scriptoriumadmin/GitNexus
1
0
Fork 0
mirror of https://github.com/abhigyanpatwari/GitNexus.git synced 2026-09-10 22:43:40 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 7

refactor(eval): simplify sweep internals and needrestart check (#2785)

Browse source 
Reuse the incumbent skill digest instead of walking the tree twice, and
keep the needrestart grep a literal match that actionlint accepts.

Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
Gergo Magyar 2026-09-03 08:19:28 +00:00
parent 2cc27dcaa8
commit 45f97045e0
2 changed files with 27 additions and 34 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

5
.github/workflows/gitnexus-skill-evolution.yml vendored
View file

@ -171,8 +171,9 @@ jobs:
run: |
set -euo pipefail
needrestart_policy=/etc/needrestart/conf.d/90-gitnexus-evolution.conf
if [[ ! -r "${needrestart_policy}" ]] || ! grep -Fqx '$nrconf{restart} = '\''l'\'';' "${needrestart_policy}"; then
echo "::error::${needrestart_policy} must contain: \$nrconf{restart} = 'l';"
needrestart_line="\$nrconf{restart} = 'l';"
if [[ ! -r "${needrestart_policy}" ]] || ! grep -Fqx "${needrestart_line}" "${needrestart_policy}"; then
echo "::error::${needrestart_policy} must contain: ${needrestart_line}"
exit 1
fi
oom_score_adjustment="$(</proc/self/oom_score_adj)"

56
eval/workflow_bench/runner.py
View file

@ -35,7 +35,7 @@ model).
from __future__ import annotations
import argparse
from concurrent.futures import Future, ThreadPoolExecutor, wait
from concurrent.futures import ThreadPoolExecutor, wait
import hashlib
import json
import os
@ -649,15 +649,12 @@ def systemic_outage_streak(error_kind: str | None, prior_streak: int) -> int:
return prior_streak + 1 if error_kind in SYSTEMIC_ERROR_KINDS else 0
CellOutcome = tuple[dict[str, Any] | None, BaseException | None]
def _run_wave(
wave: Sequence[tuple[int, str]],
*,
workers: int,
run: Callable[[int, str], dict[str, Any]],
) -> list[CellOutcome]:
) -> list[dict[str, Any] | BaseException]:
"""Run one wave on a pool and settle every future, in submission order.
The pool is owned explicitly rather than through ``with``: the context
@ -682,15 +679,11 @@ def _run_wave(
# Nothing left to wait for — every future is done — so this only retires
# the wave's threads instead of leaking one pool's worth per wave.
pool.shutdown(wait=True)
return [_settle(future) for future in futures]
def _settle(future: Future[dict[str, Any]]) -> CellOutcome:
"""A completed future as (record, error) — exactly one of them is set."""
error = future.exception()
if error is not None:
return None, error
return future.result(), None
outcomes: list[dict[str, Any] | BaseException] = []
for future in futures:
error = future.exception()
outcomes.append(error if error is not None else future.result())
return outcomes
def sweep_task_cells(
@ -727,35 +720,31 @@ def sweep_task_cells(
records = [run(run_idx, arm) for run_idx, arm in wave]
else:
outcomes = _run_wave(wave, workers=workers, run=run)
failure = next((error for _, error in outcomes if error is not None), None)
failure = next((outcome for outcome in outcomes if isinstance(outcome, BaseException)), None)
if failure is not None:
# The siblings of the failing cell have already completed and
# spent their budget. Persist their rows, in submission order,
# before the harness bug takes the process down — otherwise a
# crash in one cell silently erases the evidence of the others.
for (run_idx, arm), (record, error) in zip(wave, outcomes, strict=True):
if error is None:
on_record(run_idx, arm, record)
for (run_idx, arm), outcome in zip(wave, outcomes, strict=True):
if not isinstance(outcome, BaseException):
on_record(run_idx, arm, outcome)
raise failure
records = [record for record, _ in outcomes]
wave_tripped = False
records = [outcome for outcome in outcomes if not isinstance(outcome, BaseException)]
for (run_idx, arm), record in zip(wave, records, strict=True):
# Every future in this wave has already completed and incurred its
# cost. Persist all of them in canonical submission order even if
# an earlier row trips the breaker; only later waves are skipped.
on_record(run_idx, arm, record)
if wave_tripped:
continue
for record in records:
outage_streak = systemic_outage_streak(record.get("error_kind"), outage_streak)
if outage_limit and outage_streak >= outage_limit:
wave_tripped = True
if wave_tripped:
print(
f"[systemic-outage] {outage_streak} consecutive session/infra/cleanup "
"failures — aborting the remaining sweep; report and promotion are written "
"from partial evidence and the run exits non-zero."
)
return outage_streak, True
print(
f"[systemic-outage] {outage_streak} consecutive session/infra/cleanup "
"failures — aborting the remaining sweep; report and promotion are written "
"from partial evidence and the run exits non-zero."
)
return outage_streak, True
return outage_streak, False
@ -878,7 +867,8 @@ def run_cell(ctx: TaskCellContext, run_idx: int, arm: str) -> dict[str, Any]:
phase="task setup",
)
if arm in CANDIDATE_ARMS:
assert ctx.candidate_overlay is not None
if ctx.candidate_overlay is None:
raise RuntimeError("candidate overlay is unavailable")
applied_digest = apply_candidate_overlay(
ctx.candidate_overlay,
worktree,
@ -889,7 +879,9 @@ def run_cell(ctx: TaskCellContext, run_idx: int, arm: str) -> dict[str, Any]:
# The digest the model must preserve during its run is the
# post-overlay skill surface (candidate skills for
# candidate arms; unchanged base skills otherwise).
expected_skill_digest = skill_fingerprint(worktree, execution_arm)
expected_skill_digest = (
skill_fingerprint(worktree, execution_arm) if arm in CANDIDATE_ARMS else base_skill_digest
)
orig_sha = _sandbox_git(sandbox, ["rev-parse", "HEAD"]).strip()
if not re.fullmatch(r"[0-9a-fA-F]{40,64}", orig_sha):
raise RuntimeError("sandboxed candidate setup did not produce an immutable commit")