| 1 | #!/usr/bin/env python3
|
| 2 | """The full task loop, end to end — the product's signature motion.
|
| 3 |
|
| 4 | oak: mount -> edit -> commit -> push -> desc -> end, then REMOUNT for a
|
| 5 | follow-up task (the warm-cache story) and repeat edit/commit/push/end.
|
| 6 | git: worktree add -> branch edit -> add/commit -> push -> worktree remove,
|
| 7 | then a second worktree for the follow-up. The task description lives in
|
| 8 | the commit message git already paid for; oak pays a separate `desc`
|
| 9 | call. That asymmetry is the measurement, not a harness gap
|
| 10 | (docs/command-semantics.md): this is agent-default-track, whole-task
|
| 11 | evidence, never a step-by-step equivalence claim.
|
| 12 |
|
| 13 | Transport honesty: oak rows always cross the network to a real oak server.
|
| 14 | git runs against GitHub (--git-remote github, transport-comparable) or a
|
| 15 | local bare repo (--git-remote local, the core-equivalent track). Rows carry
|
| 16 | remote_transport/remote_server; latency comparisons across transports are
|
| 17 | forbidden — cross-subject evidence at unlike transports is tool calls and
|
| 18 | tokens only.
|
| 19 |
|
| 20 | Setup:
|
| 21 | OAK_BENCH_SYNC_REPO=oak/bench-sync-tmp \\
|
| 22 | [email protected]:oakdotspace/bench-task-loop-tmp.git \\
|
| 23 | python3 scripts/task_loop.py --runs 3
|
| 24 |
|
| 25 | Unconfigured remotes emit skip rows (returncode 77), never silent absence.
|
| 26 | """
|
| 27 |
|
| 28 | from __future__ import annotations
|
| 29 |
|
| 30 | import argparse
|
| 31 | import hashlib
|
| 32 | import json
|
| 33 | import os
|
| 34 | import platform
|
| 35 | import random
|
| 36 | import re
|
| 37 | import shutil
|
| 38 | import socket
|
| 39 | import sqlite3
|
| 40 | import stat
|
| 41 | import subprocess
|
| 42 | import tempfile
|
| 43 | import threading
|
| 44 | import time
|
| 45 | import urllib.error
|
| 46 | import urllib.parse
|
| 47 | import urllib.request
|
| 48 | from datetime import datetime, timezone
|
| 49 | from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
|
| 50 | from pathlib import Path
|
| 51 | from typing import Any
|
| 52 |
|
| 53 | from oakbench import remotes as oakbench_remotes
|
| 54 | from oakbench import tokens as oakbench_tokens
|
| 55 | from oakbench import merge_history_fixture, shallow_refresh
|
| 56 | from oakbench.byteproxy import CountingProxy
|
| 57 | from oakbench.gitserver import GitHTTPServer
|
| 58 | from oakbench.httpcount import RequestCountingProxy
|
| 59 | from oakbench.diskprobe import bounded_tree_usage
|
| 60 | from oakbench.environment import ENV_ISOLATION_VERSION, base_env as oakbench_base_env
|
| 61 | from oakbench.environment import command_display
|
| 62 | from oakbench.execution import run_timed, stop_owned_process
|
| 63 | from oakbench.results import ResultsStore
|
| 64 | from oakbench.rows import row_returncode
|
| 65 | from oakbench.runlock import measurement_lock
|
