Twitter | Search | |
Brian Grant
Google Cloud. Kubernetes Steering Committee emeritus, K8s SIG Architecture co-Chair emeritus, CNCF TOC member emeritus
783
Tweets
256
Following
6,009
Followers
Tweets
Brian Grant Oct 13
Time flies, but now is the time to migrate, before November 13! A year ago the end of Helm v2 was announced: . "Download links for the Helm 2 client...and...the stable and incubator chart repositories may no longer work". They are scheduled to be GONE.
Reply Retweet Like
Brian Grant Sep 23
Replying to @thockin @detiber and 2 others
Yes, control all the things. I believe has looked at this some.
Reply Retweet Like
Brian Grant retweeted
Drew Bradstock Aug 14
Replying to @uhoelzle
We are incredibly excited to celebrate GKE's 5th bday. All Kubernetes are definitely not created equal and we are quite proud of Google's ongoing contribution to making k8s and especially Anthos & GKE the future OS of the cloud
Reply Retweet Like
Brian Grant Aug 14
GKE launched in 2014: . The first managed Kubernetes service. It has come along way since then: . I always enjoy learning about what our customers do with it:
Reply Retweet Like
Brian Grant Aug 10
Replying to @KarlKFI
I have one of these. I only use it at the 1000 lumens setting, since otherwise the reflection can be blinding. Outdoors, it can light up a huge area. Definitely my go-to flashlight.
Reply Retweet Like
Brian Grant Aug 7
Replying to @tsaha
There are lots of other ways to usefully contribute: behavioral improvements that don't require API changes, performance optimizations, observability and logging improvements, new tests, bug fixes, better error messages, refactoring, libraries, tech debt, docs, blog posts, ...
Reply Retweet Like
Brian Grant Aug 7
Replying to @KarlKFI @thockin
Yes, it takes at least a year to take a new API from inception to GA, and longer to address feedback once people start using it, and then it needs to be maintained forever. So not really feasible for most independent contributors.
Reply Retweet Like
Brian Grant Aug 6
Replying to @KarlKFI
I also wish that was completed. It's even more obvious now that we need a way to register non-pod endpoints, and that we need to monitor system components and add-ons.
Reply Retweet Like
Brian Grant Aug 6
Replying to @embano1
Thanks. I refer to it often. It's one of my shorter documents. I wanted it to be more concise than .
Reply Retweet Like
Brian Grant Aug 5
Replying to @apuchitnis
This doesn't have a Tweet-sized answer. :-) VPA () is part of it, but the level of acceptable performance needs to be specified and measured. Quasar () was an interesting automated approach from a few years ago.
Reply Retweet Like
Brian Grant retweeted
Kubernetes Podcast from Google Aug 4
This week we check in on the Minikube project, with maintainer Thomas Strömberg (). Since reaching 1.0, the team has put a lot of effort into finding and fixing things that block users, especially the new kind! 📃 🔊
Reply Retweet Like
Brian Grant retweeted
Google Open Source Aug 3
Developers often use open source code that originates outside their organization. Since security is always top of mind, we’re happy to announce that Google is joining the Open Source Security Foundation (OpenSSF)!
Reply Retweet Like
Brian Grant Aug 4
As described in , the Kubernetes control plane is designed for automation in ways that many others are not: change notification (watch), optimistic concurrency, extensible admission control, extensible state (annotations), eventual consistency, ...
Reply Retweet Like
Brian Grant retweeted
Kubernetes Podcast from Google Jul 28
Schedule some time soon to listen to this week’s show; a tour of Kubernetes’ scheduling, with David Oppenheimer (). David worked on Borg and Omega, and explains how those systems influenced . 📃 🔊
Reply Retweet Like
Brian Grant Jul 16
Replying to @jimmyzelinskie @quayio
There are tradeoffs. I would factor that into an initContainer, which is similar to what we did in Borg for a long time. Some sysctls can be specified in pod spec (), but this couples the app and the OS. We're missing something like StorageClass for nodes.
Reply Retweet Like
Brian Grant Jul 16
This is pretty amazing, too. If you were ever wondering how to use some GCP load-balancing feature in GKE, this (and the linked docs) probably has the answer: internal load balancing, multi-cluster ingress, Cloud Armor, Identity-aware Proxy, ...
Reply Retweet Like
Brian Grant Jul 16
This is awesome. I've been looking forward to this for a long time. A number of networking-intensive workloads need to tune sysctls for peak performance. This will provide a declarative way to express that.
Reply Retweet Like
Brian Grant Jul 14
~All of the fooz pages are informational. quitquitquit, huphuphup, abortabortabort, and killkillkill impact running processes. I'd guess repeating the command 3 times was to make extra sure you meant it. Or maybe the creator was frustrated :-). May also simplify access control
Reply Retweet Like
Brian Grant Jul 14
Replying to @thockin @jbeda and 2 others
According to a comment by Craig Silverstein on an internal blog post, statusz dates back to 1998. The CL description matches, though the diff is gone.
Reply Retweet Like
Brian Grant Jul 9
Replying to @tsaha
Also, as yet largely under-exploited.
Reply Retweet Like