Outshift Logo

PRODUCT

7 min read

Blog thumbnail
Published on 06/10/2018
Last updated on 03/21/2024

Vertical pod autoscaler

Share

At Banzai Cloud we provision all kinds of applications to Kubernetes and we try to autoscale these clusters with Pipeline and/or properly size application resources as needed. As promised in an earlier blog post, How to correctly size containers for Java 10 applications, we'll share our findings on the Vertical Pod Autoscaler(VPA) used with Java 10. VPA sets resource requests on pod containers automatically, based on historical usage, thus ensuring that pods are scheduled onto nodes where appropriate resource amounts are available for each pod.
Kubernetes supports three different kind of autoscalers - cluster, horizontal and vertical. This post is part of our autoscaling series: Autoscaling Kubernetes clusters > Vertical pod autoscaler > Horizontal pod autoscaler
For an overview of autoscaling flow please see this (static) diagram. For further information and a dynamic version of vertical autoscaling flow, read. vertical-pod-autoscaler

Prerequisites for using VPA

  • VPA requires MutatingAdmissionWebhooks to be enabled on the Kubernetes cluster. This can be verified quickly via:
    $ kubectl api-versions | grep
    admissionregistration admissionregistration.k8s.io/v1beta1
    As of Kubernetes version 1.9 MutatingAdmissionWebhooks is enabled by default. If your cluster doesn't have it enabled follow these instructions.
  • Install the components that comprise VPA by following this installation guide. If the VPA installation has been successful, you should see something like:
    $ kubectl get po -n
    kube-system NAME READY STATUS RESTARTS AGE ...
    vpa-admission-controller-7b449b69c-rrs5p 1/1 Running 0 1m
    vpa-recommender-bf6577cdd-zm7rf 1/1 Running 0 1m
    vpa-updater-5dd9968676-gm28g 1/1 Running 0 1m
    $ kubectl get
    crd NAME AGE
    verticalpodautoscalercheckpoints.poc.autoscaling.k8s.io 1m
    verticalpodautoscalers.poc.autoscaling.k8s.io 1m
    As stated in documentation, VPA pulls resource usage metrics related to pods and containers from Prometheus. VPA Recommender is the component that gathers metrics from Prometheus and makes recommendations for watched pods. In the current implementation, VPA Recommender expects the Prometheus Server to be reachable at a specific location: http://prometheus.monitoring.svc. For details see the Dockerfile of VPA Recommender. Since this is a work in progress, I expect it to be made configurable in the future.
Note: we doeffortless monitoring of Java applicationsdeployed to Kubernetes without code changes
As we can see **Prometheus Server** must be deployed to `monitoring` [namespace](https://kubernetes.io/docs/concepts/overview/working-with-objects/namespaces/) and there must be a [Kubernetes service](https://kubernetes.io/docs/concepts/services-networking/service/) named `prometheus` pointing to it.


```shell

$ helm init -c

```

```shell

$ helm repo list
  NAME                    URL
  stable                  https://kubernetes-charts.storage.googleapis.com

```

```shell

$ helm install --name prometheus  --namespace monitoring  stable/prometheus

```


```shell

kubectl create -f - <<EOF
apiVersion: v1
kind: Service
metadata:
  labels:
    app: prometheus
    chart: prometheus-6.6.1
    component: server
    heritage: Tiller
    release: prometheus
  name: prometheus
  namespace: monitoring
spec:
  ports:
  - name: http
    port: 80
    protocol: TCP
    targetPort: 9090
  selector:
    app: prometheus
    component: server
    release: prometheus
  sessionAffinity: None
  type: ClusterIP
  EOF
  
  ```

Configuring VPA

Once VPA is up and running, we need to configure it. A VPA configuration contains the following settings:
  1. label selector, through which it identifies the Pods it should handle
  2. optional update policy, configures how VPA applies resource related changes to Pods. If not specified, the default - Auto - is used.
  3. optional resource policy, configures how the recommender computes recommended resources for Pods. If not specified, the default is used.

