Horizontal pod autoscaling not working: `unable to get metrics for resource cpu: no metrics returned from heapster`

Question

I'm trying to create an horizontal pod autoscaling after installing Kubernetes with kubeadm.

The main symptom is that kubectl get hpa returns the CPU metric in the column TARGETS as "undefined":

$ kubectl get hpa
NAME        REFERENCE              TARGETS           MINPODS   MAXPODS   REPLICAS   AGE
fibonacci   Deployment/fibonacci   <unknown> / 50%   1         3         1          1h

On further investigation, it appears that hpa is trying to receive the CPU metric from Heapster - but on my configuration the cpu metric is being provided by cAdvisor.

I am making this assumption based on the output of kubectl describe hpa fibonacci:

Name:                           fibonacci
Namespace:                      default
Labels:                         <none>
Annotations:                        <none>
CreationTimestamp:                  Sun, 14 May 2017 18:08:53 +0000
Reference:                      Deployment/fibonacci
Metrics:                        ( current / target )
  resource cpu on pods  (as a percentage of request):   <unknown> / 50%
Min replicas:                       1
Max replicas:                       3
Events:
  FirstSeen LastSeen    Count   From                SubObjectPath   Type        Reason              Message
  --------- --------    -----   ----                -------------   --------    ------              -------
  1h        3s      148 horizontal-pod-autoscaler           Warning     FailedGetResourceMetric     unable to get metrics for resource cpu: no metrics returned from heapster
  1h        3s      148 horizontal-pod-autoscaler           Warning     FailedComputeMetricsReplicas    failed to get cpu utilization: unable to get metrics for resource cpu: no metrics returned from heapster

Why does hpa try to receive this metric from heapster instead of cAdvisor?

How can I fix this?

Please find below my deployment, along with the contents of /var/log/container/kube-controller-manager.log and the output of kubectl get pods --namespace=kube-system and kubectl describe pods

apiVersion: extensions/v1beta1
kind: Deployment
metadata:
  name: fibonacci
  labels:
    app: fibonacci
spec:
  template:
    metadata:
      labels:
        app: fibonacci
    spec:
      containers:
      - name: fibonacci
        image: oghma/fibonacci
        ports:
          - containerPort: 8088
        resources:
          requests:
            memory: "64Mi"
            cpu: "75m"
          limits:
            memory: "128Mi"
            cpu: "100m"

---
kind: Service
apiVersion: v1
metadata:
  name: fibonacci
spec:
  selector:
    app: fibonacci
  ports:
    - protocol: TCP
      port: 8088
      targetPort: 8088
  externalIPs: 
    - 192.168.66.103

---
apiVersion: autoscaling/v1
kind: HorizontalPodAutoscaler
metadata:
  name: fibonacci
spec:
  scaleTargetRef:
    apiVersion: apps/v1beta1
    kind: Deployment
    name: fibonacci
  minReplicas: 1
  maxReplicas: 3
  targetCPUUtilizationPercentage: 50

---

$ kubectl describe pods
Name:       fibonacci-1503002127-3k755
Namespace:  default
Node:       kubernetesnode1/192.168.66.101
Start Time: Sun, 14 May 2017 17:47:08 +0000
Labels:     app=fibonacci
        pod-template-hash=1503002127
Annotations:    kubernetes.io/created-by={"kind":"SerializedReference","apiVersion":"v1","reference":{"kind":"ReplicaSet","namespace":"default","name":"fibonacci-1503002127","uid":"59ea64bb-38cd-11e7-b345-fa163edb1ca...
Status:     Running
IP:     192.168.202.1
Controllers:    ReplicaSet/fibonacci-1503002127
Containers:
  fibonacci:
    Container ID:   docker://315375c6a978fd689f4ba61919c15f15035deb9139982844cefcd46092fbec14
    Image:      oghma/fibonacci
    Image ID:       docker://sha256:26f9b6b2c0073c766b472ec476fbcd2599969b6e5e7f564c3c0a03f8355ba9f6
    Port:       8088/TCP
    State:      Running
      Started:      Sun, 14 May 2017 17:47:16 +0000
    Ready:      True
    Restart Count:  0
    Limits:
      cpu:  100m
      memory:   128Mi
    Requests:
      cpu:      75m
      memory:       64Mi
    Environment:    <none>
    Mounts:
      /var/run/secrets/kubernetes.io/serviceaccount from default-token-45kp8 (ro)
Conditions:
  Type      Status
  Initialized   True 
  Ready     True 
  PodScheduled  True 
Volumes:
  default-token-45kp8:
    Type:   Secret (a volume populated by a Secret)
    SecretName: default-token-45kp8
    Optional:   false
QoS Class:  Burstable
Node-Selectors: <none>
Tolerations:    node.alpha.kubernetes.io/notReady=:Exists:NoExecute for 300s
        node.alpha.kubernetes.io/unreachable=:Exists:NoExecute for 300s
Events:     <none>

