|
| 1 | +# SIG Node Meeting Notes |
| 2 | + |
| 3 | +## Jan 14, 2025 |
| 4 | + |
| 5 | +* \[Tim Allclair\] Changing the CRI contract for `UpdateContainerResources`, to require it to not intentionally restart a container. Redefine (and probably rename) the "NotRequired" resize restart policy to match, meaning containers won't be deliberately restarted (could still trigger an OOM). Runtimes that can't support in-place resize should return an error (e.g. VM runtimes). |
| 6 | + |
| 7 | +## Jan 7, 2025 |
| 8 | + |
| 9 | +* \[Tim Allclair\] Shrinking memory limits bellow memory usage (in-place resize) |
| 10 | + * See [In Place Pod Resize - Handling Memory Limit Decreases](https://docs.google.com/document/d/1cEFLXKwNOSNLAkzyhoJUgkBW0OiX-9bXB_aJV7OAypw/edit?tab=t.0) |
| 11 | + * \[sergey\] to not interrupt the flow posting some questions: |
| 12 | + * should we disable resize on cgroupv1 to avoid diff behavior? |
| 13 | + * Mrunal: true, but we like the v1 behavior better |
| 14 | + * How VPA can use the feature safely? This is another example of VPA cannot guarantee non-disruptive pod resize (best effort) |
| 15 | + * Tim to follow up with VPA folks |
| 16 | + * All checks suggested still have possibility for the race condition, the user experience will be super annoying. Should we reconsider the application of memory.high? |
| 17 | + * Mrunal: Ask Waiman (Red Hat memory subsystem maintainer) if switching to v1 behavior is possible |
| 18 | + * Peter: we probably need to do this to fully avoid TOCTOU / OOM kills |
| 19 | + * Tim: Pause container to avoid TOCTOU? |
| 20 | +* \[Tim Allclair\] Pod generation, status generation (design proposal) |
| 21 | + * See proposal section of[\[PUBLIC\] Design proposal: Pod ResizeStatus Improvements](https://docs.google.com/document/d/10m0vdWbjqF_q_f1_N3gVOD5YeNn09BwFcVVycG_MoEY/edit?tab=t.0#heading=h.3rtsrq1ip5xv) |
| 22 | +* \[Zeel Patel\] [https://github.com/kubernetes/kubernetes/issues/129447](https://github.com/kubernetes/kubernetes/issues/129447) |
| 23 | + * Create a KEP PR: [https://github.com/kubernetes/enhancements/pull/5022](https://github.com/kubernetes/enhancements/pull/5022) |
| 24 | + * \[fromani\] I’d be happy to review |
| 25 | +* \[HirazawaUi\]: Addressing the Issue of Slow Pod State Transitions When Evented PLEG is Enabled |
| 26 | + * Due to the time zone difference, I regret that I am unable to attend the meeting. I have compiled the entire process of how @hshiina and I attempted to address this issue, as well as my proposal, into a detailed document: [https://docs.google.com/document/d/1TPrY56q9MNW8r1FuzKDFkBBhOjQ0hqi7wJAbIP1O-4g](https://docs.google.com/document/d/1TPrY56q9MNW8r1FuzKDFkBBhOjQ0hqi7wJAbIP1O-4g). I sincerely hope this can be discussed during the meeting. |
| 27 | +* \[SergeyKanzhelev\] [Kubernetes feature development and Container runtimes](https://docs.google.com/document/d/1y42XrUPrm-6DZby1RQjexYYoNn822IRR6igsOiy_62c/edit?tab=t.0#heading=h.2bbangbf1ha4) |
| 28 | + * State containderd and cri-o explicitly |
| 29 | +* \[kannon92\] [Deprecation of NodeFeature in sig-node testing](https://github.com/kubernetes/kubernetes/pull/129166) |
| 30 | +* \[marquiz\] KEP-4112 Pass down resources to CRI, request for review: |
| 31 | + [https://github.com/kubernetes/enhancements/pull/4113](https://github.com/kubernetes/enhancements/pull/4113) |
| 32 | +* 1.33 time is coming\! [https://github.com/kubernetes/sig-release/pull/2706/](https://github.com/kubernetes/sig-release/pull/2706/files) |
0 commit comments