Skip to content

[v26.2.x] reactor: exempt 6.18 from auto-overprovisioned only from 6.18.45 - #328

Merged
travisdowns merged 1 commit into
redpanda-data:v26.2.xfrom
travisdowns:td-eevdf-exempt-6.18.45-v26.2.x
Nov 18, 2026
Merged

travisdowns merged 1 commit into
redpanda-data:v26.2.xfrom
travisdowns:td-eevdf-exempt-6.18.45-v26.2.x

Conversation

@travisdowns

Copy link
Copy Markdown
Member

Backport of #327.

The CIAINFRA-2890 EEVDF hang is only fixed from 6.18.45.

The 6.12 bound is left at 6.12.94 even though the 6.12.y fix
(e98394f7bc91) is in 6.12.11, no real need to loosen this.

Add a uname test for the exemption against the kernel strings involved.

(cherry picked from commit cf7b253)

The CIAINFRA-2890 EEVDF hang is only fixed from 6.18.45.

The 6.12 bound is left at 6.12.94 even though the 6.12.y fix
(e98394f7bc91) is in 6.12.11, no real need to loosen this.

Add a uname test for the exemption against the kernel strings involved.

(cherry picked from commit cf7b253)

Copilot AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🟢 Approval recommended

The kernel exemption-threshold backport has no active review issues.

No open issues 🎉

What changed in this PR

Backports #327 to v26.2.x, keeping the Kubernetes auto-overprovisioning workaround enabled for Linux 6.18 releases before 6.18.45.

Changes:

  • Raises the 6.18 exemption threshold from 6.18.38 to 6.18.45, leaving 6.12.94 unchanged.
  • Documents the relevant kernel fix and adds exemption tests for boundary versions and distribution-specific kernel strings.
File Description
tests/​unit/​uname_test.cc Adds coverage for exempt and non-exempt kernel versions.
src/​core/​reactor.cc Updates the 6.18 exemption threshold and explains the kernel fix.

Review effort: Balanced

@StephanDollberg

Copy link
Copy Markdown
Member

I am not entirely fond of backporting this as it can potentially massively change the perf profile which is something I wouldn't want to do in a backport.

Realistically this will affect approximately no-one but not sure.

@travisdowns

Copy link
Copy Markdown
Member Author

@StephanDollberg

I am not entirely fond of backporting this as it can potentially massively change the perf profile which is something I wouldn't want to do in a backport.

I think it's the opposite? Not changing this produces can produce massive swings, see INC-3073.

Customer is on 6.18.20, your mitigation is in place, then they upgrade to 6.18.44 then they are suddenly hit by the EEVDF bug and performance collapses.

I don't think materially increases the exposure to "massive perf changes" in the way you suggest, as folks are generally already going to be getting this setting/mitigation "on the left" as they upgrade through versions. This only moves the boundary slightly to avoid collapse.

@travisdowns
travisdowns merged commit 29ebaa7 into redpanda-data:v26.2.x Nov 18, 2026
36 of 37 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants