Add activator DaemonSet

config/activator-ds.yaml

@@ -0,0 +1 @@
+./core/activator-ds.yaml

config/core/activator-ds.yaml

@@ -0,0 +1,104 @@
+# Copyright 2020 The Knative Authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     https://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+apiVersion: apps/v1
+kind: DaemonSet
+metadata:
+  name: activator
+  namespace: knative-serving
+  labels:
+    knative.dev/release: devel
+spec:
+  updateStrategy:
+    type: RollingUpdate
+    rollingUpdate:
+      maxUnavailable: 50%
+  selector:
+    matchLabels:
+      app: activator
+      role: activator
+  template:
+    metadata:
+      labels:
+        app: activator
+        role: activator
+        knative.dev/release: devel
+    spec:
+      serviceAccountName: controller
+      containers:
+      - name: activator
+        # This is the Go import path for the binary that is containerized
+        # and substituted here.
+        image: knative.dev/serving/cmd/activator
+
+        resources:
+          requests:
+            cpu: 300m
+            memory: 60Mi
+          limits:
+            cpu: 500m
+            memory: 200Mi
+        env:
+        # Run Activator with GC collection when newly generated memory is 500%.
+        - name: GOGC
+          value: "500"
+        - name: POD_NAME
+          valueFrom:
+            fieldRef:
+              fieldPath: metadata.name
+        - name: POD_IP
+          valueFrom:
+            fieldRef:
+              fieldPath: status.podIP
+        - name: SYSTEM_NAMESPACE
+          valueFrom:
+            fieldRef:
+              fieldPath: metadata.namespace
+        - name: CONFIG_LOGGING_NAME
+          value: config-logging
+        - name: CONFIG_OBSERVABILITY_NAME
+          value: config-observability
+        # TODO(https://github.com/knative/pkg/pull/953): Remove stackdriver specific config
+        - name: METRICS_DOMAIN
+          value: knative.dev/internal/serving
+
+        securityContext:
+          allowPrivilegeEscalation: false
+
+        ports:
+        - name: metrics
+          containerPort: 9090
+        - name: profiling
+          containerPort: 8008
+        - name: http1
+          containerPort: 8012
+        - name: h2c
+          containerPort: 8013
+
+        readinessProbe: &probe
+          httpGet:
+            port: 8012
+            httpHeaders:
+            - name: k-kubelet-probe
+              value: "activator"
+        livenessProbe: *probe
+      # The activator (often) sits on the dataplane, and may proxy long (e.g.
+      # streaming, websockets) requests.  We give a long grace period for the
+      # activator to "lame duck" and drain outstanding requests before we
+      # forcibly terminate the pod (and outstanding connections).  This value
+      # should be at least as large as the upper bound on the Revision's
+      # timeoutSeconds property to avoid servicing events disrupting
+      # connections.
+      terminationGracePeriodSeconds: 300
+

config/core/deployments/activator-hpa.yaml

@@ -21,7 +21,7 @@ metadata:
     serving.knative.dev/release: devel
 spec:
     minReplicas: 1
-    maxReplicas: 20
+    maxReplicas: 1
     scaleTargetRef:
       apiVersion: apps/v1
       kind: Deployment

config/core/deployments/activator.yaml

@@ -44,11 +44,11 @@ spec:
         # https://github.com/knative/serving/issues/1625#issuecomment-511930023
         resources:
           requests:
-            cpu: 300m
+            cpu: 100m
             memory: 60Mi
           limits:
-            cpu: 1000m
-            memory: 600Mi
+            cpu: 300m
+            memory: 200Mi
 
         env:
         # Run Activator with GC collection when newly generated memory is 500%.

pkg/activator/net/revision_backends.go

@@ -251,7 +251,7 @@ func (rw *revisionWatcher) checkDests(dests sets.String) {
 		// precise load balancing in the throttler.
 		hs, noop, err := rw.probePodIPs(dests)
 		if err != nil {
-			rw.logger.Errorw("Failed probing", zap.Error(err))
+			rw.logger.With(zap.Error(err)).Warnf("Failed probing: %+v", dests)
 			// We dont want to return here as an error still affects health states.
 		}
 

test/e2e-common.sh

@@ -366,10 +366,6 @@ metadata:
 data:
   profiling.enable: "true"
 EOF
-
-  echo ">> Patching activator HPA"
-  # We set min replicas to 2 for testing multiple activator pods.
-  kubectl -n knative-serving patch hpa activator --patch '{"spec":{"minReplicas":2}}' || return 1
 }
 
 # Check if we should use --resolvabledomain.  In case the ingress only has

test/e2e/scale.go

@@ -110,11 +110,11 @@ func ScaleToWithin(t *testing.T, scale int, duration time.Duration, latencies La
 					},
 				}),
 				v1alpha1testing.WithRevisionTimeoutSeconds(10))
-
 			if err != nil {
 				t.Errorf("CreateLatestService() = %v", err)
 				return fmt.Errorf("CreateLatestService() failed: %w", err)
 			}
+
 			// Record the time it took to create the service.
 			latencies.Add("time-to-create", start)
 			names.Route = serviceresourcenames.Route(svc)
@@ -135,6 +135,7 @@ func ScaleToWithin(t *testing.T, scale int, duration time.Duration, latencies La
 				url = s.Status.URL.URL()
 				return v1a1test.IsServiceReady(s)
 			}, "ServiceUpdatedWithURL")
+
 			if err != nil {
 				t.Errorf("WaitForServiceState(w/ Domain) = %v", err)
 				return fmt.Errorf("WaitForServiceState(w/ Domain) failed: %w", err)