Twitter | Search | |
Brian Grant
Kubernetes and GKE
275
Tweets
153
Following
2,257
Followers
Tweets
Brian Grant Nov 7
Yes, pods are critical to getting operational concerns out of application containers and enabling the 1 app per container paradigm.
Reply Retweet Like
Brian Grant Nov 7
Alloc was already present in the very first commit of the Borg API at the beginning of 2004.
Reply Retweet Like
Brian Grant Nov 7
Replying to @jbeda @littleidea and 5 others
Yes, dynamic port allocation had pervasive impact, on scheduling, configuration, discovery, monitoring, health checking, load balancing, various proxies, authentication, network isolation, and probably other things.
Reply Retweet Like
Brian Grant Nov 7
Replying to @jbeda @littleidea and 5 others
Yes, there were other capabilities that required IPs rather than reusing host networking.
Reply Retweet Like
Brian Grant Nov 7
Replying to @thockin @jbeda and 5 others
"AllocVM" (not the usual notion VM). Alloc was an array of AllocVMs. SchedulingUnit was also prototyped in Borg and implemented in a couple other components/systems. It came out of a discussion with a senior SRE about how users used Allocs.
Reply Retweet Like
Brian Grant Oct 2
"Multi-workload"?
Reply Retweet Like
Brian Grant Sep 7
Yeah, I just use a normal VM, which I can then set up and snapshot and reuse or replicate.
Reply Retweet Like
Brian Grant Aug 29
Yes, I just lost a lot of respect for Techcrunch
Reply Retweet Like
Brian Grant Aug 12
Happy to see this move forward. Ubiquitous, standardized metrics format inside Google yielded huge benefits to cloud-native application management: observability facilitates control.
Reply Retweet Like
Brian Grant Aug 9
Replying to @josebiro
But, yes, many Omega ideas live on in : separate store, async controllers, generic resources, taints, tolerations, forgiveness, disruption budgets, ... I was more targeting Borg with pods (aka scheduling units), labels, and watch.
Reply Retweet Like
Brian Grant Aug 9
Replying to @josebiro
Hey! It was the K8s logo for a while, but people confused it with our bot. I then tried to upload a photo, but it was the wrong resolution or something, and I decided not to spend time on that. :-)
Reply Retweet Like
Brian Grant Aug 9
Congrats to Prometheus, the 2nd project in CNCF and the 2nd to graduate. Origins of Prometheus and its entry into CNCF were covered here:
Reply Retweet Like
Brian Grant Aug 7
Relaxed JSON and HJSON have some of those features, but chose to go the other way on commas and quoting string values.
Reply Retweet Like
Brian Grant Aug 3
Replying to @try_except_
Yes, that's also why we created kubedash:
Reply Retweet Like
Brian Grant Aug 3
Replying to @jrnt30
You can enforce this via maxLimitRequestRatio.
Reply Retweet Like
Brian Grant Aug 1
Replying to @bgrant0607
There are some blog posts (e.g., ) which have led to a misunderstanding of ResourceQuota, which bounds aggregate usage by all pods within a namespace, as well as limits on numbers of objects.
Reply Retweet Like
Brian Grant Aug 1
PSA: Use pod resource requests to prevent too many pods from being scheduled on the same node, resource limits to ensure containers don't use unlimited amounts of resources, and LimitRange to apply minimum and maximum requests and limits to each pod within a namespace
Reply Retweet Like
Brian Grant Jul 31
Replying to @mattfarina
Yes, kustomize and Helm are apples and oranges. Encapsulation/abstraction, dependency management, and providing an all-inclusive end-to-end solution are non-goals for kustomize. kustomize is a composable tool for configuration customization.
Reply Retweet Like
Brian Grant retweeted
Oren Teich Jul 24
Replying to @teich
My goal: run anywhere. Any language. Any software. Any code, from a function to an app to a container. Fully managed, serverless, with a great DX. We've got a long way to go. I'm excited about these steps today.
Reply Retweet Like
Brian Grant Jul 24
Great to see finally unveiled. Serverless with the right primitives. A huge leap forward. Congratulations!
Reply Retweet Like