---

$ kubectl get pods --namespace=kube-system

NAME                                        READY     STATUS    RESTARTS   AGE
calico-etcd-k1g53                           1/1       Running   0          2h
calico-node-6n4gp                           2/2       Running   1          2h
calico-node-nhmz7                           2/2       Running   0          2h
calico-policy-controller-1324707180-65m78   1/1       Running   0          2h
etcd-kubernetesmaster                       1/1       Running   0          2h
heapster-1428305041-zjzd1                   1/1       Running   0          1h
kube-apiserver-kubernetesmaster             1/1       Running   0          2h
kube-controller-manager-kubernetesmaster    1/1       Running   0          2h
kube-dns-3913472980-gbg5h                   3/3       Running   0          2h
kube-proxy-1dt3c                            1/1       Running   0          2h
kube-proxy-tfhr9                            1/1       Running   0          2h
kube-scheduler-kubernetesmaster             1/1       Running   0          2h
monitoring-grafana-3975459543-9q189         1/1       Running   0          1h
monitoring-influxdb-3480804314-7bvr3        1/1       Running   0          1h

---

$ cat /var/log/container/kube-controller-manager.log

"log":"I0514 17:47:08.631314       1 event.go:217] Event(v1.ObjectReference{Kind:\"Deployment\", Namespace:\"default\", Name:\"fibonacci\", UID:\"59e980d9-38cd-11e7-b345-fa163edb1ca6\", APIVersion:\"extensions\", ResourceVersion:\"1303\", FieldPath:\"\"}): type: 'Normal' reason: 'ScalingReplicaSet' Scaled up replica set fibonacci-1503002127 to 1\n","stream":"stderr","time":"2017-05-14T17:47:08.63177467Z"}
{"log":"I0514 17:47:08.650662       1 event.go:217] Event(v1.ObjectReference{Kind:\"ReplicaSet\", Namespace:\"default\", Name:\"fibonacci-1503002127\", UID:\"59ea64bb-38cd-11e7-b345-fa163edb1ca6\", APIVersion:\"extensions\", ResourceVersion:\"1304\", FieldPath:\"\"}): type: 'Normal' reason: 'SuccessfulCreate' Created pod: fibonacci-1503002127-3k755\n","stream":"stderr","time":"2017-05-14T17:47:08.650826398Z"}
{"log":"E0514 17:49:00.873703       1 horizontal.go:201] failed to compute desired number of replicas based on listed metrics for Deployment/default/fibonacci: failed to get cpu utilization: unable to get metrics for resource cpu: failed to get pod resource metrics: the server could not find the requested resource (get services http:heapster:)\n","stream":"stderr","time":"2017-05-14T17:49:00.874034952Z"}
{"log":"E0514 17:49:30.884078       1 horizontal.go:201] failed to compute desired number of replicas based on listed metrics for Deployment/default/fibonacci: failed to get cpu utilization: unable to get metrics for resource cpu: failed to get pod resource metrics: the server could not find the requested resource (get services http:heapster:)\n","stream":"stderr","time":"2017-05-14T17:49:30.884546461Z"}
{"log":"E0514 17:50:00.896563       1 horizontal.go:201] failed to compute desired number of replicas based on listed metrics for Deployment/default/fibonacci: failed to get cpu utilization: unable to get metrics for resource cpu: failed to get pod resource metrics: the server could not find the requested resource (get services http:heapster:)\n","stream":"stderr","time":"2017-05-14T17:50:00.89688734Z"}
{"log":"E0514 17:50:30.906293       1 horizontal.go:201] failed to compute desired number of replicas based on listed metrics for Deployment/default/fibonacci: failed to get cpu utilization: unable to get metrics for resource cpu: failed to get pod resource metrics: the server could not find the requested resource (get services http:heapster:)\n","stream":"stderr","time":"2017-05-14T17:50:30.906825794Z"}
{"log":"E0514 17:51:00.915996       1 horizontal.go:201] failed to compute desired number of replicas based on listed metrics for Deployment/default/fibonacci: failed to get cpu utilization: unable to get metrics for resource cpu: failed to get pod resource metrics: the server could not find the requested resource (get services http:heapster:)\n","stream":"stderr","time":"2017-05-14T17:51:00.916348218Z"}
{"log":"E0514 17:51:30.926043       1 horizontal.go:201] failed to compute desired number of replicas based on listed metrics for Deployment/default/fibonacci: failed to get cpu utilization: unable to get metrics for resource cpu: failed to get pod resource metrics: the server could not find the requested resource (get services http:heapster:)\n","stream":"stderr","time":"2017-05-14T17:51:30.926367623Z"}
{"log":"E0514 17:52:00.936574       1 horizontal.go:201] failed to compute desired number of replicas based on listed metrics for Deployment/default/fibonacci: failed to get cpu utilization: unable to get metrics for resource cpu: failed to get pod resource metrics: the server could not find the requested resource (get services http:heapster:)\n","stream":"stderr","time":"2017-05-14T17:52:00.936903072Z"}
{"log":"E0514 17:52:30.944724       1 horizontal.go:201] failed to compute desired number of replicas based on listed metrics for Deployment/default/fibonacci: failed to get cpu utilization: unable to get metrics for resource cpu: failed to get pod resource metrics: the server could not find the requested resource (get services http:heapster:)\n","stream":"stderr","time":"2017-05-14T17:52:30.945120508Z"}
