AWS Node Termination Handler

Gracefully handle EC2 instance shutdown within Kubernetes

Project Summary

This project ensures that the Kubernetes control plane responds appropriately to events that can cause your EC2 instance to become unavailable, such as EC2 maintenance events, EC2 Spot interruptions, ASG Scale-In, ASG AZ Rebalance, and EC2 Instance Termination via the API or Console. If not handled, your application code may not stop gracefully, take longer to recover full availability, or accidentally schedule work to nodes that are going down.

The aws-node-termination-handler (NTH) can operate in two different modes: Instance Metadata Service (IMDS) or the Queue Processor.

The aws-node-termination-handler Instance Metadata Service Monitor will run a small pod on each host to perform monitoring of IMDS paths like /spot or /events and react accordingly to drain and/or cordon the corresponding node.

The aws-node-termination-handler Queue Processor will monitor an SQS queue of events from Amazon EventBridge for ASG lifecycle events, EC2 status change events, Spot Interruption Termination Notice events, and Spot Rebalance Recommendation events. When NTH detects an instance is going down, we use the Kubernetes API to cordon the node to ensure no new work is scheduled there, then drain it, removing any existing work. The termination handler Queue Processor requires AWS IAM permissions to monitor and manage the SQS queue and to query the EC2 API.

You can run the termination handler on any Kubernetes cluster running on AWS, including self-managed clusters and those created with Amazon Elastic Kubernetes Service. If you're using EKS managed node groups, you don't need the aws-node-termination-handler.

Major Features

Both modes (IMDS and Queue Processor) monitor for events affecting your EC2 instances, but each supports different types of events. Both modes have the following:

Helm installation and event configuration support
Webhook feature to send shutdown or restart notification messages
Unit & integration tests

Instance Metadata Service Processor

Must be deployed as a Kubernetes DaemonSet.

Monitors EC2 Instance Metadata for:

Queue Processor

Must be deployed as a Kubernetes Deployment. Also requires some additional infrastructure setup (including SQS queue, EventBridge rules).

Monitors an SQS Queue for:
- Spot Instance Termination Notifications
- Scheduled Events (via AWS Health)
- Instance Rebalance Recommendations
- ASG Termination Lifecycle Hooks to handle the following:
  - ASG Scale-In
  - Availability Zone Rebalance
  - Unhealthy Instances, and more
- Instance State Change events

Which one should I use?

Feature	IMDS Processor	Queue Processor
Spot Instance Termination Notifications (ITN)	✅	✅
Scheduled Events	✅	✅
Instance Rebalance Recommendation	✅	✅
AZ Rebalance Recommendation	❌	✅
ASG Termination Lifecycle Hooks	❌	✅
Instance State Change Events	❌	✅

Kubernetes Compatibility

NTH Release	K8s v1.26	K8s v1.25	K8s v1.24	K8s v1.23	K8s v1.22	K8s v1.21
v1.20.0	✅	✅	✅	✅	✅	❌
v1.19.0	❌	❌	❌	✅	✅	✅
v1.18.3	❌	❌	❌	✅	✅	✅
v1.18.2	❌	❌	❌	✅	✅	✅
v1.18.1	❌	❌	❌	✅	✅	✅
v1.18.0	❌	❌	❌	✅	✅	✅

Installation and Configuration

The aws-node-termination-handler can operate in two different modes: IMDS Processor and Queue Processor. The enableSqsTerminationDraining helm configuration key or the ENABLE_SQS_TERMINATION_DRAINING environment variable are used to enable the Queue Processor mode of operation. If enableSqsTerminationDraining is set to true, then IMDS paths will NOT be monitored. If the enableSqsTerminationDraining is set to false, then IMDS Processor Mode will be enabled. Queue Processor Mode and IMDS Processor Mode cannot be run at the same time.

