Kubernetes best practices: upgrading your clusters with zero downtime | Hacker News Reader

Kubernetes best practices: upgrading your clusters with zero downtime

cloudplatform.googleblog.com

79 pointsrbanffy22 comments

Unfortunately, this is NOT really about Kubernetes - this is about GKE and GKE specifically. Nothing in this article except for the very general idea ("setup HA for masters then upgrade one master at a time") is useful for non-Google installations.

As someone who recently upgrade a GKE cluster with two mouse clicks, this whole post gave me a laugh. Basically an ad veiled as a blog post

It is an ad

"In this episode of Kubernetes Best Practices, let’s take a look at how Google Kubernetes Engine can make upgrading your Kubernetes cluster painless."

This is the chief complaint I have about the Kubernetes community - vendors are vying to lock up the operational complexity in their commercial solution then leave no breadcrumbs to self-implementation.

Notice that nothing in that post can be used outside of Google except for higher level order of operations.

As a comment somewhere a week back said: Kubernetes seems like a platform designed for consultants.

Author here. Using Node Pools to upgrade your cluster isn't something that some people are aware is possible. Same with using Regional Clusters to ensure HA (as it just came out of beta). Sure, the rolling update is one-click (or even zero-clicks with automated upgrades) on GKE, but I wanted to let people know there is another method as well.

Check out the other posts in the series[1] if you feel this one is too much like an ad and let me know what you think. Trying to keep the series a good mix of OSS and GKE specific things with the majority being OSS.

[1] https://www.youtube.com/playlist?list=PLIivdWyY5sqL3xfXz5xJv...

To be fair, this is on the GCP blog, so it's targeted to a GCP audience, not to the broader Kubernetes community. The headline could be more specific, but I think that's really more an issue of the OP's choice in titling or the lack of context provided by HN for subdomains.

It probably also applies to EKS which is launching on Twitch at 2pm EST today

... link? (edit: thanks!)

Found it on Reddit https://m.twitch.tv/events/-5UFZXuCQEK9X7avTUh_Mg

Where did you hear that?

Got an email yesterday that pretty much said everything but "we're launching EKS tomorrow". It's really the only container-related announcement they could be making.

Uh. Sorry, did I get it right, Amazon is launching EKS for public access (as in "general availability") today/tomorrow?

Today at 2PM EST on twitch.tv/aws

The email just said "We can hardly contain ourselves. Tune in tomorrow for a live announcement about running containers on AWS." But that almost certainly means EKS.

Unfortunately Twitch is blocked by Ye Olde Corporate Firewalle.

Any other place to view it?

It's on now, announcement is that EKS is generally available, they said so within the first 30 seconds. Anonymous source did not disappoint!

Just heard that Control plane is priced at $0.20/hr, yikes... that's not bad for HA masters in every AZ, but I think it's certainly going to have a hard time competing with "you don't pay for masters at all, on GKE"

Blog post author here. This is the only post in the series that is GKE specific, really strange it's the only one that ended up on Hacker News. I actually thought it would be the least popular.

You can see the other six episodes here: https://www.youtube.com/playlist?list=PLIivdWyY5sqL3xfXz5xJv...

It's a good series. Too bad people didn't take a look at the others before dragging it down.

This is just a GKE advert

https://blog.manifold.co/migrating-to-kubernetes-with-zero-d... << a less GKE specific article

Does this mean federation is solid enough to call GA / Production ready ?

No, federation is still dead in the water.

its all about GKE , running multi-region masters for HA!

Reply on news.ycombinator.com