The Kubernetes service on the Catalyst Cloud has an optional feature called
cluster auto-scaling capable of dynamically increasing or reducing the
number of worker nodes, according their current resource allocation.
Auto-scaling does not use the current CPU and memory usage as metrics to resize the cluster, but rather the CPU and memory reservation (resource requests) in the pod specification to determine if the capacity allocated to a worker node.
cluster creation time (with minimum and maximum node count be adjusted accordingly):
The minimum number of nodes must be greater than zero and lower than the maximum. The maximum number of nodes must be greater than the minimum.
The auto-scaling feature requires the use of resource requests for CPU and memory in the pod specification. The following deployment template illustrates the use of resource requests:
apiVersion: v1 kind: Pod metadata: name: webserver spec: containers: - name: webserver-ctr image: nginx resources: requests: memory: "128Mi" cpu: "0.5"
The conditions that trigger a cluster resize are explained below:
Scale out: a worker node is added to the cluster when the Kubernetes scheduler is unable to allocate a pod to any existing worker node due to insufficient capacity.
Scale in: a worker node is removed from the cluster when the cluster resource usage drops below the defined threshold (by default 50%) for a period of time.
Let’s see how this works in practice by launching a new cluster that has cluster-autoscaling enabled.
The first thing we need to do is get a list of the current labels associated with a template we wish to deploy from. To do this run the following command.
$ openstack coe cluster template show <template_name> -c labels -f yaml labels: cloud_provider_enabled: 'true' cloud_provider_tag: 1.14.0-catalyst container_infra_prefix: docker.io/catalystcloud/ heat_container_agent_tag: stein-dev ingress_controller: octavia kube_tag: v1.12.7 octavia_ingress_controller_tag: 1.14.0-catalyst
When adding or altering a label it is necessary, at the current time, to supply all of the other labels that are present in the template.
We now need to convert the list of labels obtained in the previous step, to a
comma separated list of key value pairs that include
auto_scaling_enabled=true,min_node_count=1,max_node_count=2,cloud_provider_enabled=true,cloud_provider_tag=1.14.0-catalyst,... <output truncated>
These are then passed as the label parameter to our cluster create command as shown here.
$ openstack coe cluster create my-cluster \ --cluster-template kubernetes-v1.12.7-prod-20190403 \ --keypair mykey \ --master-count 3 \ --node-count 3 \ --labels auto_scaling_enabled=true,min_node_count=1,max_node_count=2,<existing-labels>
There are several parameters that could change the auto-scaling behaviour, such as:
scale-down-utilization-threshold This is the Node utilization level,
which is defined as the sum of requested resources divided by capacity,
below which a node can be considered for scale down. By default this is
scale-down-unneeded-time This is how long a node should be unneeded
before it is eligible to be scaled down. By default this is 10 minutes.
To change the scale down parameters we need to edit the cluster-autoscaler’s
current deployment settings. We can do this using
kubectl -n kube-system edit deployment cluster-autoscaler
This will open the corresponding YAML file in an editor. Locate the
section as shown below.
spec: containers: - command: - ./cluster-autoscaler - --alsologtostderr - --cloud-provider=magnum - --cluster-name=cad28c31-cf1c-40a7-b8c8-b9fff91a1113 - --cloud-config=/config/cloud-config - --nodes=1:4:default-worker - --scale-down-unneeded-time=10m - --scale-down-delay-after-failure=3m - --scale-down-delay-after-add=10m image: docker.io/catalystcloud/cluster-autoscaler:v1.0
If we wanted to change an existing vlue, simply edit it in place. If you need
to override one of the default values that may not display be default, add it
image: line making sure to match the indenting and formatting
In the example below we have changed the following:
scale-down-unneeded-time parameter has been reduced to 8 minutes.
scale-down-utilization-threshold has been added in, with a value of
spec: containers: - command: - ./cluster-autoscaler - --alsologtostderr - --cloud-provider=magnum - --cluster-name=cad28c31-cf1c-40a7-b8c8-b9fff91a1113 - --cloud-config=/config/cloud-config - --nodes=1:4:default-worker - --scale-down-unneeded-time=8m - --scale-down-delay-after-failure=3m - --scale-down-delay-after-add=10m - --scale-down-utilization-threshold=0.4 image: docker.io/catalystcloud/cluster-autoscaler:v1.0
Once the required changes have been made save the file and exit. This will
cause the deployment to create a new
cluster-autoscaler pod and once it is
RUNNING it will remove the original one .
For more detailed information about the Cluster-Autoscaler please take a look at the FAQ .