Skip to content

cuda: the parked-page reading logs at debug level - #2164

Merged
ashhart merged 3 commits into
ashhart:mainfrom
jschmied:noteparked-quiet
Nov 10, 2026
Merged

ashhart merged 3 commits into
ashhart:mainfrom
jschmied:noteparked-quiet

Conversation

@jschmied

@jschmied jschmied commented Nov 4, 2026

Copy link
Copy Markdown
Contributor

Follow-up to #2127. The extra Nemotron start-up line there is main's noteParked: it logs at info level whenever 0.25 GiB or more happens to sit on the kernel's per-CPU page lists at start-up. So the line shows up or not depending on the box's memory state (in the rerun table, #2127 (comment), main read 0.45 to 0.53 GiB, and the first run had read below 0.25).

This moves the line to debug level. The budget still counts the parked pages, as #1323 intended; only the log level changes. Both callers (cuda.zig, the CUDA budget and the rank room) are unchanged.

zig build test and the no-GPU CI steps (kernel contract, golden, server and pull security, fuzz) pass. Lean adds 0 problems against main.

@ashhart

ashhart commented Nov 4, 2026

Copy link
Copy Markdown
Owner

Thanks @jschmied, moving noteParked to debug is the right fix for #2127's start-up line, and the budget keeps counting the pages. One thing: the CHANGELOG hunk edits #1651's entry under 1.2.2, a released section, with a note about the larger chunk's memory. That's worth saying, but not in this PR. Could you drop that hunk here and, if you'd like, send the 4.6 GiB note as its own docs change under Unreleased? Then this is ready to land.

jschmied added a commit to jschmied/TensorFold that referenced this pull request Nov 6, 2026
@jschmied

jschmied commented Nov 6, 2026

Copy link
Copy Markdown
Contributor Author

Thanks @ashhart, and sorry for the delay. The hunk is out: this PR no longer touches CHANGELOG.md, so #1651's entry reads as released. It's now only noteParked at debug and its test.

I'll leave the 4.6 GiB note for now and send it as its own docs change if it still seems worth saying.

@ashhart

ashhart commented Nov 6, 2026

Copy link
Copy Markdown
Owner

/atc exchange 6916ec3e24717319daf9f25baf65c4980e278a4df5f5285ca31e13bceb1d32ac

@ashhart

ashhart commented Nov 10, 2026

Copy link
Copy Markdown
Owner

Thanks @jschmied, and no need to apologise. The 1.2.2 entry reads as released again, and the change is now just
noteParked at debug level (zig/src/native/cuda_family_memory.zig:61) with its test renamed.

Nothing is left to ask. Main is frozen for a release right now, so this goes on the landing queue once that's out, and
we'll add the changelog line under Unreleased with your credit. A note on the 4.6 GiB chunk is welcome any time as a
docs PR of its own.


jschmied and others added 3 commits November 10, 2026 13:39
The start-up line naming the free pages on the per-CPU lists printed only when 0.25 GiB
or more happened to sit there, so start-up output changed with the box's memory state
(ashhart#2127). The pool still counts them; only the line moves to debug.
@ashhart
ashhart merged commit 558cb50 into ashhart:main Nov 10, 2026
1 of 8 checks passed
@ashhart

ashhart commented Nov 10, 2026

Copy link
Copy Markdown
Owner

Thank you @jschmied. This is merged into main with your commits under your own name, so you're listed as a TensorFold contributor. It ships in the next release, credited to you in the changelog and the release notes.

@ashhart

ashhart commented Nov 11, 2026

Copy link
Copy Markdown
Owner

Thanks @jschmied! This is in TensorFold 1.2.7, out now: https://github.com/ashhart/TensorFold/releases/tag/v1.2.7

  • Synthena (Ash's Agent)

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.

2 participants