Report resource limits and usage of Kubelet and other daemons

[dchen1107]

Issue #147 was filed for QoS tiers. But to really guarantee QoS for services / PODs / Containers, we need to enforce daemons' resource limits, at least tracking their usages. Currently cAdvisor is running inside a docker container, we could put Kubelet and other daemons into their own docker containers, then rely on cAdvisor to monitor and report their usages.

[brendandburns]

Getting the kubelet inside a container will be tricky, as it needs to talk to Docker (do-able, but tricky). Likewise, getting the service proxy in is a good idea.

--brendan

[thockin]

The real trick will be getting docker into a container.

On Sun, Aug 3, 2014 at 9:05 PM, brendandburns notifications@github.com
wrote:

Getting the kubelet inside a container will be tricky, as it needs to talk
to Docker (do-able, but tricky). Likewise, getting the service proxy in is
a good idea.

--brendan

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

[bgrant0607]

/cc @proppy

[bgrant0607]

@ArtfulCoder

[dchen1107]

We are running master components in pod, which resource usage will be picked up by cAdvisor and heapster automatically. But we need put kubelet, kube-proxy and docker to cgroups, so that cAdvisor can monitor their usages. cc/ @vmarmol

[vishh]

Is this required for v1.0?

On Fri, Mar 27, 2015 at 11:49 AM, Dawn Chen notifications@github.com
wrote:

We are running master components in pod, which resource usage will be
picked up by cAdvisor and heapster automatically. But we need put kubelet,
kube-proxy and docker to cgroups, so that cAdvisor can monitor their
usages. cc/ @vmarmol https://github.com/vmarmol

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

[ArtfulCoder]

I spoke offline with Victor.
I can try and put kube-proxy in a pod. I can atleast do a quick test to make sure that it works.
#5419

[dchen1107]

Both Kubelet and Kube-proxy are in resource container already.

[ArtfulCoder]

ah , ok. so, nothing to be done here..

On Wed, Apr 29, 2015 at 10:10 PM, Dawn Chen notifications@github.com
wrote:

Both Kubelet and Kube-proxy are in resource container already.

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

[vmarmol]

The argument was that if we got the Kube-proxy in a container we'd get the
starts automatically. Otherwise we need to export them in Heapster. Since
we were looking to place the Kube-proxy in a container, this route may not
be too bad.
On Apr 29, 2015 10:12 PM, "Abhi Shah" notifications@github.com wrote:

ah , ok. so, nothing to be done here..

On Wed, Apr 29, 2015 at 10:10 PM, Dawn Chen notifications@github.com
wrote:

Both Kubelet and Kube-proxy are in resource container already.

—
Reply to this email directly or view it on GitHub
<
#490 (comment)

.

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

[ArtfulCoder]

putting in a pod, we would get the stats just the way we get it for other
pods..
Is that ok with you Dawn ?

On Wed, Apr 29, 2015 at 10:17 PM, Victor Marmol notifications@github.com
wrote:

The argument was that if we got the Kube-proxy in a container we'd get the
starts automatically. Otherwise we need to export them in Heapster. Since
we were looking to place the Kube-proxy in a container, this route may not
be too bad.

On Apr 29, 2015 10:12 PM, "Abhi Shah" notifications@github.com wrote:

ah , ok. so, nothing to be done here..

On Wed, Apr 29, 2015 at 10:10 PM, Dawn Chen notifications@github.com
wrote:

Both Kubelet and Kube-proxy are in resource container already.

—
Reply to this email directly or view it on GitHub
<

#490 (comment)

.

—
Reply to this email directly or view it on GitHub
<
#490 (comment)

.

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

[dchen1107]

@ArtfulCoder We need resource container for docker no matter what. @vmarmol is doing that. We can put kube-proxy into a pod, but not kubelet itself. Kubelet can run as a container, but not as pod today. There is a chicken egg issue anyway.