fix: prepare v1.0.10.post5 with pool-mode failure, numa.h prereq, doc fixes - #301
Open
drunkcoding wants to merge 1 commit into
Open
fix: prepare v1.0.10.post5 with pool-mode failure, numa.h prereq, doc fixes#301drunkcoding wants to merge 1 commit into
drunkcoding wants to merge 1 commit into
Conversation
… fixes Bundle the three still-open release blockers ahead of a v1.0.10.post5 tag so a single release ships them together: - fix(#299): the pool-mode batch-failure paths called a non-existent storage.update_batch(), which raised AttributeError instead of failing the batch (leaving it stuck). Use the canonical update_batch_status(batch_id, BatchStatus.FAILED, error=...) and keep the documented capacity_exceeded / batch_failed codes in the error string. Add a regression test covering both paths (isolated import, no CUDA needed). - infra(#166): install_deps.sh now fails early if libnuma headers (numa.h) are missing, and INSTALL.md / manual-installation.md list libnuma-dev as a prerequisite for the first-launch core-engine JIT. - docs(#300): fix the MMLU-Pro example path (test/ -> tests/e2e/). Also bump the setup.py version fallback and INSTALL.md Option C wheel pins to 1.0.10.post5 (batchgen_kernels 0.3.3+sm90a). Signed-off-by: leyang.xue <leyang.xue@ed.ac.uk>
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Description
Prepares a
v1.0.10.post5release by bundling the three still-open blockers found in the full fresh re-verification of thev1.0.10.post4release docs, and bumping the version fallback + Option C wheel pins topost5.mainalready carries the fixes for the other release-tree failures (#283–#290 via #291–#298, #277/#278, #279/#280), but no tag ships them and thepost4release only uploaded 2 of 5 wheels. This PR closes the remaining gaps so a singlepost5tag is installable end-to-end.Closes #299, closes #300, closes #166.
Motivation
Verified against
origin/main@d171a15:_process_batch_pool_mode(L910) and_wait_and_finalize_batch(L977) calledself.storage.update_batch(...), which does not exist (StorageManageronly hasupdate_batch_status). Both are reachable pool-mode failure paths (intake-pool capacity rejection; batch timeout), so instead of marking the batchfailedthe scheduler task raisedAttributeErrorand the batch was left non-terminal. Fixed to the canonicalupdate_batch_status(batch_id, BatchStatus.FAILED, error=...), preserving the documentedcapacity_exceeded/batch_failedcodes in the error string. Adds a regression test.posix_shm.cpp(#include <numa.h>) at first launch; only the Dockerfile installedlibnuma-dev, so every fresh bare-metal install (incl. Hopper) died at firstlaunch_http_server. Adds an earlyinstall_deps.shprerequisite check and documentslibnuma-dev.client-api.mdpointed attest/r1_mmlu_pro_test/...; the file is attests/e2e/r1_mmlu_pro_test/....Milestone
None.
Type of Change
modelkernelcorefix— narrow bug fix (+ a regression test) ← primaryinfra— build / CI / packaging / scripts (version bump,install_deps.sh)docs— INSTALL.md / manual-installation.md / client-api.mdFile changes
batchgen/server/batch_scheduler.pyupdate_batch_status(BatchStatus.FAILED, …)on both pool-mode failure pathstests/test_pool_mode_failure_status.pyscripts/install_deps.shnuma.his missing, with an apt hintdocs/INSTALL.mdlibnuma-devprereq; Option C wheel pins →post5(batchgen_kernels 0.3.3+sm90a)docs/manual-installation.mdlibnuma-devprereqdocs/client-api.mdtest/→tests/e2e/setup.py1.0.10.post5Release checklist (after merge, before tagging
v1.0.10.post5)bash scripts/build_wheels.sh --output-dir ./wheels(Hopper, CUDA 12.8) → build all 5 wheels.gh release create v1.0.10.post5and upload all 5 wheels (post4 shipped only 2) +SHA256SUMS.Checklist
2 passed).PR Merge Policy Contract — pre-merge checklist
git diff --stat origin/mainreviewed; every file traces to the task — no unrelated files (§3.1).debug_* / check_* / scratch_* / tmp_*scripts added (§1.1).test_*.pyadded inside the runtime package; the test is undertests/(§1.2, §2.1).BATCHGEN_*env-var debug guard (§1.3).Notes
ruff I001import-sort warnings atbatch_scheduler.py:3,318,1244are untouched (onmainalready) — left out of scope per §3 diff discipline.RELEASE_DOC_VERIFICATION_v1.0.10.post4_rerun-2026-07-04.md(attached in the worktree).