IMDS Processor Mode allows for a fine-grained configuration of IMDS paths that are monitored. There are currently 3 paths supported that can be enabled or disabled by using the following helm configuration keys:

enableSpotInterruptionDraining
enableRebalanceMonitoring
enableScheduledEventDraining

By default, IMDS mode will only Cordon in response to a Rebalance Recommendation event (all other events are Cordoned and Drained). Cordon is the default for a rebalance event because it's not known if an ASG is being utilized and if that ASG is configured to replace the instance on a rebalance event. If you are using an ASG w/ rebalance recommendations enabled, then you can set the enableRebalanceDraining flag to true to perform a Cordon and Drain when a rebalance event is received.

Rebalance Recommendation is an early indicator to notify the Spot Instances that they can be interrupted soon. Node Termination Handler supports AZ Rebalance Recommendation only in Queue Processor mode using ASG Lifecycle Hooks. For AZ rebalances the instances are just terminated, using Lifecycle Hooks and EventBridge rule for EC2 Instance-terminate Lifecycle Action we can handle OD Instances.

The enableSqsTerminationDraining must be set to false for these configuration values to be considered.

The Queue Processor Mode does not allow for fine-grained configuration of which events are handled through helm configuration keys. Instead, you can modify your Amazon EventBridge rules to not send certain types of events to the SQS Queue so that NTH does not process those events. All events when operating in Queue Processor mode are Cordoned and Drained unless the cordon-only flag is set to true.

The enableSqsTerminationDraining flag turns on Queue Processor Mode. When Queue Processor Mode is enabled, IMDS mode will be disabled, even if you explicitly enabled any of the IMDS configuration keys. NTH cannot respond to queue events AND monitor IMDS paths. In this case, it is safe to disable IMDS for the NTH pod.

AWS Node Termination Handler - IMDS Processor

Installation and Configuration

The termination handler DaemonSet installs into your cluster a ServiceAccount, ClusterRole, ClusterRoleBinding, and a DaemonSet. All four of these Kubernetes constructs are required for the termination handler to run properly.

Pod Security Admission

When using Kubernetes Pod Security Admission it is recommended to assign the [privileged](https://kubernetes.io/docs/concepts/security/pod-security-standards/#privileged) level.

Kubectl Apply

You can use kubectl to directly add all of the above resources with the default configuration into your cluster.

kubectl apply -f https://github.com/aws/aws-node-termination-handler/releases/download/v1.20.0/all-resources.yaml

For a full list of releases and associated artifacts see our releases page.

Helm

The easiest way to configure the various options of the termination handler is via helm. The chart for this project is hosted in helm/aws-node-termination-handler

To get started you need to authenticate your helm client

aws ecr-public get-login-password \
  --region us-east-1 | helm registry login \
  --username AWS \
  --password-stdin public.ecr.aws

Once that is complete you can install the termination handler. We've provided some sample setup options below. Make sure to replace CHART_VERSION with the version you want to install.

Zero Config:

helm upgrade --install aws-node-termination-handler \
  --namespace kube-system \
  oci://public.ecr.aws/aws-ec2/helm/aws-node-termination-handler --version $CHART_VERSION

Enabling Features:

helm upgrade --install aws-node-termination-handler \
  --namespace kube-system \
  --set enableSpotInterruptionDraining="true" \
  --set enableRebalanceMonitoring="true" \
  --set enableScheduledEventDraining="false" \
  oci://public.ecr.aws/aws-ec2/helm/aws-node-termination-handler --version $CHART_VERSION

The enable* configuration flags above enable or disable IMDS monitoring paths.

Running Only On Specific Nodes:

helm upgrade --install aws-node-termination-handler \
  --namespace kube-system \
  --set nodeSelector.lifecycle=spot \
  oci://public.ecr.aws/aws-ec2/helm/aws-node-termination-handler --version $CHART_VERSION

Webhook Configuration:

helm upgrade --install aws-node-termination-handler \
  --namespace kube-system \
  --set webhookURL=https://hooks.slack.com/services/YOUR/SLACK/URL \
  oci://public.ecr.aws/aws-ec2/helm/aws-node-termination-handler --version $CHART_VERSION

Alternatively, pass Webhook URL as a Secret:

WEBHOOKURL_LITERAL="webhookurl=https://hooks.slack.com/services/YOUR/SLACK/URL"

kubectl create secret -n kube-system generic webhooksecret --from-literal=$WEBHOOKURL_LITERAL

helm upgrade --install aws-node-termination-handler \
  --namespace kube-system \
  --set webhookURLSecretName=webhooksecret \
  oci://public.ecr.aws/aws-ec2/helm/aws-node-termination-handler --version $CHART_VERSION

For a full list of configuration options see our Helm readme.

