Scale down the Cluster

This lesson focuses on how to scale down the Cluster and the rules which govern it.

Scale down the nodes #

Scaling up the cluster to meet the demand is essential since it allows us to host all the replicas we need to fulfill (some of) our SLAs. When the demand drops and our nodes become underutilized, we should scale down. That is not essential given that our users will not experience problems caused by having too much hardware in our cluster. Nevertheless, we shouldn’t have underutilized nodes if we are to reduce expenses. Unused nodes result in wasted money. That is true in all situations, especially when running in Cloud and paying only for the resources we used. Even on-prem, where we already purchased hardware, it is essential to scale down and release resources so that they can be used by other clusters.

We’ll simulate a decrease in demand by applying a new definition that will redefine the HPAs threshold to 2 (min) and 5 (max).

kubectl apply \
    -f scaling/go-demo-5.yml \

kubectl -n go-demo-5 get hpa

The output of the latter command is as follows.

api  Deployment/api 0%/80%, 0%/80%   2       5       15       2m56s
db   StatefulSet/db 56%/80%, 10%/80% 3       5       3        2m57s

We can see that the min and max values of the api HPA changed to 2 and 5. The current number of replicas is still 15, but that will drop to 5 soon. The HPA already changed the replicas of the Deployment, so let’s wait until it rolls out and take another look at the Pods.

kubectl -n go-demo-5 rollout status \
    deployment api

kubectl -n go-demo-5 get pods

The output of the latter command is as follows.

api-... 1/1   Running 0        104s
api-... 1/1   Running 0        104s
api-... 1/1   Running 0        104s
api-... 1/1   Running 0        94s
api-... 1/1   Running 0        104s
db-0    2/2   Running 0        4m37s
db-1    2/2   Running 0        3m57s
db-2    2/2   Running 0        3m18s

Let’s see what happened to the nodes.

kubectl get nodes

The output shows that we still have four nodes (or whatever was your number before we de-scaled the Deployment). Given that we haven’t yet reached the desired state of only three nodes, we might want to take another look at the cluster-autoscaler-status ConfigMap.

kubectl -n kube-system \
    get configmap \
    cluster-autoscaler-status \
    -o yaml

The output, limited to the relevant parts, is as follows.

apiVersion: v1
  status: |+
    Cluster-autoscaler status at 2018-10-03 ...
      Health: Healthy (ready=4 ...)
      ScaleDown: CandidatesPresent (candidates=1)
      Name:      ...gke-devops25-default-pool-f4c233dd-grp
      ScaleDown: CandidatesPresent (candidates=1)
                 LastProbeTime:      2018-10-03 23:06:...
                 LastTransitionTime: 2018-10-03 23:05:...

If your output does not contain ScaleDown: CandidatesPresent, you might need to wait a bit and repeat the previous command.

If we focus on the Health section of the cluster-wide status, all four nodes are still ready.

Candidate for removal #

Judging by the cluster-wide section of the status, we can see that there is one candidate to ScaleDown (it might be more in your case). If we move to the NodeGroups, we can observe that one of them has CandidatesPresent set to 1 in the ScaleDown section (or whatever was your initial value before scaling up). In other words, one of the nodes is the candidate for removal. If it remains so for ten minutes, the node will be drained first to allow graceful shutdown of the Pods running inside it. After that, it will be physically removed through manipulation of the scaling group.

Get hands-on with 1000+ tech skills courses.