Improve Kubernetes Fault Tolerance

Kubernetes: Until today, I don't know this. It's crazy. Because you can control pod eviction time.

First, let's understand this.

Why does a Pod even get evicted?

Maybe because of:

1. Resource pressure on the node

2. Pods exceeding resource limits.

Fair enough?

Now how to control eviction?

By setting evictionPressureTransitionPeriod in Kubelet config.

Let's say to 120 seconds.

Hmm, what does it do?

It enables a 120 second grace period before a Pod is automatically evicted.

An evicted Pod in a deployment, goes to a new node.

Is it Good?

Ofcourse, it's stupendously useful for your app in production.

This will support system fault tolerance.

Alternatively, on your Dev/QA

set "evictionPressureTransitionPeriod: 0s" to speed up eviction process.

Restart Kubelet after making changes.

That's it for today.

If this is useful, do a Repost. It really helps ♻️

I'm

Mutha Nagavamsi

. Follow me for Kubernetes, Devops and tech content.

Before you leave, don't forget to SMILE 😁

---

Thank you so much for reading this. If you found it interesting, do spread the word about it. You may also find my other content interesting, find them below.

1. Mutha Nagavamsi on Youtube. (Subscribe, it really helps)

2. Me on Medium.

3. Me on Twitter.