AWS Node Termination Handler - Queue Processor (requires AWS IAM Permissions)

Infrastructure Setup

The termination handler requires some infrastructure prepared before deploying the application. In a multi-cluster environment, you will need to repeat the following steps for each cluster.

You'll need the following AWS infrastructure components:

Amazon Simple Queue Service (SQS) Queue
AutoScaling Group Termination Lifecycle Hook
Amazon EventBridge Rule
IAM Role for the aws-node-termination-handler Queue Processing Pods

1. Create an SQS Queue:

Here is the AWS CLI command to create an SQS queue to hold termination events from ASG and EC2, although this should really be configured via your favorite infrastructure-as-code tool like CloudFormation (template here) or Terraform:

## Queue Policy
QUEUE_POLICY=$(cat <<EOF
{
    "Version": "2012-10-17",
    "Id": "MyQueuePolicy",
    "Statement": [{
        "Effect": "Allow",
        "Principal": {
            "Service": ["events.amazonaws.com", "sqs.amazonaws.com"]
        },
        "Action": "sqs:SendMessage",
        "Resource": [
            "arn:aws:sqs:${AWS_REGION}:${ACCOUNT_ID}:${SQS_QUEUE_NAME}"
        ]
    }]
}
EOF
)

## make sure the queue policy is valid JSON
echo "$QUEUE_POLICY" | jq .

## Save queue attributes to a temp file
cat << EOF > /tmp/queue-attributes.json
{
  "MessageRetentionPeriod": "300",
  "Policy": "$(echo $QUEUE_POLICY | sed 's/\"/\\"/g' | tr -d -s '\n' " ")",
  "SqsManagedSseEnabled": "true"
}
EOF

aws sqs create-queue --queue-name "${SQS_QUEUE_NAME}" --attributes file:///tmp/queue-attributes.json

If you are sending Lifecycle termination events from ASG directly to SQS, instead of through EventBridge, then you will also need to create an IAM service role to give Amazon EC2 Auto Scaling access to your SQS queue. Please follow these linked instructions to create the IAM service role: link. Note the ARNs for the SQS queue and the associated IAM role for Step 2.

There are some caveats when using server side encryption with SQS:

using SSE-KMS with a customer managed key requires changing the KMS key policy to allow EventBridge to publish events to SQS.
using SSE-KMS with an AWS managed key is not supported as the KMS key policy can't be updated to allow EventBridge to publish events to SQS.
using SSE-SQS doesn't require extra setup and works out of the box as SQS queues without encryption at rest.

2. Create an ASG Termination Lifecycle Hook:

Here is the AWS CLI command to create a termination lifecycle hook on an existing ASG when using EventBridge, although this should really be configured via your favorite infrastructure-as-code tool like CloudFormation or Terraform:

aws autoscaling put-lifecycle-hook \
  --lifecycle-hook-name=my-k8s-term-hook \
  --auto-scaling-group-name=my-k8s-asg \
  --lifecycle-transition=autoscaling:EC2_INSTANCE_TERMINATING \
  --default-result=CONTINUE \
  --heartbeat-timeout=300

If you want to avoid using EventBridge and instead send ASG Lifecycle events directly to SQS, instead use the following command, using the ARNs from Step 1:

aws autoscaling put-lifecycle-hook \
  --lifecycle-hook-name=my-k8s-term-hook \
  --auto-scaling-group-name=my-k8s-asg \
  --lifecycle-transition=autoscaling:EC2_INSTANCE_TERMINATING \
  --default-result=CONTINUE \
  --heartbeat-timeout=300 \
  --notification-target-arn <your test queue ARN here> \
  --role-arn <your SQS access role ARN here>

3. Tag the Instances:

By default the aws-node-termination-handler will only manage terminations for instances tagged with key=aws-node-termination-handler/managed. The value of the key does not matter.

