Fix cpu burst test failure on newer kernels#4374
Conversation
1. Rename current -> got, expected -> want. 2. check_cgroup_value: add file name to output. 3. Improve functions description. This is mostly to simplify debugging test failures. Example output before: current 500000 !? 500 After: cpu.max.burst: got 500000, want 500 Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
A kernel bug which resulted in cpu.max.burst value read which is 1000 times smaller than it should be has recently been fixed (see [1]). Adapt the test so it works with either broken or fixed kernel. [1]: https://lore.kernel.org/all/20240424132438.514720-1-serein.chengyu@huawei.com/ Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
|
No need to backport this PR to 1.1 as this functionality (initially added by #3749) was not backported. |
|
I think you should add the commits in this PR to #4360, these two PRs are depended on each other. |
Why do you think they are dependent? They are definitely not: this one is about fixing a test for newer kernel, while #4360 is about removing Go 1.21 and adding Go 1.23. It just happened that we have CI broken in two places now, and each of these PRs fixes one of such places. This is why I opened #4375 -- to see that CI is green again with both this PR and #4360 are in, and for reviewers of both this PR and #4360 to make sure it's green. Note we don't have to have a green CI in order to merge a PR. So a plan is to have two LGTMs on either this one or #4360, merge one, rebase the other. |
rata
left a comment
rata
left a comment
There was a problem hiding this comment.
LGTM, great finding!
I guess the upstream commit will be applied to all stable branches, so maybe in a few weeks/months we can remove this?
|
2 LGTM and @kolyshkin said we don't need to wait for CI to be green to merge this. This, combined with another PR, is what will make the CI green. Furthermore, cirrus (the only failing checks) failed to install dependencies. Merging then. Please let me know if I missed something! |
4 participants
A kernel bug which resulted in cpu.max.burst value read which is 1000
times smaller than it should be has recently been fixed (see 1).
Adapt the test so it works with either broken or fixed kernel.
Fixes: #4373
🔔 For CI, see #4375.