Enable scaling fluentd-gcp resources using ScalingPolicy.

See https://github.com/justinsb/scaler for more details about ScalingPolicy resource.

cluster/addons/fluentd-gcp/fluentd-gcp-ds.yaml

@@ -31,12 +31,6 @@ spec:
         env:
         - name: FLUENTD_ARGS
           value: --no-supervisor -q
-        resources:
-          limits:
-            memory: 300Mi
-          requests:
-            cpu: 100m
-            memory: 200Mi
         volumeMounts:
         - name: varlog
           mountPath: /var/log

cluster/addons/fluentd-gcp/scaler-deployment.yaml

@@ -0,0 +1,34 @@
+apiVersion: apps/v1
+kind: Deployment
+metadata:
+  name: fluentd-gcp-scaler
+  namespace: kube-system
+  labels:
+    k8s-app: fluentd-gcp-scaler
+    version: v0.1
+    addonmanager.kubernetes.io/mode: Reconcile
+spec:
+  selector:
+    matchLabels:
+      k8s-app: fluentd-gcp-scaler
+  template:
+    metadata:
+      labels:
+        k8s-app: fluentd-gcp-scaler
+    spec:
+      serviceAccountName: fluentd-gcp-scaler
+      containers:
+      - name: fluentd-gcp-scaler
+        image: gcr.io/google-containers/fluentd-gcp-scaler:0.1
+        command:
+          - /scaler.sh
+          - --ds-name=fluentd-gcp-v2.0.13
+          - --scaling-policy=fluentd-gcp-scaling-policy
+        env:
+        # Defaults, used if no overrides are found in fluentd-gcp-scaling-policy
+        - name: CPU_REQUEST
+          value: 100m
+        - name: MEMORY_REQUEST
+          value: 200Mi
+        - name: MEMORY_LIMIT
+          value: 300Mi

cluster/addons/fluentd-gcp/scaler-policy.yaml

@@ -0,0 +1,13 @@
+apiVersion: apiextensions.k8s.io/v1beta1
+kind: CustomResourceDefinition
+metadata:
+  name: scalingpolicies.scalingpolicy.kope.io
+  labels:
+    addonmanager.kubernetes.io/mode: Reconcile
+spec:
+  group: scalingpolicy.kope.io
+  version: v1alpha1
+  names:
+    kind: ScalingPolicy
+    plural: scalingpolicies
+  scope: Namespaced

cluster/addons/fluentd-gcp/scaler-rbac.yaml

@@ -0,0 +1,48 @@
+apiVersion: v1
+kind: ServiceAccount
+metadata:
+  name: fluentd-gcp-scaler
+  namespace: kube-system
+  labels:
+    kubernetes.io/cluster-service: "true"
+    addonmanager.kubernetes.io/mode: Reconcile
+---
+apiVersion: rbac.authorization.k8s.io/v1
+kind: Role
+metadata:
+  name: system:fluentd-gcp-scaler
+  namespace: kube-system
+  labels:
+    kubernetes.io/cluster-service: "true"
+    addonmanager.kubernetes.io/mode: Reconcile
+rules:
+- apiGroups:
+  - "extensions"
+  resources:
+  - daemonsets
+  verbs:
+  - get
+  - patch
+- apiGroups:
+  - "scalingpolicy.kope.io"
+  resources:
+  - scalingpolicies
+  verbs:
+  - get
+---
+apiVersion: rbac.authorization.k8s.io/v1
+kind: RoleBinding
+metadata:
+  name: fluentd-gcp-scaler-binding
+  namespace: kube-system
+  labels:
+    kubernetes.io/cluster-service: "true"
+    addonmanager.kubernetes.io/mode: Reconcile
+roleRef:
+  apiGroup: rbac.authorization.k8s.io
+  kind: Role
+  name: system:fluentd-gcp-scaler
+subjects:
+- kind: ServiceAccount
+  name: fluentd-gcp-scaler
+  namespace: kube-system

cluster/gce/config-default.sh

@@ -353,9 +353,9 @@ if [[ -n "${LOGROTATE_MAX_SIZE:-}" ]]; then
 fi
 
 # Fluentd requirements
-FLUENTD_GCP_MEMORY_LIMIT="${FLUENTD_GCP_MEMORY_LIMIT:-300Mi}"
+FLUENTD_GCP_MEMORY_LIMIT="${FLUENTD_GCP_MEMORY_LIMIT:-}"
-FLUENTD_GCP_CPU_REQUEST="${FLUENTD_GCP_CPU_REQUEST:-100m}"
+FLUENTD_GCP_CPU_REQUEST="${FLUENTD_GCP_CPU_REQUEST:-}"
-FLUENTD_GCP_MEMORY_REQUEST="${FLUENTD_GCP_MEMORY_REQUEST:-200Mi}"
+FLUENTD_GCP_MEMORY_REQUEST="${FLUENTD_GCP_MEMORY_REQUEST:-}"
 