Let's see these in action

For a dynamic overview of how the vertical cluster autoscaler works, please see the diagram below: vertical-pod-autoscaler We're going to use the same test application we did in How to correctly size containers for Java 10 applications. We deploy the test application using:
$ kubectl
create -f - <<EOF apiVersion: apps/v1 kind: Deployment
metadata: name: dyn-class-gen-deployment labels: app:
dyn-class-gen spec: replicas: 1 selector: matchLabels: app:
dyn-class-gen template: metadata: labels: app: dyn-class-gen
spec: containers: - name: dyn-class-gen-container image:
banzaicloud/dynclassgen:1.0 env: - name: DYN_CLASS_COUNT
value: "256" - name: MEM_USAGE_PER_OBJECT_MB value: "1"
resources: requests: memory: "64Mi" cpu: 1 limits: memory:
"1Gi" cpu: 2 EOF

$ deployment "dyn-class-gen-deployment" created
The container's upper memory limit is set to 1GB. The max heap size of the application will be automatically set to 1GB / 4 = 256MB. So, 256MB of max heap size is clearly not enough, since the application will try to consume 256 * 1MB of heap space, plus it needs space for internal objects in loaded libraries, etc. Thus we can expect to see the application quit due to java.lang.OutOfMemoryError.
$ kubectl get po NAME READY STATUS
RESTARTS AGE dyn-class-gen-deployment-5c75c8c555-gzcdq 0/1



```shell
 kubectl logs
dyn-class-gen-deployment-7f4f95b94b-cbrx6

... DynClassBase243 instance consuming 1MB DynClassBase244
instance consuming 1MB Exception in thread "main"
java.lang.OutOfMemoryError: Java heap space at
com.banzaicloud.dynclassgen.DynClassBase245.consumeSomeMemory(DynClassBase.java:25)
at
com.banzaicloud.dynclassgen.DynamicClassGen.main(DynamicClassGen.java:72)
Now let's see how VPA would handle our pod failing due to java.lang.OutOfMemoryError. We have to configure VPA first to find our pod.
$ kubectl create
-f - <<EOF apiVersion: poc.autoscaling.k8s.io/v1alpha1 kind:
VerticalPodAutoscaler metadata: name: dyn-class-gen-vpa
spec: selector: matchLabels: app: dyn-class-gen
updatePolicy: updateMode: "Auto" EOF

verticalpodautoscaler "dyn-class-gen-vpa" created
After waiting some time, then checking the logs of VPA Recommender, we can see that it doesn't provide any recommendations for our dyn-class-gen-vpa pod. My educated guess is that the pod is failing so quickly that Prometheus is unable to collect valuable data on resource usage from the pod, which means there is not enough input data for VPA Recommender to be able to come up with a recommendation. Let's modify the pod such as it's not failing with java.lang.OutOfMemoryError by increasing the upper limit of the heap to 300MB :
$ kubectl edit
deployment dyn-class-gen-deployment ... spec: containers:

- env: - name: DYN_CLASS_COUNT value: "256" - name: JVM_OPTS
  value: -Xmx300M - name: MEM_USAGE_PER_OBJECT_MB value: "1"
After letting our pod run a little longer, let's see what VPA Recommender tells us:
$ kubectl get
VerticalPodAutoscaler dyn-class-gen-vpa -o yaml

apiVersion: poc.autoscaling.k8s.io/v1alpha1 kind:
VerticalPodAutoscaler metadata: clusterName: ""
creationTimestamp: 2018-06-05T19:36:09Z generation: 0 name:
dyn-class-gen-vpa namespace: default resourceVersion:
"48550" selfLink:
/apis/poc.autoscaling.k8s.io/v1alpha1/namespaces/default/verticalpodautoscalers/dyn-class-gen-vpa
uid: b238081d-68f7-11e8-973e-42010a800fe7 spec: selector:
matchLabels: app: dyn-class-gen updatePolicy: updateMode:
Auto status: conditions:

- lastTransitionTime: 2018-06-05T19:36:22Z status: "True"
  type: Configured
- lastTransitionTime: 2018-06-05T19:36:22Z status: "True"
  type: RecommendationProvided lastUpdateTime:
  2018-06-06T06:26:43Z recommendation:
  containerRecommendations: - maxRecommended: cpu: 4806m
  memory: "12344993833" minRecommended: cpu: 241m memory:
  "619256043" name: dyn-class-gen-container target: cpu:
  250m memory: "642037204"
The VPA recommender recommends:
  • cpu: 250m
  • memory: "642037204" - aprox. 642Mi
for resource requests versus
  • cpu: 1
