Thanks to visit codestin.com
Credit goes to github.com

Skip to content

Conversation

markturansky
Copy link
Contributor

See https://github.com/GoogleCloudPlatform/kubernetes/blob/master/pkg/kubelet/kubelet.go#L1836

When cadvisor is unavailable, the node gets 0 CPU and RAM. Capacity['pods'] does not exist.

Because there is no pod capacity, Kubelet running on OS X and using boot2docker cannot schedule any pods. All fail because there are no pod resources available.

I don't know if this is the correct solution, but it fixes running on OS X.

@vishh
Copy link
Contributor

vishh commented May 26, 2015

Why is max pods set to 0 on OSX? Are you running a kubelet built from HEAD?

@markturansky
Copy link
Contributor Author

@vishh yes, HEAD.

Capacity['pods'] wasn't set at all, which was the problem.

@vishh
Copy link
Contributor

vishh commented May 26, 2015

May be the fix is to set max pods when cadvisor is not running?

Overall, making kubelet control a docker daemon on another host doesn't
sound like a use case that is being considered whenever node design
decisions are being made.

On Tue, May 26, 2015 at 1:51 PM, Saad Ali [email protected] wrote:

Assigned #8826
#8826 to @vishh
https://github.com/vishh.


Reply to this email directly or view it on GitHub
#8826 (comment)
.

@markturansky
Copy link
Contributor Author

I think this is a scheduling issue, with the scheduler running in a process in OSX just like Kubelet.

The node has no pod capacity (CPU and memory set to 0 on line 1836). Are you suggesting I move the fix to that section?

I'm looking for advice for how to fix this (if this is a bug worth fixing) :)

@vishh
Copy link
Contributor

vishh commented May 26, 2015

Yes. Can you try setting Capacity[pods] here:
https://github.com/GoogleCloudPlatform/kubernetes/blob/master/pkg/kubelet/kubelet.go#L1836
https://github.com/GoogleCloudPlatform/kubernetes/blob/master/pkg/kubelet/kubelet.go#L1836
?

On Tue, May 26, 2015 at 2:14 PM, Mark Turansky [email protected]
wrote:

I think this is a scheduling issue, with the scheduler running in a
process in OSX just like Kubelet.

The node has no pod capacity (CPU and memory set to 0 on line 1836). Are
you suggesting I move the fix to that section?

I'm looking for advice for how to fix this (if this is a bug worth fixing)
:)


Reply to this email directly or view it on GitHub
#8826 (comment)
.

@markturansky
Copy link
Contributor Author

@vishh It is moved, per your request.

@vishh
Copy link
Contributor

vishh commented May 26, 2015

LGTM.

@vishh vishh added the lgtm "Looks good to me", indicates that a PR is ready to be merged. label May 26, 2015
@thockin
Copy link
Member

thockin commented May 27, 2015

CL fails.

@thockin thockin removed the lgtm "Looks good to me", indicates that a PR is ready to be merged. label May 27, 2015
@thockin
Copy link
Member

thockin commented May 27, 2015

poke when CI passes and we can LGTM then

@markturansky
Copy link
Contributor Author

@thockin Poke. Travis just returned green.

@markturansky
Copy link
Contributor Author

@thockin poke again, rebased, and all returned green (the Jenkins builder flaked last time)

@vishh vishh added the lgtm "Looks good to me", indicates that a PR is ready to be merged. label May 28, 2015
thockin added a commit that referenced this pull request May 28, 2015
Fixed setNodeStatus to allow running on pods on OS X
@thockin thockin merged commit bade36c into kubernetes:master May 28, 2015
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
lgtm "Looks good to me", indicates that a PR is ready to be merged.
Projects
None yet
Development

Successfully merging this pull request may close these issues.

4 participants