Skip to content

A Kubernetes DaemonSet to gracefully delete pods 2 minutes before an EC2 Spot Instance gets terminated

License

Notifications You must be signed in to change notification settings

engineering-bjs/kube-spot-termination-notice-handler

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

92 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

A Kubernetes DaemonSet to run 1 container per node to periodically polls the EC2 Spot Instance Termination Notices endpoint. Once a termination notice is received, it will try to gracefully stop all the pods running on the Kubernetes node, up to 2 minutes before the EC2 Spot Instance backing the node is terminated.

Installation

Helm

A helm chart has been created for this tool, and at time of writing was in the stable repository.

$ helm install stable/k8s-spot-termination-handler

Available docker images/tags

Tags denotes Kubernetes/kubectl versions. Using the same version for your Kubernetes cluster and spot-termination-notice-handler is recommended. Note that the -1 (or similar) is the revision of this tool, in case we need versioning.

  • kubeaws/kube-spot-termination-notice-handler:1.8.5-1
  • kubeaws/kube-spot-termination-notice-handler:1.9.0-1
  • kubeaws/kube-spot-termination-notice-handler:1.10.11-2
  • kubeaws/kube-spot-termination-notice-handler:1.11.3-1
  • kubeaws/kube-spot-termination-notice-handler:1.12.0-2
  • kubeaws/kube-spot-termination-notice-handler:1.13.7-1

Why use it

  • So that your kubernetes jobs backed by spot instances can keep running on another instances (typically on-demand instances)

How it works

Each spot-termination-notice-handler pod polls the notice endpoint until it returns a http status 200. That status means a termination is scheduled for the EC2 spot instance running the handler pod, according to my study).

Run kubectl logs against the handler pod to watch how it works.

$ kubectl logs --namespace kube-system spot-termination-notice-handler-ibyo6
This script polls the "EC2 Spot Instance Termination Notices" endpoint to gracefully stop and then reschedule all the pods running on this Kubernetes node, up to 2 minutes before the EC2 Spot Instance backing the node is terminated.
See https://aws.amazon.com/jp/blogs/aws/new-ec2-spot-instance-termination-notices/ for more information.
`kubectl drain minikubevm` will be executed once a termination notice is made.
Polling http://169.254.169.254/latest/meta-data/spot/termination-time every 5 second(s)
Fri Jul 29 07:38:59 UTC 2016: 404
Fri Jul 29 07:39:04 UTC 2016: 404
Fri Jul 29 07:39:09 UTC 2016: 404
Fri Jul 29 07:39:14 UTC 2016: 404
...
Fri Jul 29 hh:mm:ss UTC 2016: 200

Building against a specific version of Kubernetes

Run KUBE_VERSION=<your desired k8s version> make build to specify the version number of k8s/kubectl.

Slack Notifications

Introduced in version 0.9.2 of this application (the @mumoshu version), you are able to setup a Slack incoming web hook in order to send slack notifications to a channel, notifying the users that an instance has been terminated.

Incoming WebHooks require that you set the SLACK_URL environmental variable as part of your PodSpec.

You can also set SLACK_CHANNEL to send message to different slack channel insisted of default slack webhook url's channel.

The URL should look something like: https://hooks.slack.com/services/T67UBFNHQ/B4Q7WQM52/1ctEoFjkjdjwsa22934

Slack Setup:

Show where things are happening by setting the CLUSTER environment variable to whatever you call your cluster. Very handy if you have several clusters that report to the same Slack channel.

Example Pod Spec:

        env:
          - name: POD_NAME
            valueFrom:
              fieldRef:
                fieldPath: metadata.name
          - name: NAMESPACE
            valueFrom:
              fieldRef:
                fieldPath: metadata.namespace
          - name: SLACK_URL
            value: "https://hooks.slack.com/services/T67UBFNHQ/B4Q7WQM52/1ctEoFjkjdjwsa22934"
          - name: SLACK_CHANNEL
          - value: "#devops"
          - name: CLUSTER
            value: development

Sematext Cloud Event Notifications

The Sematext Cloud event URL is different for Europe and USA and includes the application token for your monitored App.

Sematext Setup:

Show where things are happening by setting the CLUSTER environment variable to whatever you call your cluster. Very handy if you have several clusters that report to the same Slack channel.

Example Pod Spec:

        env:
          - name: POD_NAME
            valueFrom:
              fieldRef:
                fieldPath: metadata.name
          - name: NAMESPACE
            valueFrom:
              fieldRef:
                fieldPath: metadata.namespace
          - name: SEMATEXT_URL
            value: "https://event-receiver.sematext.com/APPLICATION_TOKEN/event"
          - name: CLUSTER
            value: development
          - name: DETACH_ASG
            value: "true"

Wechat Notifications

Incoming WebHooks require that you set the WECHAT_URL and WECHAT_KEY environmental variables as part of your PodSpec.

The URL should look something like: https://pushbear.ftqq.com/sub?key=3488-876437815599e06514b2bbc3864bc96a&text=SpotTermination&desp=SpotInstanceDetainInfo

Wechat Setup:

Example Pod Spec:

        env:
          - name: POD_NAME
            valueFrom:
              fieldRef:
                fieldPath: metadata.name
          - name: NAMESPACE
            valueFrom:
              fieldRef:
                fieldPath: metadata.namespace
          - name: WECHAT_URL
            value: "https://pushbear.ftqq.com/sub"
          - name: WECHAT_KEY
            value: "3488-876437815599e06514b2bbc3864bc96a"
          - name: CLUSTER
            value: development
          - name: DETACH_ASG
            value: "true"

AutoScaling Detachment

This feature currently only supports simple autoscaling - no spot fleet or similar.

If you set the environment variable DETACH_ASG to any value other than false, the handler will detach the instance from the ASG, which may bring a replacement instance up sooner.

The autoscaling group name is automatically detected by the handler.

Credits

kube-spot-termination-notice-handler is a collaborative project to unify @mumoshu and @kylegato's initial work and @egeland's fork with various enhancements and simplifications.

The project is currently maintained by:

  • @egeland
  • @kylegato
  • @mumoshu

About

A Kubernetes DaemonSet to gracefully delete pods 2 minutes before an EC2 Spot Instance gets terminated

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages

  • Shell 78.6%
  • Python 15.3%
  • Dockerfile 6.1%