To tag ASGs and propagate the tags to your instances (recommended):

aws autoscaling create-or-update-tags \
  --tags ResourceId=my-auto-scaling-group,ResourceType=auto-scaling-group,Key=aws-node-termination-handler/managed,Value=,PropagateAtLaunch=true

To tag an individual EC2 instance:

aws ec2 create-tags \
    --resources i-1234567890abcdef0 \
    --tags 'Key="aws-node-termination-handler/managed",Value='

Tagging your EC2 instances in this way is helpful if you only want aws-node-termination-handler to manage the lifecycle of instances in certain ASGs. For example, if your account also has other ASGs that do not contain Kubernetes nodes, this tagging mechanism will ensure that NTH does not manage the lifecycle of any instances in those non-Kubernetes ASGs.

However, if the only ASGs in your account are for your Kubernetes cluster, then you can turn off the tag check by setting the flag --check-tag-before-draining=false or environment variable CHECK_TAG_BEFORE_DRAINING=false.

You can also control what resources NTH manages by adding the resource ARNs to your Amazon EventBridge rules.

Take a look at the docs on how to create rules that only manage certain ASGs, and read about all the supported ASG events.

4. Create Amazon EventBridge Rules

You may skip this step if sending events from ASG to SQS directly.

If we use ASG with capacity-rebalance enabled on ASG, then we do not need Spot and Rebalance events enabled with EventBridge. ASG will send a termination lifecycle hook for spot interrruptions while it's launching a new instance and for Rebalance events ASG will send a termination lifecycle hook after it brings a new node in the ASG.

If we use ASG without capacity-rebalance enabled, then spot interruptions will cause a termination lifecycle hook after the interruption occurs but not while launching the new instance.

Here are AWS CLI commands to create Amazon EventBridge rules so that ASG termination events, Spot Interruptions, Instance state changes, Rebalance Recommendations, and AWS Health Scheduled Changes are sent to the SQS queue created in the previous step. This should really be configured via your favorite infrastructure-as-code tool like CloudFormation (template here) or Terraform:

aws events put-rule \
  --name MyK8sASGTermRule \
  --event-pattern "{\"source\":[\"aws.autoscaling\"],\"detail-type\":[\"EC2 Instance-terminate Lifecycle Action\"]}"

aws events put-targets --rule MyK8sASGTermRule \
  --targets "Id"="1","Arn"="arn:aws:sqs:us-east-1:123456789012:MyK8sTermQueue"

aws events put-rule \
  --name MyK8sSpotTermRule \
  --event-pattern "{\"source\": [\"aws.ec2\"],\"detail-type\": [\"EC2 Spot Instance Interruption Warning\"]}"

aws events put-targets --rule MyK8sSpotTermRule \
  --targets "Id"="1","Arn"="arn:aws:sqs:us-east-1:123456789012:MyK8sTermQueue"

aws events put-rule \
  --name MyK8sRebalanceRule \
  --event-pattern "{\"source\": [\"aws.ec2\"],\"detail-type\": [\"EC2 Instance Rebalance Recommendation\"]}"

aws events put-targets --rule MyK8sRebalanceRule \
  --targets "Id"="1","Arn"="arn:aws:sqs:us-east-1:123456789012:MyK8sTermQueue"

aws events put-rule \
  --name MyK8sInstanceStateChangeRule \
  --event-pattern "{\"source\": [\"aws.ec2\"],\"detail-type\": [\"EC2 Instance State-change Notification\"]}"

aws events put-targets --rule MyK8sInstanceStateChangeRule \
  --targets "Id"="1","Arn"="arn:aws:sqs:us-east-1:123456789012:MyK8sTermQueue"

aws events put-rule \
  --name MyK8sScheduledChangeRule \
  --event-pattern "{\"source\": [\"aws.health\"],\"detail-type\": [\"AWS Health Event\"],\"detail\": {\"service\": [\"EC2\"],\"eventTypeCategory\": [\"scheduledChange\"]}}"

aws events put-targets --rule MyK8sScheduledChangeRule \
  --targets "Id"="1","Arn"="arn:aws:sqs:us-east-1:123456789012:MyK8sTermQueue"