| 66 | from oakbench.subjects import (
|
| 67 | DEFAULT_OAK_REPO,
|
| 68 | Subject,
|
| 69 | binary_sha256,
|
| 70 | load_subjects,
|
| 71 | source_metadata,
|
| 72 | subject_details,
|
| 73 | subject_version,
|
| 74 | subject_versions,
|
| 75 | )
|
| 76 |
|
| 77 | ROOT = Path(__file__).resolve().parents[1]
|
| 78 | DEFAULT_WORKDIR = Path(tempfile.gettempdir()) / "oak-task-loop"
|
| 79 | SKIP_RETURNCODE = 77
|
| 80 | VIRTUAL_BRANCH_RE = re.compile(r"virtual branch ([^\s)]+)")
|
| 81 | PHASES = ("task", "followup")
|
| 82 | TASK_LOOP_SCENARIO = "task_loop"
|
| 83 | AGENT_TASK_SCENARIO = "agent_task_baseline_v1"
|
| 84 | AGENT_TASK_OAK_LOCAL_SCENARIO = "agent_task_oak_local_v1"
|
| 85 | INTERRUPTED_PUBLICATION_SCENARIO = "interrupted_publication_after_request_v1"
|
| 86 | LEGACY_SELECTED_MATERIALIZATION_SCENARIO = "legacy_selected_clone_materialization_v1"
|
| 87 | SHALLOW_REFRESH_SCENARIO = "shallow_worker_branch_refresh_v1"
|
| 88 | SHALLOW_REFRESH_DEFAULT_FIXTURES = "1000:1,2000:2"
|
| 89 | SHALLOW_REFRESH_TASK_BRANCH = "task"
|
| 90 | SHALLOW_REFRESH_STEP_TIMEOUT_S = 900.0
|
| 91 | COMMITS_INFO_SUFFIX = "/commits/info"
|
| 92 | INTERRUPTED_PUBLICATION_REQUEST_TIMEOUT_S = 2.0
|
| 93 | INTERRUPTED_PUBLICATION_PROCESS_TIMEOUT_S = 30.0
|
| 94 | INTERRUPTED_PUBLICATION_MAX_REQUEST_BYTES = 64 * 1024 * 1024
|
| 95 | LOCAL_OAK_SERVER_READY_TIMEOUT_S = 10.0
|
| 96 |
|
| 97 |
|
| 98 | def legacy_selected_receipt_evidence(receipt: dict[str, Any]) -> dict[str, Any]:
|
| 99 | result = receipt.get("result") if isinstance(receipt.get("result"), dict) else {}
|
| 100 | observed = receipt.get("observed") if isinstance(receipt.get("observed"), dict) else {}
|
| 101 | phases = observed.get("pull_phases") if isinstance(observed.get("pull_phases"), list) else []
|
| 102 | phase_names = [phase.get("name") for phase in phases if isinstance(phase, dict)]
|
| 103 | materialization = (
|
| 104 | observed.get("materialization")
|
| 105 | if isinstance(observed.get("materialization"), dict)
|
| 106 | else {}
|
| 107 | )
|
| 108 | return {
|
| 109 | "legacy_route_observed": phase_names == ["initial", "legacy_selected_branch"],
|
| 110 | "pull_phase_names": phase_names,
|
| 111 | "pull_phases": phases,
|
| 112 | "materialization_passes_observed": materialization.get("passes_observed"),
|
| 113 | "written_paths": materialization.get("written_paths"),
|
| 114 | "snapshot_bound": result.get("snapshot_bound"),
|
| 115 | "head": result.get("head"),
|
| 116 | "manifest": result.get("manifest"),
|
| 117 | }
|
| 118 |
|
| 119 |
|
| 120 | def legacy_selected_receipt_contract(evidence: dict[str, Any]) -> dict[str, Any]:
|
| 121 | phases = []
|
| 122 | for phase in evidence.get("pull_phases", []):
|
| 123 | if not isinstance(phase, dict):
|
| 124 | continue
|
| 125 | phases.append(
|
| 126 | {
|
| 127 | key: value
|
| 128 | for key, value in phase.items()
|
| 129 | if key not in {"materialization_passes"}
|
| 130 | }
|
| 131 | )