{"log":"E0514 17:53:00.954785       1 horizontal.go:201] failed to compute desired number of replicas based on listed metrics for Deployment/default/fibonacci: failed to get cpu utilization: unable to get metrics for resource cpu: failed to get pod resource metrics: the server could not find the requested resource (get services http:heapster:)\n","stream":"stderr","time":"2017-05-14T17:53:00.955126309Z"}
{"log":"E0514 17:53:30.970454       1 horizontal.go:201] failed to compute desired number of replicas based on listed metrics for Deployment/default/fibonacci: failed to get cpu utilization: unable to get metrics for resource cpu: failed to get pod resource metrics: the server could not find the requested resource (get services http:heapster:)\n","stream":"stderr","time":"2017-05-14T17:53:30.972996568Z"}
{"log":"E0514 17:54:00.980735       1 horizontal.go:201] failed to compute desired number of replicas based on listed metrics for Deployment/default/fibonacci: failed to get cpu utilization: unable to get metrics for resource cpu: failed to get pod resource metrics: the server could not find the requested resource (get services http:heapster:)\n","stream":"stderr","time":"2017-05-14T17:54:00.981098832Z"}
{"log":"E0514 17:54:30.993176       1 horizontal.go:201] failed to compute desired number of replicas based on listed metrics for Deployment/default/fibonacci: failed to get cpu utilization: unable to get metrics for resource cpu: failed to get pod resource metrics: the server could not find the requested resource (get services http:heapster:)\n","stream":"stderr","time":"2017-05-14T17:54:30.993538841Z"}
{"log":"E0514 17:55:01.002941       1 horizontal.go:201] failed to compute desired number of replicas based on listed metrics for Deployment/default/fibonacci: failed to get cpu utilization: unable to get metrics for resource cpu: failed to get pod resource metrics: the server could not find the requested resource (get services http:heapster:)\n","stream":"stderr","time":"2017-05-14T17:55:01.003265908Z"}
{"log":"W0514 17:55:06.511756       1 reflector.go:323] k8s.io/kubernetes/pkg/controller/garbagecollector/graph_builder.go:192: watch of \u003cnil\u003e ended with: etcdserver: mvcc: required revision has been compacted\n","stream":"stderr","time":"2017-05-14T17:55:06.511957851Z"}
{"log":"E0514 17:55:31.013415       1 horizontal.go:201] failed to compute desired number of replicas based on listed metrics for Deployment/default/fibonacci: failed to get cpu utilization: unable to get metrics for resource cpu: failed to get pod resource metrics: the server could not find the requested resource (get services http:heapster:)\n","stream":"stderr","time":"2017-05-14T17:55:31.013776243Z"}
{"log":"E0514 17:56:01.024507       1 horizontal.go:201] failed to compute desired number of replicas based on listed metrics for Deployment/default/fibonacci: failed to get cpu utilization: unable to get metrics for resource cpu: failed to get pod resource metrics: the server could not find the requested resource (get services http:heapster:)\n","stream":"stderr","time":"2017-05-14T17:56:01.0248332Z"}
{"log":"E0514 17:56:31.036191       1 horizontal.go:201] failed to compute desired number of replicas based on listed metrics for Deployment/default/fibonacci: failed to get cpu utilization: unable to get metrics for resource cpu: failed to get pod resource metrics: the server could not find the requested resource (get services http:heapster:)\n","stream":"stderr","time":"2017-05-14T17:56:31.036606698Z"}
{"log":"E0514 17:57:01.049277       1 horizontal.go:201] failed to compute desired number of replicas based on listed metrics for Deployment/default/fibonacci: failed to get cpu utilization: unable to get metrics for resource cpu: failed to get pod resource metrics: the server could not find the requested resource (get services http:heapster:)\n","stream":"stderr","time":"2017-05-14T17:57:01.049616359Z"}
{"log":"E0514 17:57:31.064104       1 horizontal.go:201] failed to compute desired number of replicas based on listed metrics for Deployment/default/fibonacci: failed to get cpu utilization: unable to get metrics for resource cpu: failed to get pod resource metrics: the server could not find the requested resource (get services http:heapster:)\n","stream":"stderr","time":"2017-05-14T17:57:31.064489485Z"}
{"log":"E0514 17:58:01.073988       1 horizontal.go:201] failed to compute desired number of replicas based on listed metrics for Deployment/default/fibonacci: failed to get cpu utilization: unable to get metrics for resource cpu: failed to get pod resource metrics: the server could not find the requested resource (get services http:heapster:)\n","stream":"stderr","time":"2017-05-14T17:58:01.074339488Z"}
{"log":"E0514 17:58:31.084511       1 horizontal.go:201] failed to compute desired number of replicas based on listed metrics for Deployment/default/fibonacci: failed to get cpu utilization: unable to get metrics for resource cpu: failed to get pod resource metrics: the server could not find the requested resource (get services http:heapster:)\n","stream":"stderr","time":"2017-05-14T17:58:31.084839352Z"}
{"log":"E0514 17:59:01.096507       1 horizontal.go:201] failed to compute desired number of replicas based on listed metrics for Deployment/default/fibonacci: failed to get cpu utilization: unable to get metrics for resource cpu: failed to get pod resource metrics: the server could not find the requested resource (get services http:heapster:)\n","stream":"stderr","time":"2017-05-14T17:59:01.096896254Z"}

