Add ability to specify existing etcd cluster in options. about dcos-kubernetes-quickstart HOT 6 CLOSED

Thanks for the feedback @jzampieron. This is not on our near-term roadmap, and your feedback is noted. We will consider it for future releases.

from dcos-kubernetes-quickstart.

Nick,
Can we leave this ticket open as an enhancement request.

Alternately, who do we escalate to to ensure this makes it on the near-term roadmap.

Expecting customers to waste substantial additional cluster resources and introduce additional ops workload is unacceptable for production environments. There are major cost implications for anyone hosting in public cloud environments.

Are pull requests welcome?

Thanks,
-JZ

from dcos-kubernetes-quickstart.

@jzampieron from a technical perspective, Kubernetes officially supports, as in validates through extensive end-to-end testing, specific versions of etcd and container runtimes, such as Docker. We too need to make such a commitment and, for now, that's only possible if the package manages the dependencies.

We also believe that enabling users to maintain their own etcd clusters, while it makes sense from the resource usage perspective, wouldn't provide the UX quality we are working hard to deliver here. This etcd cluster should only be used by this framework.

from dcos-kubernetes-quickstart.

Pull-requests are most welcome! But please understand we are still in the early days and there may be some delay in coming back to you. Thank you!

from dcos-kubernetes-quickstart.

@pires Thanks for the detailed reasoning. Understood that kube wants its own etcd cluster.

Perhaps we can enhance the example directions to clarify that point since it's apparently very important.

How is the docker version requirement handled since DC/OS has similar requirements?

from dcos-kubernetes-quickstart.

Perhaps we can enhance the example directions to clarify that point since it's apparently very important.

Are you saying you are opening a PR? Sweet! 😉

How is the docker version requirement handled since DC/OS has similar requirements?

For the sake of resource isolation, we are not using Docker from the agent but rather a somewhat advanced technique that a lot of people refer to as docker-in-docker.
We are aware this is not the perfect solution but only recently did the Kubernetes community stabilize the CRI API, and even more recently are we seeing more stable implementations being released, e.g. cri-o, that we expect to use instead of Docker soon.

from dcos-kubernetes-quickstart.