|
| 132 | return {
|
| 133 | "legacy_route_observed": evidence.get("legacy_route_observed"),
|
| 134 | "pull_phases": phases,
|
| 135 | "snapshot_bound": evidence.get("snapshot_bound"),
|
| 136 | "head": evidence.get("head"),
|
| 137 | "manifest": evidence.get("manifest"),
|
| 138 | "written_paths": evidence.get("written_paths"),
|
| 139 | }
|
| 140 |
|
| 141 |
|
| 142 | def normalized_cache_contract(snapshot: dict[str, Any] | None) -> list[dict[str, Any]] | None:
|
| 143 | if not isinstance(snapshot, dict) or not isinstance(snapshot.get("rows"), list):
|
| 144 | return None
|
| 145 | return [
|
| 146 | {key: row.get(key) for key in ("path", "size", "blob_hash")}
|
| 147 | for row in snapshot["rows"]
|
| 148 | if isinstance(row, dict)
|
| 149 | ]
|
| 150 |
|
| 151 |
|
| 152 | def normalized_status(value: dict[str, Any] | list[Any] | None) -> dict[str, Any] | None:
|
| 153 | if not isinstance(value, dict):
|
| 154 | return None
|
| 155 | return {
|
| 156 | key: value.get(key)
|
| 157 | for key in ("branch", "parent", "branch_status", "head", "working_changes", "branch_changes")
|
| 158 | }
|
| 159 |
|
| 160 |
|
| 161 | def materialization_timing_controls_met(
|
| 162 | metadata: dict[str, Any], runs: int, randomize_order: bool, controls_requested: bool
|
| 163 | ) -> bool:
|
| 164 | return bool(
|
| 165 | controls_requested
|
| 166 | and metadata.get("measurement_lock") == "held"
|
| 167 | and randomize_order
|
| 168 | and runs >= 10
|
| 169 | )
|
| 170 |
|
| 171 |
|
| 172 | def isolated_oak_fixture_env(task_config_root: Path) -> dict[str, str]:
|
| 173 | task_config_root.mkdir(mode=0o700, parents=True, exist_ok=True)
|
| 174 | retained = {
|
| 175 | "PATH",
|
| 176 | "TMPDIR",
|
| 177 | "TMP",
|
| 178 | "TEMP",
|
| 179 | "HOME",
|
| 180 | "USER",
|
| 181 | "USERNAME",
|
| 182 | "SHELL",
|
| 183 | "SYSTEMROOT",
|
| 184 | "WINDIR",
|
| 185 | "COMSPEC",
|
| 186 | "PATHEXT",
|
| 187 | "SSL_CERT_FILE",
|
| 188 | "SSL_CERT_DIR",
|
| 189 | }
|
| 190 | env = {key: value for key, value in os.environ.items() if key in retained}
|
| 191 | env.update(
|
| 192 | {
|
| 193 | "GIT_AUTHOR_NAME": "Oak Materialization Bench",
|
| 194 | "GIT_AUTHOR_EMAIL": "[email protected]",
|
| 195 | "GIT_COMMITTER_NAME": "Oak Materialization Bench",
|
| 196 | "GIT_COMMITTER_EMAIL": "[email protected]",
|
| 197 | "GIT_CONFIG_GLOBAL": os.devnull,
|
| 198 | "GIT_CONFIG_NOSYSTEM": "1",
|
| 199 | "OAK_AUTHOR": "oak-materialization-bench",
|
| 200 | "OAK_NO_UPDATE_CHECK": "1",
|
| 201 | "OAK_BENCH_ENV_ISOLATION_VERSION": ENV_ISOLATION_VERSION,
|
| 202 | "NO_COLOR": "1",
|
| 203 | "CLICOLOR": "0",
|
| 204 | "LC_ALL": "C",
|
| 205 | "LANG": "C",
|
| 206 | "TZ": "UTC",
|
| 207 | "GIT_AUTHOR_DATE": "2026-01-01T00:00:00Z",
|
| 208 | "GIT_COMMITTER_DATE": "2026-01-01T00:00:00Z",
|
| 209 | }
|
| 210 | )
|
| 211 | return env
|
| 212 |
|
| 213 |
|
| 214 | def normalized_worktree_snapshot(root: Path) -> list[dict[str, Any]]:
|
