harald/nixcfg
1
0
Fork 0
Code Issues Pull requests 2 Projects Releases Packages Activity Actions

fix(cache-warm): build each target's initrd before its toplevel closure #73

Open
hermes wants to merge 1 commit from hermes/t-d497f1e2-cachewarm-initrd into main AGit
pull from: hermes/t-d497f1e2-cachewarm-initrd
merge into: harald:main
Conversation 2 Commits 1 Files changed 2 +25 -1
hermes commented 2026-11-01 18:41:56 +01:00
Contributor
Copy link
No description provided.
fix(cache-warm): build each target's initrd before its toplevel closure
Some checks failed
nix flake check / flake-check (pull_request) Failing after 4m32s
Details
ad737eff3c
hermes commented 2026-11-01 18:44:58 +01:00
Author
Contributor
Copy link

Why this change: live halo cache-warm.service 01.11 00:42/00:43 UTC bursted

error: unable to download .../nar/hl37....nar: HTTP 200 (curl error: Transferred a partial file)
error: path ... hl37...-initrd-linux-6.18.60 is required, but there is no substituter that can build it

and the same for sgx's 1856... initrd, during the attic and sgx targets. Both attics copies verify live now: full narinfo + accepted nix path-info --store, NAR decompresses to exactly NarSize with matching NarHash; repeated and 16-concurrent downloads from halo complete. The nix build still built the initrds locally and the attic/sgx targets pushed, so the rc=1 of that night came from the unrelated llama-cpp target; but the substitute line is what the acceptance wants gone.

Mechanism: a fresh-kernel initrd arrives as a bare reference of the target toplevel being substituted; in a store that never built it there is no attached drv, so a failed NAR fetch cannot fall back (the existing nix module's fallback comment documents this class from the 31.07 sgx open-webui NAR). Warming the target's initialRamdisk attr in its own nix build first attaches that drv and makes its output local before the toplevel closure walk. Measured 1.7 s with the path already local; a healthy cache makes it a substitution.

Verification: nix fmt, deadnix --fail, generated cache-warm script sh -n for halo/amd/sgx, and real builds of sgx and attic initialRamdisk plus sgx toplevel from this tree all green. CI flake-check pending.

Why this change: live halo `cache-warm.service` 01.11 00:42/00:43 UTC bursted error: unable to download .../nar/hl37....nar: HTTP 200 (curl error: Transferred a partial file) error: path ... hl37...-initrd-linux-6.18.60 is required, but there is no substituter that can build it and the same for sgx's 1856... initrd, during the attic and sgx targets. Both attics copies verify live now: full narinfo + accepted `nix path-info --store`, NAR decompresses to exactly NarSize with matching NarHash; repeated and 16-concurrent downloads from halo complete. The `nix build` still built the initrds locally and the attic/sgx targets pushed, so the rc=1 of that night came from the unrelated llama-cpp target; but the substitute line is what the acceptance wants gone. Mechanism: a fresh-kernel initrd arrives as a bare reference of the target toplevel being substituted; in a store that never built it there is no attached drv, so a failed NAR fetch cannot fall back (the existing nix module's fallback comment documents this class from the 31.07 sgx open-webui NAR). Warming the target's initialRamdisk attr in its own `nix build` first attaches that drv and makes its output local before the toplevel closure walk. Measured 1.7 s with the path already local; a healthy cache makes it a substitution. Verification: `nix fmt`, `deadnix --fail`, generated `cache-warm` script `sh -n` for halo/amd/sgx, and real builds of sgx and attic initialRamdisk plus sgx toplevel from this tree all green. CI flake-check pending.
hermes force-pushed hermes/t-d497f1e2-cachewarm-initrd from ad737eff3c
Some checks failed
nix flake check / flake-check (pull_request) Failing after 4m32s
Details
to 16a08e59a1
All checks were successful
nix flake check / flake-check (pull_request) Successful in 4m23s
Details
2026-11-01 18:48:40 +01:00 Compare
hermes commented 2026-11-01 18:54:40 +01:00
Author
Contributor
Copy link

CI on head 16a08e59 is green (nix flake check, run #75, 4m23s; local run also all checks passed!). The head differs from the first push only in treefmt: systems/x86_64-linux/amd/default.nix's virbr-rialo port list came in on base 89c31315 unformatted and the treefmt check rejected it, so nix fmt . (0 changed elsewhere) was committed alongside the module change. Diff vs origin/main: module 21 additions, amd 5 (one line reshaped). PR awaits Harald's merge.

CI on head 16a08e59 is green (`nix flake check`, run #75, 4m23s; local run also `all checks passed!`). The head differs from the first push only in treefmt: `systems/x86_64-linux/amd/default.nix`'s `virbr-rialo` port list came in on base 89c31315 unformatted and the treefmt check rejected it, so `nix fmt .` (0 changed elsewhere) was committed alongside the module change. Diff vs origin/main: module 21 additions, amd 5 (one line reshaped). PR awaits Harald's merge.
All checks were successful
nix flake check / flake-check (pull_request) Successful in 4m23s
Details
This pull request can be merged automatically.
This branch is out-of-date with the base branch
You are not authorized to merge this pull request.
View command line instructions

Checkout

From your project repository, check out a new branch and test the changes.
git fetch -u origin +refs/pull/73/head:hermes/t-d497f1e2-cachewarm-initrd
git switch hermes/t-d497f1e2-cachewarm-initrd
Sign in to join this conversation.
Reviewers
No reviewers
Labels
Clear labels
No items
No labels
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
harald/nixcfg!73
No description provided.