Perhaps this is out of place, but I'd like to ask anyway. I've been using 5.10 and 5.15 kernels w/ these AMIs for over a year, but I noticed that the GPU AMI was not compatible with these kernels. AFAIK the nvidia-gpu-operator was not able to initialize nodes. Is this being considered as well?

@mhulscher I don't think the GPU operator specifically supports Amazon Linux at this point, though that support is being worked on as far as I know. My understanding is that the operator installs pre-compiled kernel modules, and I assume you'd run into issues if there was no such module compiled against the 5.10/5.15 headers. Just a guess; I'd probably open an issue with the operator maintainers.

@cartermckinnon any updates on this PR? Do you have an estimate as to when it will land in 1.24?

Thanks!

We're currently building our own AMI that pulls in this change so we'd love to see it merged.

I'll probably file a separate issue for this, but it would be nice for 5.15 to be included as an option that we can pick at build time just so I don't have to maintain a (minor) fork to test later kernels on our nodes.

@RainofTerra yep, I plan to make that change once we get this merged 👍

We're still working on some performance testing (we were delayed around the holidays, apologies). We don't expect to find any notable differences between 5.4 and 5.10, we're just trying to get a baseline and cover our bases. I hope to share as much of that analysis as we can here. My expectation is this will be merged and released on the order of weeks, not months; I'll update here as possible.

can't wait to see this merged. :-)

oh dang - is it happening?!

Hurrah!

We’re ready to merge this! We compared the 5.4 and 5.10 kernels’:

1. Conformance, using Kubernetes' end-to-end integration test suite, the aws-k8s-tester, and long-running synthetic workloads.
   1. We observed no functional difference between the two kernels when it comes to Kubernetes.
2. Performance, using stress-ng and netperf.
   1. We observed both marginal increases and decreases across the test cases; overall representing equivalent performance.

We’re excited for users to leverage the features of 5.10 in their clusters! We encourage users to monitor their workloads for differences when upgrading, and to open an issue here if something seems off.

Thanks, all!

@cartermckinnon based on the release notes am I correct in thinking this didn't make it into v20230203?

xref - aws/containers-roadmap#1570

@stevehipwell Correct, this will land in the next release. Should be in the next week or so.