  • memory: "64Mi"
what we gave in the original deployment. In accordance with the official documentation, the values recommended by VPA Recommender will be applied to the pod by VPA Admission Controller upon the pod's creation. Thus, if we delete our pod, the Deployment will take care of spinning up a new one. The new one will have resources requests set by VPA Admission Controller, instead of inheriting values from the Deployment.
$ kubectl delete po
dyn-class-gen-deployment-7db4f5c557-l97w9
$ kubectl describe
po dyn-class-gen-deployment-7db4f5c557-pd9bc

Name: dyn-class-gen-deployment-7db4f5c557-pd9bc Namespace:
default Node:
gke-gkecluster-seba-636-pool1-f8f0d428-6n1f/10.128.0.2 Start
Time: Wed, 06 Jun 2018 08:38:01 +0200 Labels:
app=dyn-class-gen pod-template-hash=3860917113 Annotations:
vpaUpdates=Pod resources updated by dyn-class-gen-vpa:
container 0: cpu request, memory request Status: Running IP:
10.52.0.27 Controlled By:
ReplicaSet/dyn-class-gen-deployment-7db4f5c557 Containers:
dyn-class-gen-container: Container ID:
docker://688d6088efdc2045d56c4f187211e43f09f4654779bdaa3e50f6e378718cb976
Image: banzaicloud/dynclassgen:1.0 Image ID:
docker-pullable://banzaicloud/dynclassgen@sha256:134835da5696f3f56b3cc68c13421512868133bcf5aa9cd196867920f813e785
Port: <none> State: Running Started: Wed, 06 Jun 2018
08:38:03 +0200 Ready: True Restart Count: 0 Limits: cpu: 2
memory: 1Gi Requests: cpu: 250m memory: 642037204
Environment: DYN_CLASS_COUNT: 256 JVM_OPTS: -Xmx300M
MEM_USAGE_PER_OBJECT_MB: 1 Mounts:
/var/run/secrets/kubernetes.io/serviceaccount from
default-token-v7z2l (ro) Conditions: Type Status Initialized
True Ready True PodScheduled True Volumes:
default-token-v7z2l: Type: Secret (a volume populated by a
Secret) SecretName: default-token-v7z2l Optional: false QoS
Class: Burstable Node-Selectors: <none> Tolerations:
node.kubernetes.io/not-ready:NoExecute for 300s
node.kubernetes.io/unreachable:NoExecute for 300s Events:
Type Reason Age From Message

---

Normal Scheduled 31m default-scheduler Successfully assigned
dyn-class-gen-deployment-7db4f5c557-pd9bc to
gke-gkecluster-seba-636-po ol1-f8f0d428-6n1f Normal
SuccessfulMountVolume 31m kubelet,
gke-gkecluster-seba-636-pool1-f8f0d428-6n1f
MountVolume.SetUp succeeded for volume "default-token-v7z2l"
Normal Pulled 31m kubelet,
gke-gkecluster-seba-636-pool1-f8f0d428-6n1f Container image
"banzaicloud/dynclassgen:1.0" already present on machine
Normal Created 31m kubelet,
gke-gkecluster-seba-636-pool1-f8f0d428-6n1f Created
container Normal Started 31m kubelet,
gke-gkecluster-seba-636-pool1-f8f0d428-6n1f Started
container

Opinionated conclusions

  • VPA is in it's early stages and is expected to change its shape many times, so early adopters should be prepared for that. Details on known limitations can be found here and on future work here
  • VPA only adjusts the resources requests of containers based on observed past and current resource usage. It doesn't set resources limits. This can be problematic with misbehaving applications that begin using more and more resources, leading to pods being killed by Kubernetes.
Learn more about the different types of autoscaling features supported and automated by the Banzai Cloud Pipeline platform platform:
Subscribe card background
Subscribe
Subscribe to
the Shift!

Get emerging insights on emerging technology straight to your inbox.

Unlocking Multi-Cloud Security: Panoptica's Graph-Based Approach

Discover why security teams rely on Panoptica's graph-based technology to navigate and prioritize risks across multi-cloud landscapes, enhancing accuracy and resilience in safeguarding diverse ecosystems.

thumbnail
I
Subscribe
Subscribe
 to
the Shift
!
Get
emerging insights
on emerging technology straight to your inbox.

The Shift keeps you at the forefront of cloud native modern applications, application security, generative AI, quantum computing, and other groundbreaking innovations that are shaping the future of technology.

Outshift Background