This page will cover the scaling of an HPCC Kubernetes Cluster along with Vertical and Horizontal Pod Scaling.
Cluster Autoscaler automatically resizes the amount of nodes in your application based on the demands of your workload. Cluster Autoscaler removes the need for Cloud admins to manually add/remove nodes or over-position nodes.
(If you are not familiar with the Cloud SDK, I recommend you use the Cloud Console. The GUI is intuitive and will accomplish the same things as the Cloud SDK)
For this tutorial, we will be creating an HPCC Cluster called “hpcc-cluster” (very original :))
Run the command below with your own parameters:
gcloud container clusters create hpcc-cluster \
--zone <zone> \
--node-locations <location(s)> \
--num-nodes <num-nodes> --enable-autoscaling --min-nodes <min-nodes> --max-nodes <max-nodes>
Run the command below with you own parameters:
gcloud container clusters update cluster-name --enable-autoscaling \
--min-nodes 1 --max-nodes 10 --zone <zone> --node-pool default-pool
When applications are deployed in GKE, you can determine how many replicas you want to run of each application. When GKE applications are scaled, you either increase or decrease the number of replicas.
Ex: Say if your GKE application is using up too many resources that you do not require. GKE scaling will automatically decrease the number of replicas being used, thus saving money and resources.
For this example, we will be using the kubectl apply command as it is easier to control/edit through config files. there are other options that you can choose from which are listed here.
If the HPCC Systems deployments already exists, you may be better off using kubectl scale or the Google Cloud Console.
Kubectl scale lets you instantly change the number of replicas you want to use per pod. Steps:
kubectl get deployments to get the controller deployment names. You should get something like this: hthor 1/1 1 1 6d6h
mydali 1/1 1 1 6d6h
myeclccserver 1/1 1 1 6d6h
myesp 1/1 1 1 6d6h
nfsstorage-hpcc-nfs 1/1 1 1 6d7h
roxie 1/1 1 1 6d6h
roxie-cluster-slave-1 1/1 1 1 6d6h
roxie-cluster-slave-2 1/1 1 1 6d6h
roxie-cluster-toposerver 1/1 1 1 6d6h
thor-agent 1/1 1 1 6d6h
thor-thoragent 1/1 1 1 6d6h
kubectl scale deployments <my-app> --replicas 4 t0 set the replica amount to 4. This value can be changed. kubectl get deployments <my-app> The result should be: NAME DESIRED CURRENT UP-TO-DATE AVAILABLE AGE
my-app 4 4 4 4 15m
The HPCC deployments do come predefined with the amount of replicas in each respective yaml folder. You can change these values to whatever you would like by either creating or editing spec: replicas.
The replica structure may look a little confusing for HPCC Systems because occasionally there are defined in other files. For example, in the helm/hpcc/values.yaml file, you can set the replica amount for roxie, myeclcserver, myesp, and hthor. So my advice would be to use kubectl scale after deployment.
Similar to the manual scaling above, Horizontal Autoscaling increases or decreases the amount of replicas, but (as the name implies) automatically.
For this Horizontal Autoscaling tutorial, we will be using kubectl apply.
resources:
# based on CPU utilization
requests:
cpu: "250m"
(This will have to be changed in each deployment yaml file)
apiVersion: autoscaling/v1
kind: HorizontalPodAutoscaler
metadata:
name: <deployment-name>
spec:
scaleTargetRef:
apiVersion: apps/v1
kind: Deployment
name: <deployment-name>
minReplicas: 1
maxReplicas: 10
targetCPUUtilizationPercentage: 50
kubectl apply -f <deployment>-hpa.yaml(You can change the minReplicas, maxReplicas, and targetCPUUtilizationPercentage to whatever you like. However, make sure your cluster has enough resources to handle any changes you make.)
kubectl get hpa. NAME REFERENCE TARGETS MINPODS MAXPODS REPLICAS AGE
name Deployment/name 0%/50% 1 10 3 61s
Vertical Autoscaler analyzes and adjusts your containers’ cpu and memory requests. It can be configured to only give recommendations or to make automatic changes to cpu and memory requests.
Creating a new cluster with vertical autoscaling is very straightforward. Just enter in the below command in Google SDK.
gcloud container clusters create cluster-name \
--enable-vertical-pod-autoscaling --cluster-version=1.14.7
Of course, change “cluster-name” to your HPCC clusters’ name.
Enabling vertical autoscaling on an existing cluster is also very simple. Enter in below command into the Google Cloud SDK.
gcloud container clusters update cluster-name --enable-vertical-pod-autoscaling
apiVersion: autoscaling.k8s.io/v1
kind: VerticalPodAutoscaler
metadata:
name: <deployment>-vpa
spec:
targetRef:
apiVersion: "apps/v1"
kind: Deployment
name: <deployment>
updatePolicy:
updateMode: "Off"
kubectl create -f <deployment>-vpa.yamlWhen updateMode is marked off, VPA does not take action on its resource recommendations. VPA analyzes the cpu usage and outputs its results in the “status” field.
When updateMode is marked “Auto”, VPA has the ability to automatically scale based on its own recommendations. If you want to take action on resource recommendations, change “Off” to “Auto”.
You can access recommendations through the kube command kubectl get <deployment>-vpa --output yaml. (Expected Results)
...
recommendation:
containerRecommendations:
- containerName: <my-container>
lowerBound:
cpu: 25m
memory: 262144k
target:
cpu: 25m
memory: 262144k
upperBound:
cpu: 7931m
memory: 8291500k
...
After finishing this tutorial on autoscaling, the next step would be introducing yaml files into the HPCC ecosystem that will manage autoscaling through code. This offers tremendous benefits when compared to command line commands by simplifying the process.
| “Cluster Autoscaler | Kubernetes Engine Documentation | Google Cloud.” Google, Google, cloud.google.com/kubernetes-engine/docs/concepts/cluster-autoscaler. |