5. Create an IAM Role for the Pods

There are many different ways to allow the aws-node-termination-handler pods to assume a role:

IAM Policy for aws-node-termination-handler Deployment:

{
    "Version": "2012-10-17",
    "Statement": [
        {
            "Effect": "Allow",
            "Action": [
                "autoscaling:CompleteLifecycleAction",
                "autoscaling:DescribeAutoScalingInstances",
                "autoscaling:DescribeTags",
                "ec2:DescribeInstances",
                "sqs:DeleteMessage",
                "sqs:ReceiveMessage"
            ],
            "Resource": "*"
        }
    ]
}

Installation

Pod Security Admission

When using Kubernetes Pod Security Admission it is recommended to assign the [baseline](https://kubernetes.io/docs/concepts/security/pod-security-standards/#baseline) level.

Helm

The easiest way to configure the various options of the termination handler is via helm. The chart for this project is hosted in helm/aws-node-termination-handler

To get started you need to authenticate your helm client

aws ecr-public get-login-password \
     --region us-east-1 | helm registry login \
     --username AWS \
     --password-stdin public.ecr.aws

Once that is complete you can install the termination handler. We've provided some sample setup options below. Make sure to replace CHART_VERSION with the version you want to install.

Minimal Config:

helm upgrade --install aws-node-termination-handler \
  --namespace kube-system \
  --set enableSqsTerminationDraining=true \
  --set queueURL=https://sqs.us-east-1.amazonaws.com/0123456789/my-term-queue \
  oci://public.ecr.aws/aws-ec2/helm/aws-node-termination-handler --version $CHART_VERSION

Webhook Configuration:

helm upgrade --install aws-node-termination-handler \
  --namespace kube-system \
  --set enableSqsTerminationDraining=true \
  --set queueURL=https://sqs.us-east-1.amazonaws.com/0123456789/my-term-queue \
  --set webhookURL=https://hooks.slack.com/services/YOUR/SLACK/URL \
  oci://public.ecr.aws/aws-ec2/helm/aws-node-termination-handler --version $CHART_VERSION

Alternatively, pass Webhook URL as a Secret:

WEBHOOKURL_LITERAL="webhookurl=https://hooks.slack.com/services/YOUR/SLACK/URL"

kubectl create secret -n kube-system generic webhooksecret --from-literal=$WEBHOOKURL_LITERAL

helm upgrade --install aws-node-termination-handler \
  --namespace kube-system \
  --set enableSqsTerminationDraining=true \
  --set queueURL=https://sqs.us-east-1.amazonaws.com/0123456789/my-term-queue \
  --set webhookURLSecretName=webhooksecret \
  oci://public.ecr.aws/aws-ec2/helm/aws-node-termination-handler --version $CHART_VERSION

For a full list of configuration options see our Helm readme.

Single Instance vs Multiple Replicas

The Helm chart, by default, will deploy a single instance of Amazon Node Termination Handler. With the minimizing of resource usage, a single instance still provides good responsiveness in processing SQS messages.

When should multiple instances of Amazon Node Termination Handler be used?

Responsiveness: Amazon Node Termination Handler may be taking longer than desired to process certain events, potentially in processing numerous concurrent events or taking too long to drain Pods. The deployment of multiple Amazon Node Termination Handler instances may help.
Availability: The deployment of multiple Amazon Node Termination Handler instances provides mitigation in the case that Amazon Node Termination Handler itself is drained. Replica Amazon Node Termination Handlers will process SQS messages, avoiding a delay until the Deployment can start another instance.

