Harden the executable build cells against uv-cache corruption and disk exhaustion#544
Merged
Merged
Conversation
A corrupted uv cache (hardlink-fallback race) can install a mixed-version litellm that only an import catches; the extras check now purges the cache and re-syncs once before failing the cell. And the onefile build leaves multi-GB intermediates that have filled a runner's disk while the smoke models restored, after a successful build; drop them before the models come down, after the Windows Nuitka cache save. Both cost a 2.5h cell rerun on dev721 for what is now a seconds-scale retry or a free deletion.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Problem
Two runner failure modes each cost a 2.5 hour cell rerun on the last release. A corrupted uv cache (the hardlink-fallback race) installed a litellm whose files came from two different versions, which only the import check catches; and the onefile build's multi-GB intermediates filled the runner's disk while the smoke models restored, after the build itself had succeeded.
Solution
The extras import check now purges the uv cache and re-syncs once before failing the cell, in both the release and CUDA build workflows. The release build drops the Nuitka intermediates (keeping the final binary) before restoring smoke models, after the Windows Nuitka cache save so nothing cache-worthy is lost; ccache is untouched since its post-job hook still saves it.