 # Heapster requirements
 HEAPSTER_GCP_BASE_MEMORY="${HEAPSTER_GCP_BASE_MEMORY:-140Mi}"

cluster/gce/config-test.sh

@@ -388,9 +388,9 @@ if [[ -n "${LOGROTATE_MAX_SIZE:-}" ]]; then
 fi
 
 # Fluentd requirements
-FLUENTD_GCP_MEMORY_LIMIT="${FLUENTD_GCP_MEMORY_LIMIT:-300Mi}"
+FLUENTD_GCP_MEMORY_LIMIT="${FLUENTD_GCP_MEMORY_LIMIT:-}"
-FLUENTD_GCP_CPU_REQUEST="${FLUENTD_GCP_CPU_REQUEST:-100m}"
+FLUENTD_GCP_CPU_REQUEST="${FLUENTD_GCP_CPU_REQUEST:-}"
-FLUENTD_GCP_MEMORY_REQUEST="${FLUENTD_GCP_MEMORY_REQUEST:-200Mi}"
+FLUENTD_GCP_MEMORY_REQUEST="${FLUENTD_GCP_MEMORY_REQUEST:-}"
 
 # Heapster requirements
 HEAPSTER_GCP_BASE_MEMORY="${HEAPSTER_GCP_BASE_MEMORY:-140Mi}"

cluster/gce/gci/configure-helper.sh

@@ -1979,50 +1979,57 @@ function copy-manifests {
   chmod 644 "${dst_dir}"/*
 }
 
-# Fluentd manifest is modified using kubectl, which may not be available at
+# Fluentd resources are modified using ScalingPolicy CR, which may not be
-# this point. Run this as a background process.
+# available at this point. Run this as a background process.
 function wait-for-apiserver-and-update-fluentd {
-  local -r fluentd_gcp_yaml="${1}"
+  local any_overrides=false
-
-  local modifying_flags=""
   if [[ -n "${FLUENTD_GCP_MEMORY_LIMIT:-}" ]]; then
-    modifying_flags="${modifying_flags} --limits=memory=${FLUENTD_GCP_MEMORY_LIMIT}"
+    any_overrides=true
   fi
-  local request_resources=""
   if [[ -n "${FLUENTD_GCP_CPU_REQUEST:-}" ]]; then
-    request_resources="cpu=${FLUENTD_GCP_CPU_REQUEST}"
+    any_overrides=true
   fi
   if [[ -n "${FLUENTD_GCP_MEMORY_REQUEST:-}" ]]; then
-    if [[ -n "${request_resources}" ]]; then
+    any_overrides=true
-      request_resources="${request_resources},"
-    fi
-    request_resources="memory=${FLUENTD_GCP_MEMORY_REQUEST}"
   fi
-  if [[ -n "${request_resources}" ]]; then
+  if ! $any_overrides; then
-    modifying_flags="${modifying_flags} --requests=${request_resources}"
+    # Nothing to do here.
+    exit
   fi
 
-  until kubectl get nodes
+  # Wait until ScalingPolicy CRD is in place.
+  until kubectl get scalingpolicies
   do
     sleep 10
   done
 
-  local -r temp_fluentd_gcp_yaml="${fluentd_gcp_yaml}.tmp"
+  # Single-shot, not managed by addon manager. Can be later modified or removed
-  if kubectl set resources --dry-run --local -f ${fluentd_gcp_yaml} ${modifying_flags} \
+  # at will.
-      --containers=fluentd-gcp -o yaml > ${temp_fluentd_gcp_yaml}; then
+  cat <<EOF | kubectl apply -f -
-    mv ${temp_fluentd_gcp_yaml} ${fluentd_gcp_yaml}
+apiVersion: scalingpolicy.kope.io/v1alpha1
-  else
+kind: ScalingPolicy
-    (echo "Failed to update fluentd resources. Used manifest:" && cat ${temp_fluentd_gcp_yaml}) >&2
+metadata:
-    rm ${temp_fluentd_gcp_yaml}
+  name: fluentd-gcp-scaling-policy
-  fi
+  namespace: kube-system
+spec:
+  containers:
+  - name: fluentd-gcp
+    resources:
+      requests:
+      - resource: cpu
+        base: ${FLUENTD_GCP_CPU_REQUEST:-}
+      - resource: memory
+        base: ${FLUENTD_GCP_MEMORY_REQUEST:-}
+      limits:
+      - resource: memory
+        base: ${FLUENTD_GCP_MEMORY_LIMIT:-}
+EOF
 }
 
 # Trigger background process that will ultimately update fluentd resource
 # requirements.
 function start-fluentd-resource-update {
-  local -r fluentd_gcp_yaml="${1}"
+  wait-for-apiserver-and-update-fluentd &
-
-  wait-for-apiserver-and-update-fluentd ${fluentd_gcp_yaml} &
 }
 
 # Updates parameters in yaml file for prometheus-to-sd configuration, or