Notes

Running multiple instances of Amazon Node Termination Handler will not load balance responding to events. Each instance will greedily consume and respond to events.
Logs from multiple instances of Amazon Node Termination Handler are not aggregated.
Multiple instances of Amazon Node Termination Handler may respond to the same event, if it takes longer than 20s to process. This is not an error case, only the first response will have an affect.

Kubectl Apply

Queue Processor needs an SQS queue URL to function; therefore, manifest changes are REQUIRED before using kubectl to directly add all of the above resources into your cluster.

Minimal Config:

curl -L https://github.com/aws/aws-node-termination-handler/releases/download/v1.20.0/all-resources-queue-processor.yaml -o all-resources-queue-processor.yaml
<open all-resources-queue-processor.yaml and update QUEUE_URL value>
kubectl apply -f ./all-resources-queue-processor.yaml

For a full list of releases and associated artifacts see our releases page.

Use with Kiam

If you are using IMDS mode which defaults to hostNetworking: true, or if you are using queue-processor mode, then this section does not apply. The configuration below only needs to be used if you are explicitly changing NTH IMDS mode to hostNetworking: false .

To use the termination handler alongside Kiam requires some extra configuration on Kiam's end. By default Kiam will block all access to the metadata address, so you need to make sure it passes through the requests the termination handler relies on.

To add a whitelist configuration, use the following fields in the Kiam Helm chart values:

agent.whiteListRouteRegexp: '^\/latest\/meta-data\/(spot\/instance-action|events\/maintenance\/scheduled|instance-(id|type)|public-(hostname|ipv4)|local-(hostname|ipv4)|placement\/availability-zone)|\/latest\/dynamic\/instance-identity\/document$'

Or just pass it as an argument to the kiam agents:

kiam agent --whitelist-route-regexp='^\/latest\/meta-data\/(spot\/instance-action|events\/maintenance\/scheduled|instance-(id|type)|public-(hostname|ipv4)|local-(hostname|ipv4)|placement\/availability-zone)|\/latest\/dynamic\/instance-identity\/document$'

Metadata endpoints

The termination handler relies on the following metadata endpoints to function properly:

/latest/dynamic/instance-identity/document
/latest/meta-data/spot/instance-action
/latest/meta-data/events/recommendations/rebalance
/latest/meta-data/events/maintenance/scheduled
/latest/meta-data/instance-id
/latest/meta-data/instance-life-cycle
/latest/meta-data/instance-type
/latest/meta-data/public-hostname
/latest/meta-data/public-ipv4
/latest/meta-data/local-hostname
/latest/meta-data/local-ipv4
/latest/meta-data/placement/availability-zone

Building

For build instructions please consult BUILD.md.

Metrics

Available Prometheus metrics:

Metric name	Description
`actions_node`	Number of actions per node
`events_error`	Number of errors in events processing

Communication

If you've run into a bug or have a new feature request, please open an issue.
You can also chat with us in the Kubernetes Slack in the #provider-aws channel
Check out the open source Amazon EC2 Spot Instances Integrations Roadmap to see what we're working on and give us feedback!

Contributing

Contributions are welcome! Please read our guidelines and our Code of Conduct

License

This project is licensed under the Apache-2.0 License.

du2016/aws-node-termination-handler

AWS Node Termination Handler

Gracefully handle EC2 instance shutdown within Kubernetes

Project Summary

Major Features

Instance Metadata Service Processor

Queue Processor

Which one should I use?

Kubernetes Compatibility

Installation and Configuration

Installation and Configuration

Pod Security Admission

Kubectl Apply

Helm

Infrastructure Setup

1. Create an SQS Queue:

2. Create an ASG Termination Lifecycle Hook:

3. Tag the Instances:

4. Create Amazon EventBridge Rules

5. Create an IAM Role for the Pods

Installation

Pod Security Admission

Helm

Single Instance vs Multiple Replicas

Kubectl Apply

Use with Kiam

Metadata endpoints

Building

Metrics

Communication

Contributing

License