-
Notifications
You must be signed in to change notification settings - Fork 1.6k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add resource awareness for node #1963
Comments
It's not clear to me that lxcfs is actually a reasonable approach here, or that kind is a good approach to running things with serious resource management (e.g. most users probably have swap enabled which would ordinarily cause kubelet to refuse to run). The system reserved approach seems possibly acceptable |
I agree https://github.com/lxc/lxcfs did not handle SWAP precisely, and I agree kind does not serious depends on resources awareness today. But I think it's a good feature for users who use kind that he/she can also uses metrics server against kind cluster -- isn't it cool to experiment more functions on kind? Right, maybe, more options need be explored for right approach... |
Sure but experimenting is possible without agreeing to ship something in the project 😅 I commented upstream about another approach. |
there's discussion in google/cadvisor#2699 and in SIG Node slack. There's significant pushback against having any way to tell cAdvisor what the actual host limits are w/o relying on auto-detection and against letting cAdvisor read it from the cgroup. To a certain extent it may still be possible to implement a bad version of this by setting a computed system-reserved at the kubelet level. Or by bind mounting static fakes / using something similar to lxcfs (bringing on their own problems...). |
we may have a new opportunity in the future kubernetes/enhancements#2371 |
/remove-kind design |
This is really a dupe of the earlier issue #877 Worth noting that kubelet seems to be moving in the direction of CRI providing the stats, not necessarily with cAdvisor. |
What would you like to be added:
Resources awareness of a container node comes from the host, it's not same as the resources limited by the cgroup. There is a PR tracked in
cadvisor
to discuss how to hack it google/cadvisor#2699, I recently find a projectlxcfs
https://github.com/lxc/lxcfs which seems meet all the requirements.Why is this needed:
With this implemented, we can use metrics to monitor a container nodes resources capability and usability.
The text was updated successfully, but these errors were encountered: