scriptoriumadmin/GitNexus
1
0
Fork 0
mirror of https://github.com/abhigyanpatwari/GitNexus.git synced 2026-08-28 05:25:25 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 9

fix(eval): size the buffered-fallback budget for the real graph index

Browse source 
trusted_gitnexus_runtime_mounts and sanitized_graph both hand the ~290 MiB
graph index through TaskAssetSnapshot.materialize(), which reflinks into
each arm clone or falls back to a buffered copy under a 16 MiB budget.
Neither ext4 (CI runners) nor 9p (this container) support FICLONE, so
every real materialization fell to the buffered path and blew the budget
instantly (CI run 29750271566).

Raises MAX_BUFFERED_FALLBACK_BYTES to 512 MiB - well above the real index
size, still a full 4x below MAX_TASK_ASSET_BYTES so a genuinely oversized
declaration still fails closed.
This commit is contained in:
Gergo Magyar 2026-07-20 14:42:50 +00:00
parent e3ed82d162
commit c723d420d5
2 changed files with 29 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

21
eval/tests/test_task_assets.py
View file

@ -113,6 +113,27 @@ def test_small_assets_use_a_bounded_buffered_fallback(monkeypatch, tmp_path: Pat
assert (clone / "second").read_bytes() == b"def"
def test_default_buffered_fallback_budget_covers_a_realistic_large_asset(
monkeypatch,
tmp_path: Path,
) -> None:
# 20 MiB exceeds the old 16 MiB default but must fit comfortably under
# the current default, proving the real (non-monkeypatched) budget
# constant is sized for a realistic large sandbox_copy asset such as the
# harness's own pre-built graph index, not just tiny fixtures.
payload = os.urandom(20 * 1024 * 1024)
repo, task = _repo_and_task(tmp_path, {"large": payload})
clone = tmp_path / "clone"
clone.mkdir()
monkeypatch.setattr(task_assets, "_try_reflink", lambda *_args: False)
with TaskAssetCache(tmp_path / "cache") as cache:
snapshot = cache.prepare(task, repo=repo, resolved_sha=SHA)
snapshot.materialize(clone)
assert (clone / "large").read_bytes() == payload
def test_large_asset_without_reflink_fails_before_publish_and_cleans_staging(
monkeypatch,
tmp_path: Path,

11
eval/workflow_bench/task_assets.py
View file

@ -39,9 +39,14 @@ MAX_TASK_ASSET_ENTRIES = 100_000
MAX_TASK_ASSET_PATH_BYTES = 4_096
MAX_TASK_ASSET_BYTES = 2 * 1024 * 1024 * 1024
# A filesystem without reflink support may still run tiny fixtures. Large
# assets fail closed instead of silently returning to one full copy per arm.
MAX_BUFFERED_FALLBACK_BYTES = 16 * 1024 * 1024
# The largest known real sandbox_copy asset in this harness is the shipped
# index above (~428 MiB estimated, ~290 MiB measured); budget comfortably
# above that so it can still materialize via buffered copy on a filesystem
# that cannot reflink (ext4 CI runners, 9p-backed dev mounts), while staying
# well below MAX_TASK_ASSET_BYTES so a genuinely oversized or malformed
# declaration still fails closed instead of silently paying for a slow full
# copy.
MAX_BUFFERED_FALLBACK_BYTES = 512 * 1024 * 1024
COPY_CHUNK_BYTES = 1024 * 1024
# linux/fs.h: #define FICLONE _IOW(0x94, 9, int)