4 ms·

None if that explains why it’s so, how do I put it, fiddly? A flexible tool could still have reasonable defaults. If you rent a managed K8s cluster from AWS, yo

None if that explains why it’s so, how do I put it, fiddly? A flexible tool could still have reasonable defaults. If you rent a managed K8s cluster from AWS, you’re of course fine, but why does deploying it yourself have to be so complicated that the recommendation is now to use a “Kubernetes distribution” like K3s. Why can’t it just be a daemon you install using standard mechanisms?

lbriner 5y ago
Probably because it was designed for scale and big companies that can afford lots of hardware don't need to invest on standalone instances.
KptMarchewa 5y ago
There are 3 node clusters on RPi and 15000 node bare metal clusters. For which will you optimize?
- codeflo 5y ago
  First of all, are you sure there isn't a way to improve the ergonomics of one without harming the other? Second of all, to the extent that there is a tradeoff, this is essentially a solved problem in every other kind of daemon (from web servers to databases). Make the simple case secure and simple, while making the complicated case (which will require a lot of custom configuration and reading the documentation anyway, not matter what you do) as straightforward as possible.
pas 5y ago
Because it was quickly thrown together in an experimential/evolving language, in a design by committee way, by a very non-diverse group. And it's getting much better at ergonomics/UX/DX (go is also getting much better), but it's very similar to OpenStack. (Python 2, resource hog, impossibly fragile, basic setup works, but uselessly incomplete without certain critical 3rd party components - eg. without Ceph and ovs it was just too big of a mess, similarly the CNI stuff and the LoadBalancers for k8s are the important missing batteries.) That said both microk8s (snap packaged by Canonical) and kind (kubernetes in docker) are turnkey solutions. k3s is more of a vanguard :) .. also, it can be (is already) packaged into a simple RPM/DEB easily, like the GitLab Omnibus.

lbriner 5y ago

Probably because it was designed for scale and big companies that can afford lots of hardware don't need to invest on standalone instances.

KptMarchewa 5y ago

There are 3 node clusters on RPi and 15000 node bare metal clusters. For which will you optimize?

codeflo 5y ago
First of all, are you sure there isn't a way to improve the ergonomics of one without harming the other? Second of all, to the extent that there is a tradeoff, this is essentially a solved problem in every other kind of daemon (from web servers to databases). Make the simple case secure and simple, while making the complicated case (which will require a lot of custom configuration and reading the documentation anyway, not matter what you do) as straightforward as possible.

pas 5y ago

Because it was quickly thrown together in an experimential/evolving language, in a design by committee way, by a very non-diverse group. And it's getting much better at ergonomics/UX/DX (go is also getting much better), but it's very similar to OpenStack. (Python 2, resource hog, impossibly fragile, basic setup works, but uselessly incomplete without certain critical 3rd party components - eg. without Ceph and ovs it was just too big of a mess, similarly the CNI stuff and the LoadBalancers for k8s are the important missing batteries.) That said both microk8s (snap packaged by Canonical) and kind (kubernetes in docker) are turnkey solutions. k3s is more of a vanguard :) .. also, it can be (is already) packaged into a simple RPM/DEB easily, like the GitLab Omnibus.