Answer 1

If in your deployment you have more than one container, make sure that you have specified the resource limits in all of them.

Answer 2

I have seen this also at other apps: There seems to be a bug in the HPA API.

Solution can be to use a replication controller scaleref instead:

apiVersion: autoscaling/v1
kind: HorizontalPodAutoscaler
metadata:
  name: fibonacci
  namespace: ....
spec:
  scaleRef:
    kind: ReplicationController
    name: fibonacci
    subresource: scale
  minReplicas: 1
  maxReplicas: 3
  targetCPUUtilizationPercentage: 50

Untested, so might need some edit on scaleRef (You used scaleTargetRef)

Answer 3

In case if you are using GKE 1.9.x

There is some bug, One needs to disable auto-scaling first and then re-enable it. This will provide current value in place of unknown

Try updating to latest GKE available.

Answer 4

Tl;dr: If you're using AWS EKS and specifying .spec.templates.spec.containers.<resources|limits> didn't work, the problem might be that you don't have Kubernetes Metrics Server installed.

I ran into this problem with Kubernetes HPAs while using AWS EKS. While hunting for solutions, I ran into the command below and decided to run it to see if I had Metrics Server installed:

kubectl get pods -n kube-system

I didn't have it installed. And it turns out that AWS has this doc that states that by default Metrics Server isn't installed on EKS clusters. So I followed the steps the doc advised for installing the server:

- Deploy the Metrics Server with the following command:

    kubectl apply -f https://github.com/kubernetes-sigs/metrics-server/releases/latest/download/components.yaml

- Verify that the metrics-server deployment is running the desired number of pods with the following command.

    kubectl get deployment metrics-server -n kube-system

Output

NAME             READY   UP-TO-DATE   AVAILABLE   AGE
metrics-server   1/1     1 

That was the solution for me. Once Metric Server was on my cluster, I succeeded in creating HPAs that were able to get usage info about their target pods/resources.

PS: You can run kubectl get pods -n kube-system again too so as to confirm the installation.

PPS: HPA = horizontal pod autoscaler

Answer 5

You can remove the LIMITS from your deployments and try it. In my deployment, I used only REQUESTS for RESOURCES and it worked. If you see the Horizontal Pod Autoscaler (HPA) is working then later you can play with LIMITS as well. This discussion tells you that only using REQUESTS is sufficient to do the HPA.

Answer 6

there is an option to enable autscaling on cluster pool make sure to turn it on first.

and then apply your hpa, and don't forgot to set cpu, memory requests,limits on k8s controllers

one thing to note is if you have multiple containers on your pod, you should then to specify cpu, memory requests,limits for each container

Answer 7

I faced a similar issue, hope this helps:

1. make sure the ApiVersion of the HPA is correct as syntax changes slightly version to version
2. Do kubectl autoscale deploy -n --cpu-percent= --min= --max= --dry-run -o yaml

Now this will give you the exact syntax for the HPA in accordance with the ApiVersion of the cluster. Amend your helm hpa.yaml file as per the output and that should do the trick.