Do not trigger a watch of pods for status owner ref (#754)

* Do not trigger a watch of pods for status owner ref

Signed-off-by: Max Smythe <smythe@google.com>

* Run goimports

Signed-off-by: Max Smythe <smythe@google.com>

* Do not use cache for retrieving pod

Signed-off-by: Max Smythe <smythe@google.com>

* Unify getPod and add flag to toggle enabling fake get pod

Signed-off-by: Max Smythe <smythe@google.com>

* Fix lint errors

Signed-off-by: Max Smythe <smythe@google.com>

* Fix nil pointer error

Signed-off-by: Max Smythe <smythe@google.com>

pkg/controller/constraint/constraint_controller.go

@@ -317,14 +317,9 @@ func (r *ReconcileConstraint) Reconcile(request reconcile.Request) (reconcile.Re
 }
 
 func (r *ReconcileConstraint) defaultGetPod() (*corev1.Pod, error) {
-	ns := util.GetNamespace()
-	name := util.GetPodName()
-	key := types.NamespacedName{Namespace: ns, Name: name}
-	pod := &corev1.Pod{}
-	if err := r.reader.Get(context.TODO(), key, pod); err != nil {
-		return nil, err
-	}
-	return pod, nil
+	// require injection of GetPod in order to control what client we use to
+	// guarantee we don't inadvertently create a watch
+	panic("GetPod must be injected")
 }
 
 func (r *ReconcileConstraint) getOrCreatePodStatus(constraint *unstructured.Unstructured) (*constraintstatusv1beta1.ConstraintPodStatus, error) {

pkg/controller/constrainttemplate/constrainttemplate_controller.go

@@ -485,14 +485,9 @@ func (r *ReconcileConstraintTemplate) handleDelete(
 }
 
 func (r *ReconcileConstraintTemplate) defaultGetPod() (*corev1.Pod, error) {
-	ns := util.GetNamespace()
-	name := util.GetPodName()
-	key := types.NamespacedName{Namespace: ns, Name: name}
-	pod := &corev1.Pod{}
-	if err := r.Get(context.TODO(), key, pod); err != nil {
-		return nil, err
-	}
-	return pod, nil
+	// require injection of GetPod in order to control what client we use to
+	// guarantee we don't inadvertently create a watch
+	panic("GetPod must be injected")
 }
 
 func (r *ReconcileConstraintTemplate) deleteAllStatus(ctName string) error {

pkg/controller/controller.go

@@ -17,15 +17,29 @@ package controller
 
 import (
 	"context"
+	"flag"
+	"os"
+	"sync"
 
 	opa "github.com/open-policy-agent/frameworks/constraint/pkg/client"
+	podstatus "github.com/open-policy-agent/gatekeeper/apis/status/v1beta1"
 	"github.com/open-policy-agent/gatekeeper/pkg/controller/config/process"
 	"github.com/open-policy-agent/gatekeeper/pkg/readiness"
+	"github.com/open-policy-agent/gatekeeper/pkg/util"
 	"github.com/open-policy-agent/gatekeeper/pkg/watch"
 	corev1 "k8s.io/api/core/v1"
+	"k8s.io/apimachinery/pkg/apis/meta/v1/unstructured"
+	"k8s.io/apimachinery/pkg/runtime"
+	"k8s.io/apimachinery/pkg/types"
+	"sigs.k8s.io/controller-runtime/pkg/client"
+	"sigs.k8s.io/controller-runtime/pkg/client/apiutil"
 	"sigs.k8s.io/controller-runtime/pkg/manager"
 )
 
+var (
+	debugUseFakePod = flag.Bool("debug-use-fake-pod", false, "Use a fake pod name so the Gatekeeper executable can be run outside of Kubernetes")
+)
+
 type Injector interface {
 	InjectOpa(*opa.Client)
 	InjectWatchManager(*watch.Manager)
@@ -59,12 +73,73 @@ type Dependencies struct {
 	ProcessExcluder  *process.Excluder
 }
 
+type defaultPodGetter struct {
+	client client.Client
+	scheme *runtime.Scheme
+	pod    *corev1.Pod
+	mux    sync.RWMutex
+}
+
+func (g *defaultPodGetter) GetPod() (*corev1.Pod, error) {
+	pod := func() *corev1.Pod {
+		g.mux.RLock()
+		defer g.mux.RUnlock()
+		return g.pod
+	}()
+	if pod != nil {
+		return pod.DeepCopy(), nil
+	}
+	g.mux.Lock()
+	defer g.mux.Unlock()
+	// guard against the race condition where the pod has been retrieved
+	// between releasing the read lock and acquiring the write lock
+	if g.pod != nil {
+		return g.pod.DeepCopy(), nil
+	}
+	pod = &corev1.Pod{}
+	ns := util.GetNamespace()
+	name := util.GetPodName()
+	key := types.NamespacedName{Namespace: ns, Name: name}
+	// use unstructured to avoid inadvertently creating a watch on pods
+	uPod := &unstructured.Unstructured{}
+	gvk, err := apiutil.GVKForObject(pod, g.scheme)
+	if err != nil {
+		return nil, err
+	}
+	uPod.SetGroupVersionKind(gvk)
+	if err := g.client.Get(context.TODO(), key, uPod); err != nil {
+		return nil, err
+	}
+	if err := g.scheme.Convert(uPod, pod, nil); err != nil {
+		return nil, err
+	}
+	g.pod = pod
+	return pod.DeepCopy(), nil
+}
+
 // AddToManager adds all Controllers to the Manager
 func AddToManager(m manager.Manager, deps Dependencies) error {
 	// Reset cache on start - this is to allow for the future possibility that the OPA cache is stored remotely
 	if err := deps.Opa.Reset(context.Background()); err != nil {
 		return err
 	}
+	if deps.GetPod == nil {
+		podGetter := &defaultPodGetter{
+			scheme: m.GetScheme(),
+			client: m.GetClient(),
+		}
+		deps.GetPod = podGetter.GetPod
+	}
+	if *debugUseFakePod {
+		os.Setenv("POD_NAME", "no-pod")
+		podstatus.DisablePodOwnership()
+		fakePodGetter := func() (*corev1.Pod, error) {
+			pod := &corev1.Pod{}
+			pod.Name = util.GetPodName()
+			return pod, nil
+		}
+		deps.GetPod = fakePodGetter
+	}
 	for _, a := range Injectors {
 		a.InjectOpa(deps.Opa)
 		a.InjectWatchManager(deps.WatchManger)