This project is an operator allowing Kubernetes to automatically manage the lifecycle of instances within a cluster based on specific events.
This tool is not intended to replace existing tools such as the Cluster Autoscaler but rather to supplement them in order to have more control and responsiveness over the provisioning of instances in a cluster.
Kubestitute only works with clusters deployed on AWS using Auto Scaling Groups at the moment.
The standard use case for this tool is to provision on-demand fallback instances in case Spot instances cannot be scheduled.
To do so, configure an Auto Scaling Group of Spot instances managed by the Cluster Autoscaler and another one of on-demand fallback instances managed by Kubestitute.
Kubestitute will scale up the on-demand Auto Scaling Group according to events on the Spot instances Auto Scaling Group retrieved from the cluster-autoscaler status. It will also drain fallback instances and detach them from the Auto Scaling Group according to events (typically when the Spot instances have finally been scheduled).
A Kubernetes cluster of version v1.11.3+ is required. If you are just starting out with Kubestitute, it is highly recommended to use the latest version.
To be used with AWS and interact with Auto Scaling Groups, an AWS account or IAM role with the following permissions on Auto Scaling Groups managed by Kubestitute is required:
{
"Version": "2012-10-17",
"Statement": [
{
"Sid": "AllObjectActions",
"Effect": "Allow",
"Action": [
"autoscaling:DescribeAutoScalingGroups",
"autoscaling:SetDesiredCapacity",
"autoscaling:TerminateInstanceInAutoScalingGroup"
],
"Resource": "*"
}
]
}Follow Kubestitute documentation for Helm deployment here.
The kubestitute container takes as argument the parameters below.
| Key | Description | Default |
|---|---|---|
| clusterautoscaler-namespace | The namespace the clusterautoscaler belongs to. | kube-system |
| clusterautoscaler-status-name | The names of the clusterautoscaler status configmap. | cluster-autoscaler-status |
| cluster-autoscaler-priority-expander-config-map | The name of the clusterautoscaler priority expander config map. | cluster-autoscaler-priority-expander |
| priority-expander-enabled | Is the PriorityExpander controller enabled. | false |
| priority-expander-namespace | The namespace the unique priority expander object belongs to. | kubestitute-system |
| priority-expander-name | The only accepted name for the priority expander object. | priority-expander-default |
| dev | Enable dev mode for logging. | false |
| v | Logs verbosity. 0 => panic, 1 => error, 2 => warning, 3 => info, 4 => debug | 3 |
| asg-poll-interval | AutoScaling Groups polling interval (used to generate custom metrics about ASGs). | 30 |
| eviction-timeout | The timeout in seconds for pods eviction on Instance deletion. | 300 |
| instances-max-concurrent-reconciles | The maximum number of concurrent Reconciles which can be run for Instances. | 10 |
| metrics-bind-address | The address the metric endpoint binds to. | :8080 |
| health-probe-bind-address | The address the probe endpoint binds to. | :8081 |
| leader-elect | Enable leader election for controller manager. Enabling this will ensure there is only one active controller manager. | false |
A core feature of Kubestitute is to monitor the Kubernetes API server for changes to specific objects and ensure that the current cluster infrastructure match these objects.
The Operator acts on the following custom resource definitions (CRDs):
Instance defines a desired Instance (only AWS EC2 instances in Auto Scaling Groups supported at the moment).
Scheduler defines a scheduler for Instances (only AWS EC2 instances in Auto Scaling Groups supported at the moment). This resource is used to configure advanced instances scheduling based on node groups events.
PriorityExpander defines a template that will be used to dynamically create cluster-autoscaler configmap. More information here
You can find examples of CRDs defined by Kubestitute here.
Full API documentation is available here.
By default, Kubestitute produces structured logs, with "Info" verbosity. These settings can be configured as described here.
Kubestitute being built from Kubebuilder, it natively exposes a collection of performance metrics for each controller. Kubebuilder documentation about metrics can be found here.
We also expose custom metrics as described here:
All the metrics are prefixed with kubestitute_
| Metric name | Metric type | Labels | Description |
|---|---|---|---|
| scaled_up_nodes_total | Counter | autoscaling_group_name, scheduler_name |
Number of nodes added by kubestitute. |
| scaled_down_nodes_total | Counter | autoscaling_group_name, scheduler_name |
Number of nodes removed by kubestitute. |
| evicted_pods_total | Counter | autoscaling_group_name, node_name, scheduler_name |
Number of pods evicted by kubestitute. |
| autoscaling_group_desired_capacity | Gauge | autoscaling_group_name |
The desired size of the autoscaling group. |
| autoscaling_group_capacity | Gauge | autoscaling_group_name |
The current autoscaling group capacity (Pending and InService instances). |
| autoscaling_group_min_size | Gauge | autoscaling_group_name |
The minimum size of the autoscaling group. |
| autoscaling_group_max_size | Gauge | autoscaling_group_name |
The maximum size of the autoscaling group. |
| priority_expander_template_error | Gauge | Returns 1 if template can't be parsed. |
Distributed under the Apache 2.0 License. See LICENSE for more information.
We use SemVer for versioning.
To contribute to this project, please first consult the contribution rules guide.
Got a question? File a GitHub issue