| 215 | rows: list[dict[str, Any]] = []
|
| 216 | for path in sorted(root.rglob("*")):
|
| 217 | relative = path.relative_to(root)
|
| 218 | if relative.parts and relative.parts[0] in {".oak", ".git"}:
|
| 219 | continue
|
| 220 | info = path.lstat()
|
| 221 | if stat.S_ISDIR(info.st_mode):
|
| 222 | continue
|
| 223 | row: dict[str, Any] = {
|
| 224 | "path": relative.as_posix(),
|
| 225 | "mode": stat.S_IMODE(info.st_mode),
|
| 226 | }
|
| 227 | if stat.S_ISLNK(info.st_mode):
|
| 228 | row.update({"kind": "symlink", "target": os.readlink(path)})
|
| 229 | else:
|
| 230 | row.update({"kind": "file", "sha256": hashlib.sha256(path.read_bytes()).hexdigest()})
|
| 231 | rows.append(row)
|
| 232 | return rows
|
| 233 |
|
| 234 |
|
| 235 | def json_command(
|
| 236 | command: list[str], cwd: Path, env: dict[str, str]
|
| 237 | ) -> dict[str, Any] | list[Any] | None:
|
| 238 | result = run_bounded_harness_command(command, cwd, env=env)
|
| 239 | if result.returncode != 0:
|
| 240 | return None
|
| 241 | try:
|
| 242 | value = json.loads(result.stdout)
|
| 243 | except (TypeError, ValueError):
|
| 244 | return None
|
| 245 | return value if isinstance(value, (dict, list)) else None
|
| 246 |
|
| 247 |
|
| 248 | def normalized_history(value: dict[str, Any] | list[Any] | None) -> list[dict[str, Any]] | None:
|
| 249 | if not isinstance(value, list):
|
| 250 | return None
|
| 251 | rows: list[dict[str, Any]] = []
|
| 252 | for item in value:
|
| 253 | if not isinstance(item, dict):
|
| 254 | return None
|
| 255 | rows.append(
|
| 256 | {
|
| 257 | key: item.get(key)
|
| 258 | for key in (
|
| 259 | "hash",
|
| 260 | "timestamp",
|
| 261 | "branch",
|
| 262 | "description_or_subject",
|
| 263 | "files_changed",
|
| 264 | )
|
| 265 | }
|
| 266 | )
|
| 267 | return rows
|
| 268 |
|
| 269 |
|
| 270 | def normalized_oak_artifacts(root: Path) -> list[str] | None:
|
| 271 | oak_dir = root / ".oak"
|
| 272 | if not oak_dir.is_dir():
|
| 273 | return None
|
| 274 | return sorted(
|
| 275 | path.relative_to(oak_dir).as_posix() + ("/" if path.is_dir() else "")
|
| 276 | for path in oak_dir.rglob("*")
|
| 277 | )
|
| 278 |
|
| 279 |
|
| 280 | def stat_cache_snapshot(root: Path) -> dict[str, Any] | None:
|
| 281 | db = root / ".oak/oak.db"
|
| 282 | if not db.is_file():
|
| 283 | return {"rows": None, "all_metadata_valid": False, "error": "oak_db_missing"}
|
| 284 | try:
|
| 285 | connection = sqlite3.connect(str(db))
|
| 286 | connection.execute("PRAGMA query_only = ON")
|
| 287 | raw_rows = connection.execute(
|
| 288 | "SELECT path, mtime_ns, ctime_ns, size, blob_hash FROM stat_cache ORDER BY path"
|
| 289 | ).fetchall()
|
| 290 | except sqlite3.Error as error:
|
| 291 | return {"rows": None, "all_metadata_valid": False, "error": str(error)}
|
| 292 | finally:
|
| 293 | if "connection" in locals():
|
| 294 | connection.close()
|
| 295 | rows: list[dict[str, Any]] = []
|
| 296 | valid = True
|
