Container technologies at Coinbase: Why Kubernetes is not part of our stack
1–10 of 414 posts
Re: Container technologies at Coinbase: Why Kubernetes is not part of our stack
#2This isn't really about the article, but why does it take 6 seconds to load the page.
Re: Container technologies at Coinbase: Why Kubernetes is not part of our stack
#3One thing that is regrettable about K8s winning the orchestration wars so remarkably, is that it pretty much killed all other solutions. Swarm is dead, Nomad doesn't seem like it has much community support and Mesos feels like it's on life support. Mesos still has a lot of people working on it however, but the perception feels different.
Personally I've found Mesos much easier to manage, secure, and operate than k8s. However, when it first came out all the cool kids were using it, then most of them jumped ship to k8s. AirBnB's Chronos is now pretty much a dead project, Mesosphere's Marathon is now gimped (no UI) and major features moved into DCOS. At the same time, Mesosphere (now D2iQ?), now seems like it's more focused on k8s.
k8s is everything plus the kitchen sink, and managed k8s isn't the killer feature I thought it would be. I don't blame people for not jumping on the k8s bandwagon at all.
Re: Container technologies at Coinbase: Why Kubernetes is not part of our stack
#4This isn't really about the article, but why does it take 6 seconds to load the page.
Because: medium.com ...
Re: Container technologies at Coinbase: Why Kubernetes is not part of our stack
#5This isn't really about the article, but why does it take 6 seconds to load the page.
I thought you were exaggerating. You weren't.
Re: Container technologies at Coinbase: Why Kubernetes is not part of our stack
#6The history is a bit fuzzy. The interesting feature introduced in 2.6.24 was PID and network namespaces. Containers were "complete" by Linux 3.8 with user namespaces. Cgroups are not that important to be able to build containers (isolation first). There were other out-of-tree technologies before that, notably VServer and OpenVZ.
Re: Container technologies at Coinbase: Why Kubernetes is not part of our stack
#7What I don't like in these kind of posts is the fact that he seems to know much more than he could.
"For example, most folks that run large-scale container orchestration platforms cannot utilize their built-in secret or configuration management."
So, with thousands (maybe more?) of "folks" out there, each with a different environment, requirement, background and so on. How can someone knows what most of them can or cannot do? Sure there is a lot of them posting about it in blogs and showing in conferences, but it still a very small subset of all the companies running containers today in the world, and the majority will not expose any info about it, for a myriad of reasons.
Re: Container technologies at Coinbase: Why Kubernetes is not part of our stack
#8tl;dr We are all in on AWS proprietary stuff instead.
Re: Container technologies at Coinbase: Why Kubernetes is not part of our stack
#9One thing that is regrettable about K8s winning the orchestration wars so remarkably, is that it pretty much killed all other solutions. Swarm is dead, Nomad doesn't seem like it has much community support and Mesos feels like it's on life support. Mesos still has a lot of people working on it however, but the perception feels different. Personally I've found Mesos much easier to manage, secure, and operate than k8s.…
I'd agree that k8s has a lot of functionality built-in, another important thing to realise is what k8s doesn't do.
In addition to the well-known integration points (Container Runtime/Network/Storage Interfaces), there's things like the lack of a good built-in user authentication mechanism with Kubernetes, which means you pretty much always need some external authentication service for your clusters.
That's not too bad if your on one of the big managed providers (GKE/AKS/EKS) but can get complex for people who want to deploy on-prem.
Re: Container technologies at Coinbase: Why Kubernetes is not part of our stack
#10One thing that is regrettable about K8s winning the orchestration wars so remarkably, is that it pretty much killed all other solutions. Swarm is dead, Nomad doesn't seem like it has much community support and Mesos feels like it's on life support. Mesos still has a lot of people working on it however, but the perception feels different. Personally I've found Mesos much easier to manage, secure, and operate than k8s.…
Nomad, Consul + Swarm is still a lovely solution and I prefer it a lot to K8s. K8s is a big monolith and often way too complex for my personal use cases.
I hope Hashicorp sooner or later builds a proper replacement for Swarm so that we can have overlay networks without hassle. I know there is Weave, but never tried it.