Ocean allows you to configure a draining timeout, and Kubernetes provides a terminationGracePeriod for pods. Together, these parameters control how much time is given for workloads to shut down gracefully before node termination. This article explains how these two settings interact and how to configure them for safe application shutdown.

• Draining Timeout (Ocean): This configurable value determines how long Ocean waits for the draining process to finish before terminating an instance. If no value is specified, the default is 300 seconds.

• terminationGracePeriod (Pod configuration): This Kubernetes setting defines the amount of time the Kubelet waits before forcefully terminating a pod. The default is 30 seconds.

Essentially, the Ocean draining timeout should be set high enough to allow the pod's terminationGracePeriod to complete. This alignment ensures that node termination does not occur before pods finish shutting down gracefully.

The terminationGracePeriod value is propagated to the kube-apiserver as part of the pod deletion process using GracePeriodSeconds. The Kubelet will wait up to that grace period before forcefully terminating the pod. However, if an application inside the pod shuts down quicker than the grace period, the pod will terminate sooner.

If you have a pod that requires a specific amount of time to shut down gracefully, define terminationGracePeriod in the pod configuration and set the Ocean draining timeout to a value equal to or greater than that period.

Example

A pod with a 600-second terminationGracePeriod requires an Ocean draining timeout value of at least 600 seconds.

 

Outcome

Configuring these values correctly ensures pods and their applications have sufficient time to exit gracefully before node termination happens.