| 297 | for path, mtime_ns, ctime_ns, size, blob_hash in raw_rows:
|
| 298 | file_path = root / path
|
| 299 | try:
|
| 300 | info = file_path.stat()
|
| 301 | except OSError:
|
| 302 | valid = False
|
| 303 | continue
|
| 304 | expected_ctime = info.st_ctime_ns if os.name == "posix" else 0
|
| 305 | metadata_valid = (
|
| 306 | int(size) == info.st_size
|
| 307 | and int(mtime_ns) == info.st_mtime_ns
|
| 308 | and int(ctime_ns) == expected_ctime
|
| 309 | )
|
| 310 | valid = valid and metadata_valid
|
| 311 | rows.append(
|
| 312 | {
|
| 313 | "path": str(path),
|
| 314 | "size": int(size),
|
| 315 | "blob_hash": str(blob_hash),
|
| 316 | "metadata_valid": metadata_valid,
|
| 317 | }
|
| 318 | )
|
| 319 | return {"rows": rows, "all_metadata_valid": valid}
|
| 320 |
|
| 321 |
|
| 322 | def base_env() -> dict[str, str]:
|
| 323 | return oakbench_base_env(
|
| 324 | author_name="Oak Task Loop",
|
| 325 | author_email="[email protected]",
|
| 326 | oak_author="oak-task-loop",
|
| 327 | )
|
| 328 |
|
| 329 |
|
| 330 | def parse_args() -> argparse.Namespace:
|
| 331 | parser = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
|
| 332 | parser.add_argument("--subjects", help="Comma-separated subject names from config/subjects.toml")
|
| 333 | parser.add_argument("--config", type=Path, default=ROOT / "config" / "subjects.toml")
|
| 334 | parser.add_argument("--git-bin", type=Path)
|
| 335 | parser.add_argument("--oak-installed-bin", type=Path)
|
| 336 | parser.add_argument("--oak-local-bin", type=Path)
|
| 337 | parser.add_argument(
|
| 338 | "--oak-serve-bin",
|
| 339 | type=Path,
|
| 340 | help="Immutable Oak binary used only as the fixed loopback server and fixture writer.",
|
| 341 | )
|
| 342 | parser.add_argument(
|
| 343 | "--oak-serve-source",
|
| 344 | type=Path,
|
| 345 | help="Read-only source checkout that produced --oak-serve-bin.",
|
| 346 | )
|
| 347 | parser.add_argument(
|
| 348 | "--oak-serve-source-head",
|
| 349 | help="Exact 64-hex Oak source head that produced --oak-serve-bin.",
|
| 350 | )
|
| 351 | parser.add_argument(
|
| 352 | "--materialization-baseline-subject",
|
| 353 | help="Named control subject required by the legacy materialization scenario.",
|
| 354 | )
|
| 355 | parser.add_argument(
|
| 356 | "--materialization-timing-controls-met",
|
| 357 | action="store_true",
|
| 358 | help=(
|
| 359 | "Record that harness-verifiable controls are met only with a held measurement lock, "
|
| 360 | "randomized subject order, and at least 10 runs. This is not a speedup claim: "
|
| 361 | "separate A/A noise and externally verified quiet-host evidence remain required."
|
| 362 | ),
|
| 363 | )
|
| 364 | parser.add_argument(
|
| 365 | "--shallow-refresh-fixtures",
|
| 366 | default=SHALLOW_REFRESH_DEFAULT_FIXTURES,
|
| 367 | help=(
|
| 368 | "shallow_worker_branch_refresh_v1 merge-history fixtures as COMMITS:SEED[,...]; "
|
| 369 | "the first is primary, the rest are held-out scaling cases."
