Please multi-node docker DNS instructions

[borg286]

I'm trying to follow the instructions for local docker here
https://github.com/kubernetes/kubernetes/blob/master/docs/getting-started-guides/docker.md
For DNS setup it refers you to the multi-node docker setup for DNS. There is has the user set up some environment variables, one of which is
export KUBE_SERVER=10.10.103.250 # your master server ip, you may change it
For single-node docker setups this should be 127.0.0.1

I request that a note in the above page mention in the DNS section that the user should swap in this export instead of the one meant for multi-node setups
export KUBE_SERVER=127.0.0.1 # your master server ip, you may change it

A side note, I tried localhost and got the following error
error: couldn't read version from server: Get localhost:8080: unsupported protocol scheme "localhost"

[a-robinson]

Thanks @borg286 -- would you be up for making this change in the docs and submitting a pull request?

[borg286]

Sure.

[luxas]

Sorry for distrurbing, but I don't think that's the case.
If you submit 127.0.0.1 to KUBE_SERVER, it will point to localhost of the pod, not the node. The only case when localhost of a pod is the same as the node localhost is when setting hostNetwork: true on the pod definition. So, as you mentioned, setting KUBE_SERVER to 127.0.0.1 will not work, not even in the single-node docker case. You must use the public ip address of the node.

However, when serviceAccounts is properly merged and included in releases, this line and KUBE_SERVER) is unnecessary.
To check, run kubectl get serviceaccounts. If there are one secret for the default serviceaccount, it should work without the -kube-master-url to kube2sky.

Cheers!

[borg286]

What is the approximate timeline for serviceAccounts to get merged in?
Contrary to what you said, setting KUBE_SERVER to 127.0.0.1 did seem to
work. I recognize now that it should be the internal address. How do I get
that internal address?

On Tue, Jan 5, 2016 at 7:25 AM, Lucas Käldström [email protected]
wrote:

Sorry for distrurbing, but I don't think that's the case.
If you submit 127.0.0.1 to KUBE_SERVER, it will point to localhost of the
pod, not the node. The only case when localhost of a pod is the same as
the node localhost is when setting hostNetwork: true on the pod
definition. So, as you mentioned, setting KUBE_SERVER to 127.0.0.1 will
not work, not even in the single-node docker case. You must use the
public ip address of the node.

However, when serviceAccounts is properly merged and included in
releases, this line
https://github.com/kubernetes/kubernetes/blob/master/docs/getting-started-guides/docker-multinode/skydns-rc.yaml.in#L51
and KUBE_SERVER) is unnecessary.
To check, run kubectl get serviceaccounts. If there are 2 secrets for the
default serviceaccount, it should work without the -kube-master-url to
kube2sky.

Cheers!

—
Reply to this email directly or view it on GitHub
#19228 (comment)
.

Integral z-squared dz
from 1 to the cube root of 3
times the cosine
of three pi over 9
equals log of the cube root of 'e'.

[luxas]

Soon #17213 is merged, I hope. But v1.1.3 has this PR already merged in, so that would work.
But instead another issue appears. The apiserver cert isn't signed for 10.0.0.10, so DNS without KUBE_SERVER won't work until we fix that.

I tested setting KUBE_SERVER to 127.0.0.1 and it didn't work. I don't know what kind of setup you have. localhost:8080 in the pod will refer to the exechealthz port.

[andreikop]

Setting KUBE_SERVER to localhost:8080 produces error described at #12534

If I understand correctly, now there are no API server accessible by kube2sky container in single-node docker mode.
I tried to use https://10.0.0.1, which is available if kube-proxy is running. But kube2sky needs https certificates. Probably this problem will be fixed by #17213

So, now docker single node tutorial doesn't work. @luxas, I'd suggest to roll back @6ab098ec56c371a9b81fce611af1bf5f7685df6e and include it to your #17213 PR to avoid problems for next users.

See also #18626

[bgrant0607]

See also #19227, #16087

[davidopp]

This is marked 1.2 with no assignee.

[davidopp]

@fgrzadkowski maybe you can take a look at this?

[luxas]

@davidopp I'm working on this

[fgrzadkowski]

Thanks @luxas

[timbunce]

I don't know if this is relevant here, but the master/docs/getting-started-guides/docker-multinode/skydns-*.yaml.in lag behind the master/cluster/addons/dns/skydns-*.yaml.in files. E.g. the former skydns-rc.yaml.in refers to gcr.io/google_containers/kube2sky:1.11 but the later refers to 1.12. I noticed this when I stumbled into #12534.

[luxas]

@timbunce I have a fix for this, #21504, and are working on docker-multinode UX that probably will start DNS somewhat automatically #21646

[timbunce]

That's great news. Thanks @luxas.