreply.github.com>

---------

Signed-off-by: jukie <10012479+Jukie@users.noreply.github.com>
Signed-off-by: Isaac <10012479+jukie@users.noreply.github.com>
Co-authored-by: zirain <zirain2009@gmail.com>

Author: jukie

internal/provider/kubernetes/kubernetes.go

@@ -10,6 +10,8 @@ import (
 	"fmt"
 	"time"
 
+	corev1 "k8s.io/api/core/v1"
+	"k8s.io/apimachinery/pkg/labels"
 	"k8s.io/client-go/rest"
 	"k8s.io/klog/v2"
 	"k8s.io/utils/ptr"
@@ -26,6 +28,7 @@ import (
 	egv1a1 "github.com/envoyproxy/gateway/api/v1alpha1"
 	"github.com/envoyproxy/gateway/internal/envoygateway"
 	ec "github.com/envoyproxy/gateway/internal/envoygateway/config"
+	"github.com/envoyproxy/gateway/internal/infrastructure/kubernetes/proxy"
 	"github.com/envoyproxy/gateway/internal/message"
 )
 
@@ -112,6 +115,16 @@ func New(ctx context.Context, restCfg *rest.Config, svrCfg *ec.Server, resources
 		mgrOpts.Cache.SyncPeriod = ptr.To(csp)
 	}
 
+	// Limit the cache to only Envoy proxy Pods to reduce memory and sync churn.
+	// ProxyTopologyInjector is the only component that interacts with Pods.
+	if mgrOpts.Cache.ByObject == nil {
+		mgrOpts.Cache.ByObject = map[client.Object]cache.ByObject{}
+	}
+
+	mgrOpts.Cache.ByObject[&corev1.Pod{}] = cache.ByObject{
+		Label: labels.SelectorFromSet(proxy.EnvoyAppLabel()),
+	}
+
 	if svrCfg.EnvoyGateway.NamespaceMode() {
 		mgrOpts.Cache.DefaultNamespaces = make(map[string]cache.Config)
 		for _, watchNS := range svrCfg.EnvoyGateway.Provider.Kubernetes.Watch.Namespaces {
@@ -134,9 +147,10 @@ func New(ctx context.Context, restCfg *rest.Config, svrCfg *ec.Server, resources
 	if svrCfg.EnvoyGateway.Provider.Kubernetes.TopologyInjector == nil || !ptr.Deref(svrCfg.EnvoyGateway.Provider.Kubernetes.TopologyInjector.Disable, false) {
 		mgr.GetWebhookServer().Register("/inject-pod-topology", &webhook.Admission{
 			Handler: &ProxyTopologyInjector{
-				Client:  mgr.GetClient(),
-				Logger:  svrCfg.Logger.WithName("proxy-topology-injector"),
-				Decoder: admission.NewDecoder(mgr.GetScheme()),
+				Client:    mgr.GetClient(),
+				APIReader: mgr.GetAPIReader(),
+				Logger:    svrCfg.Logger.WithName("proxy-topology-injector"),
+				Decoder:   admission.NewDecoder(mgr.GetScheme()),
 			},
 		})
 	}

internal/provider/kubernetes/topology_injector.go

@@ -22,9 +22,9 @@ import (
 
 type ProxyTopologyInjector struct {
 	client.Client
-	Decoder admission.Decoder
-
-	Logger logging.Logger
+	APIReader client.Reader
+	Decoder   admission.Decoder
+	Logger    logging.Logger
 }
 
 func (m *ProxyTopologyInjector) Handle(ctx context.Context, req admission.Request) admission.Response {
@@ -50,9 +50,13 @@ func (m *ProxyTopologyInjector) Handle(ctx context.Context, req admission.Reques
 
 	pod := &corev1.Pod{}
 	if err := m.Get(ctx, podName, pod); err != nil {
-		logger.Error(err, "get pod failed", "pod", podName.String())
-		topologyInjectorEventsTotal.WithFailure(metrics.ReasonError).Increment()
-		return admission.Allowed("internal error, skipped")
+		// Cache isn't guaranteed to be updated yet so if m.Get() fails
+		// try getting the pod from API server directly.
+		if err = m.APIReader.Get(ctx, podName, pod); err != nil {
+			logger.Error(err, "get pod failed", "pod", podName.String())
+			topologyInjectorEventsTotal.WithFailure(metrics.ReasonError).Increment()
+			return admission.Allowed("internal error, skipped")
+		}
 	}
 
 	// Skip non-proxy pods

release-notes/current.yaml

@@ -18,6 +18,7 @@ bug fixes: |
   Fixed the controller cannot read the EnvoyProxy attached gatewayclass only.
   Fixed indexer and controller crashing when BackendTrafficPolicy has a redirect response override.
   Fixed Lua validator log level to be suppressed by default.
+  Fixed ProxyTopologyInjector cache sync race condition that caused injection failures
 
 # Enhancements that improve performance.
 performance improvements: |