|
| 370 | ),
|
| 371 | )
|
| 372 | parser.add_argument(
|
| 373 | "--injected-request-delay-ms",
|
| 374 | type=float,
|
| 375 | default=0.0,
|
| 376 | help=(
|
| 377 | "shallow_worker_branch_refresh_v1: fixed sleep the loopback proxy adds before each "
|
| 378 | "HTTP request (labeled injected_request_delay_ms; not a shaped network)."
|
| 379 | ),
|
| 380 | )
|
| 381 | parser.add_argument("--runs", type=int, default=3)
|
| 382 | parser.add_argument(
|
| 383 | "--scenarios",
|
| 384 | default=TASK_LOOP_SCENARIO,
|
| 385 | help=(
|
| 386 | "Comma-separated scenarios: task_loop, agent_task_baseline_v1, "
|
| 387 | "agent_task_oak_local_v1, legacy_selected_clone_materialization_v1, "
|
| 388 | "shallow_worker_branch_refresh_v1, "
|
| 389 | "and/or interrupted_publication_after_request_v1. "
|
| 390 | "The interruption "
|
| 391 | "scenario is a loopback-only Oak fixture and never contacts a remote service."
|
| 392 | ),
|
| 393 | )
|
| 394 | parser.add_argument(
|
| 395 | "--git-remote",
|
| 396 | choices=("github", "local"),
|
| 397 | default="github",
|
| 398 | help=(
|
| 399 | "github: agent-default track, real push to GIT_BENCH_REMOTE "
|
| 400 | "(transport-comparable to oak). local: core-equivalent track, "
|
| 401 | "local bare remote (no network in git's loop)."
|
| 402 | ),
|
| 403 | )
|
| 404 | parser.add_argument("--workdir", type=Path, default=DEFAULT_WORKDIR)
|
| 405 | parser.add_argument("--results", type=Path, default=ROOT / "results" / "task-loop")
|
| 406 | parser.add_argument("--oak-repo", type=Path, default=DEFAULT_OAK_REPO)
|
| 407 | parser.add_argument("--admitted-output-chars", type=int, default=20_000)
|
| 408 | parser.add_argument("--keep-workdirs", action="store_true")
|
| 409 | parser.add_argument(
|
| 410 | "--randomize-subject-order",
|
| 411 | action="store_true",
|
| 412 | help="Shuffle subject order independently in each run using --random-seed.",
|
| 413 | )
|
| 414 | parser.add_argument("--random-seed", type=int, default=20260908)
|
| 415 | return parser.parse_args()
|
| 416 |
|
| 417 |
|
| 418 | def run_untimed(command: list[str], cwd: Path | None, timeout: float = 300.0) -> subprocess.CompletedProcess[str]:
|
| 419 | return subprocess.run(
|
| 420 | command,
|
| 421 | cwd=cwd,
|
| 422 | env=base_env(),
|
| 423 | text=True,
|
| 424 | stdout=subprocess.PIPE,
|
| 425 | stderr=subprocess.PIPE,
|
| 426 | timeout=timeout,
|
| 427 | check=False,
|
| 428 | )
|
| 429 |
|
| 430 |
|
| 431 | def step_row(
|
| 432 | subject: Subject,
|
| 433 | scenario: str,
|
| 434 | operation: str,
|
| 435 | command: list[str],
|
| 436 | cwd: Path,
|
| 437 | run_index: int,
|
| 438 | metadata: dict[str, Any],
|
| 439 | admitted_output_chars: int,
|
| 440 | *,
|
| 441 | kind: str = "vcs",
|
| 442 | expected_returncodes: tuple[int, ...] = (0,),
|
| 443 | preserve_stdout: bool = False,
|
| 444 | preserve_stderr: bool = False,
