Karpenter Spot Event

• https://github.com/aws/karpenter/blob/main/website/content/en/preview/getting-started/getting-started-with-karpenter/cloudformation.yaml

SQS 생성

import * as aws from "@pulumi/aws";
import * as variable from "@src/variable";

const karpenterInterruptionQueueName = "karpenter-interruption-queue";
export const karpenterInterruptionQueue = new aws.sqs.Queue(karpenterInterruptionQueueName, {
	namePrefix: `${karpenterInterruptionQueueName}-`,
	messageRetentionSeconds: 300,
	sqsManagedSseEnabled: true,
	tags: {
		Name: karpenterInterruptionQueueName,
		"loliot.net/stack": variable.stackName,
	},
});

const karpenterInterruptionQueuePolicyName = "karpenter-interruption-queue-policy";
const karpenterInterruptionQueuePolicy = new aws.sqs.QueuePolicy(karpenterInterruptionQueuePolicyName, {
	queueUrl: karpenterInterruptionQueue.id,
	policy: {
		Id: "EC2InterruptionPolicy",
		Version: "2012-10-17",
		Statement: [
			{
				Effect: "Allow",
				Principal: {
					Service: ["sqs.amazonaws.com", "events.amazonaws.com"],
				},
				Action: "sqs:SendMessage",
				Resource: [karpenterInterruptionQueue.arn],
			},
		],
	},
});

EventBridge 추가

const targetId = "KarpenterInterruptionQueueTarget";

const scheduledChangeRuleName = "scheduled-change-rule";
const scheduledChangeRule = new aws.cloudwatch.EventRule(scheduledChangeRuleName, {
	namePrefix: `${scheduledChangeRuleName}-`,
	eventPattern: JSON.stringify({
		source: ["aws.health"],
		"detail-type": ["AWS Health Event"],
	}),
	tags: {
		Name: scheduledChangeRuleName,
		"loliot.net/stack": variable.stackName,
	},
});

new aws.cloudwatch.EventTarget(`${scheduledChangeRuleName}-target`, {
	rule: scheduledChangeRule.name,
	arn: karpenterInterruptionQueue.arn,
	targetId: targetId,
});

const spotInterruptionRuleName = "spot-interruption-rule";
const spotInterruptionRule = new aws.cloudwatch.EventRule(spotInterruptionRuleName, {
	namePrefix: `${spotInterruptionRuleName}-`,
	eventPattern: JSON.stringify({
		source: ["aws.ec2"],
		"detail-type": ["EC2 Spot Instance Interruption Warning"],
	}),
	tags: {
		Name: spotInterruptionRuleName,
		"loliot.net/stack": variable.stackName,
	},
});

new aws.cloudwatch.EventTarget(`${spotInterruptionRuleName}-target`, {
	rule: spotInterruptionRule.name,
	arn: karpenterInterruptionQueue.arn,
	targetId: targetId,
});

const rebalanceRuleName = "rebalance-rule";
const rebalanceRule = new aws.cloudwatch.EventRule(rebalanceRuleName, {
	namePrefix: `${rebalanceRuleName}-`,
	eventPattern: JSON.stringify({
		source: ["aws.ec2"],
		"detail-type": ["EC2 Instance Rebalance Recommendation"],
	}),
	tags: {
		Name: rebalanceRuleName,
		"loliot.net/stack": variable.stackName,
	},
});

new aws.cloudwatch.EventTarget(`${rebalanceRuleName}-target`, {
	rule: rebalanceRule.name,
	arn: karpenterInterruptionQueue.arn,
	targetId: targetId,
});

const instanceStateChangeRuleName = "instance-state-change-rule";
const instanceStateChangeRule = new aws.cloudwatch.EventRule(instanceStateChangeRuleName, {
	namePrefix: `${instanceStateChangeRuleName}-`,
	eventPattern: JSON.stringify({
		source: ["aws.ec2"],
		"detail-type": ["EC2 Instance State-change Notification"],
	}),
	tags: {
		Name: instanceStateChangeRuleName,
		"loliot.net/stack": variable.stackName,
	},
});

new aws.cloudwatch.EventTarget(`${instanceStateChangeRuleName}-target`, {
	rule: instanceStateChangeRule.name,
	arn: karpenterInterruptionQueue.arn,
	targetId: targetId,
});

Client

현재 v0.28.0 버전에서는 EC2 Instance Rebalance Recommendation 이벤트를 처리하지 않고 있습니다. (https://github.com/aws/karpenter/blob/bf484b63ae500578ac2843f28ba6ba6ab849d83f/pkg/controllers/interruption/controller.go#L288-L295)

Karpenter 설정

const controllerPolicyName = "karpenter-controller-policy";
const controllerPolicy = new aws.iam.Policy(controllerPolicyName, {
	namePrefix: `${controllerPolicyName}-`,
	policy: {
		Version: "2012-10-17",
		Statement: [
			// ...
			{
				Sid: "AllowInterruptionQueueActions",
				Effect: "Allow",
				Resource: karpenterInterruptionQueue.arn,
				Action: ["sqs:DeleteMessage", "sqs:GetQueueAttributes", "sqs:GetQueueUrl", "sqs:ReceiveMessage"],
			},
		],
	},
	// ...
});

karpenter-values.yaml

settings:
  interruptionQueue: <sqsName>

AWS Node Termination Handler

• https://github.com/aws/aws-node-termination-handler