Natan Yellin(@aantn) 's Twitter Profileg
Natan Yellin

@aantn

Take Kubernetes monitoring to the next level with Prometheus and @RobustaDev

ID:17523402

linkhttps://robusta.dev calendar_today20-11-2008 21:48:32

2,8K Tweets

5,5K Followers

1,2K Following

Follow People
Natan Yellin(@aantn) 's Twitter Profile Photo

A question for the Kubernetes pros: Is it possible to prevent a COMPLETED Pod from being deleted? (Or at least delaying the deletion until I can read the Pod logs?)

In more detail: an app runs a job which creates a pod. The app needs to read logs but the pod is gone. Fixable?

account_circle
Natan Yellin(@aantn) 's Twitter Profile Photo

Reduce your Kubernetes spend by running one cli command.

We released a new KRR version that fixes some stability issues and runs faster than ever. If you have feedback or are missing a feature, DM me directly or leave a comment. Help us make it easy.

github.com/robusta-dev/krโ€ฆ

account_circle
Natan Yellin(@aantn) 's Twitter Profile Photo

More work on reducing Prometheus memory consumption - this time by deduping labels. Has anyone tried enabling this in production? What have you seen?

github.com/prometheus/proโ€ฆ

account_circle
Natan Yellin(@aantn) 's Twitter Profile Photo

What do people think of CEL? (Common expression language.)

Have been playing around with it a little and was reading github.com/kubernetes/enhโ€ฆ

account_circle
Micah Hausler(@micahhausler) 's Twitter Profile Photo

Every node is authorized to list all other pods and nodes in the cluster. One compromised node can result in enumeration of all other pods and nodes in the cluster.

account_circle
Ian Coldwater ๐Ÿ“ฆ๐Ÿ’ฅ(@IanColdwater) 's Twitter Profile Photo

A pod in Kubernetes is a group of containers. A pod is also the name for a group of whales. Get it? Like Docker whales? ๐Ÿณ๐Ÿ“ฆ

account_circle
Tim Hockin (thockin.yaml)(@thockin) 's Twitter Profile Photo

zamboni Sam Rose Natan Yellin Running DNS __in the cluster__ was ALWAYS meant to be an example of DNS integration, not a full solution. Like many such cases, there is nothing more durable than a quick hack.

IM(NS)HO, the way GKE does DNS offload to CloudDNS is _much_ better, especially for managed offerings.

account_circle
Michael Hausenblas(@mhausenblas) 's Twitter Profile Photo

Andy Randall ๐Ÿ‡บ๐Ÿ‡ฆ Patrick Magee Natan Yellin You are, as usual, correct. Indeed, early versions also included Klippy, a proactive AI based scheduler. I remember fondly Klippy asking โ€œIt looks like youโ€™re trying to schedule a pod here, can I offfer you a node โ€ฆโ€

account_circle
Andy Randall ๐Ÿ‡บ๐Ÿ‡ฆ(@ahrkrak) 's Twitter Profile Photo

Michael Hausenblas Patrick Magee Natan Yellin Had to be VB to quickly mock up the UI. Most people have forgotten that a pod spec was defined via a Windows dialog until we got yaml support in the 1.02 release.

account_circle
David Flanagan(@rawkode) 's Twitter Profile Photo

Natan Yellin Workers used to be called Minions, most likely due to the influence of Salt; which was the original deployment mechanism.

account_circle
Natan Yellin(@aantn) 's Twitter Profile Photo

How does kubectl really work?

You can trace any kubectl command to see what API calls it makes.

Here is a great post from Thomas Stringer on the technique. This is how I figured out which columns prints and how.

trstringer.com/kubernetes-apiโ€ฆ

account_circle