|
| 445 | env: dict[str, str] | None = None,
|
| 446 | timeout_seconds: float | None = None,
|
| 447 | ) -> dict[str, Any]:
|
| 448 | capture = run_timed(
|
| 449 | command,
|
| 450 | cwd,
|
| 451 | env or base_env(),
|
| 452 | admitted_output_chars,
|
| 453 | timeout_seconds=timeout_seconds,
|
| 454 | )
|
| 455 | expected = not capture.timed_out and capture.returncode in expected_returncodes
|
| 456 | is_vcs = 1 if kind == "vcs" else 0
|
| 457 | row = {
|
| 458 | **metadata,
|
| 459 | "subject": subject.name,
|
| 460 | "subject_kind": subject.kind,
|
| 461 | "subject_label": subject.label,
|
| 462 | "scenario": scenario,
|
| 463 | "run": run_index,
|
| 464 | "operation": operation,
|
| 465 | "step_kind": kind,
|
| 466 | "elapsed_ms": round(capture.elapsed_ms, 3),
|
| 467 | "returncode": 0 if expected else (capture.returncode or 1),
|
| 468 | "process_returncode": capture.returncode,
|
| 469 | "timed_out": capture.timed_out,
|
| 470 | "expected_returncodes": list(expected_returncodes),
|
| 471 | "command": command,
|
| 472 | "tool_call_count": 1,
|
| 473 | "terminal_tool_call_count": 1,
|
| 474 | "vcs_tool_call_count": is_vcs,
|
| 475 | **oakbench_tokens.interaction_token_fields(
|
| 476 | command_display(command),
|
| 477 | capture.stdout_text,
|
| 478 | capture.stderr_text,
|
| 479 | capture.stdout_bytes,
|
| 480 | capture.stderr_bytes,
|
| 481 | capture.stdout_truncated,
|
| 482 | capture.stderr_truncated,
|
| 483 | ),
|
| 484 | }
|
| 485 | if not expected:
|
| 486 | row["stderr"] = capture.stderr_text[-4000:]
|
| 487 | if preserve_stdout:
|
| 488 | row["stdout"] = capture.stdout_text
|
| 489 | if preserve_stderr:
|
| 490 | row["stderr"] = capture.stderr_text
|
| 491 | return row
|
| 492 |
|
| 493 |
|
| 494 | def skip_row(
|
| 495 | subject: Subject,
|
| 496 | scenario: str,
|
| 497 | run_index: int,
|
| 498 | metadata: dict[str, Any],
|
| 499 | reason: str,
|
| 500 | ) -> dict[str, Any]:
|
| 501 | return {
|
| 502 | **metadata,
|
| 503 | "subject": subject.name,
|
| 504 | "subject_kind": subject.kind,
|
| 505 | "subject_label": subject.label,
|
| 506 | "scenario": scenario,
|
| 507 | "run": run_index,
|
| 508 | "operation": "task_loop.skipped",
|
| 509 | "step_kind": "summary",
|
| 510 | "elapsed_ms": 0.0,
|
| 511 | "returncode": SKIP_RETURNCODE,
|
| 512 | "command": [],
|
| 513 | "skipped": True,
|
| 514 | "skip_reason": reason,
|
| 515 | "tool_call_count": 0,
|
| 516 | "terminal_tool_call_count": 0,
|
| 517 | "vcs_tool_call_count": 0,
|
| 518 | }
|
| 519 |
|
| 520 |
|
| 521 | def unmeasured_step_row(
|
| 522 | subject: Subject,
|
| 523 | scenario: str,
|
| 524 | operation: str,
|
| 525 | run_index: int,
|
| 526 | metadata: dict[str, Any],
|
| 527 | reason: str,
|
| 528 | ) -> dict[str, Any]:
|
| 529 | row = skip_row(subject, scenario, run_index, metadata, reason)
|
| 530 | row["operation"] = operation
|
| 531 | return row
|
| 532 |
|
| 533 |
|
| 534 | def harness_step_row(
|
| 535 | subject: Subject,
|
| 536 | scenario: str,
|
| 537 | operation: str,
|
| 538 | run_index: int,
|
| 539 | metadata: dict[str, Any],
|
| 540 | elapsed_ms: float,
|
| 541 | *,
|
| 542 | returncode: int = 0,
|
| 543 | harness_command_count: int = 0,
|
| 544 | ) -> dict[str, Any]:
|
| 545 | return {
|
| 546 | **metadata,
|
| 547 | "subject": subject.name,
|
| 548 | "subject_kind": subject.kind,
|
| 549 | "subject_label": subject.label,
|
| 550 | "scenario": scenario,
|
| 551 | "run": run_index,
|
| 552 | "operation": operation,
|
| 553 | "step_kind": "harness",
|
| 554 | "elapsed_ms": round(elapsed_ms, 3),
|
| 555 | "returncode": returncode,
|
| 556 | "command": [],
|
| 557 | "tool_call_count": 0,
|
| 558 | "terminal_tool_call_count": 0,
|
| 559 | "vcs_tool_call_count": 0,
|
| 560 | "harness_command_count": harness_command_count,
|
| 561 | "stdout_bytes": 0,
|
| 562 | "stderr_bytes": 0,
|
| 563 | "retry_count": 0,
|
| 564 | "human_action_count": 0,
|
| 565 | }
|
| 566 |
|
| 567 |
|
| 568 | def phase_total_row(
|
| 569 | rows: list[dict[str, Any]],
|
| 570 | subject: Subject,
|
| 571 | scenario: str,
|
| 572 | run_index: int,
|
| 573 | metadata: dict[str, Any],
|
| 574 | operation: str,
|
| 575 | ) -> dict[str, Any]:
|
| 576 | failures = [row for row in rows if row_returncode(row) not in (0, SKIP_RETURNCODE)]
|
| 577 | unmeasured = [
|
| 578 | str(row["operation"])
|
| 579 | for row in rows
|
| 580 | if row_returncode(row) == SKIP_RETURNCODE
|
| 581 | ]
|
| 582 | phase_elapsed_ms: dict[str, float] = {}
|
| 583 | for row in rows:
|
| 584 | phase = str(row.get("operation", "unknown")).split(".", 1)[0]
|
| 585 | phase_elapsed_ms[phase] = phase_elapsed_ms.get(phase, 0.0) + float(
|
| 586 | row.get("elapsed_ms", 0.0)
|
| 587 | )
|
| 588 | return {
|
| 589 | **metadata,
|
| 590 | "subject": subject.name,
|
| 591 | "subject_kind": subject.kind,
|
| 592 | "subject_label": subject.label,
|
| 593 | "scenario": scenario,
|
| 594 | "run": run_index,
|
| 595 | "operation": operation,
|
| 596 | "step_kind": "summary",
|
| 597 | "elapsed_ms": round(sum(float(row.get("elapsed_ms", 0)) for row in rows), 3),
|
| 598 | "returncode": 1 if failures else 0,
|
| 599 | "coverage_complete": not unmeasured,
|
| 600 | "unmeasured_operations": unmeasured,
|
| 601 | "command": [row["command"] for row in rows],
|
| 602 | "tool_call_count": sum(oakbench_tokens.int_or_zero(row.get("tool_call_count")) for row in rows),
|
| 603 | "terminal_tool_call_count": sum(
|
| 604 | oakbench_tokens.int_or_zero(row.get("terminal_tool_call_count")) for row in rows
|
| 605 | ),
|
| 606 | "vcs_tool_call_count": sum(oakbench_tokens.int_or_zero(row.get("vcs_tool_call_count")) for row in rows),
|
| 607 | "agent_tool_call_count": sum(
|
| 608 | oakbench_tokens.int_or_zero(row.get("tool_call_count")) for row in rows
|
| 609 | ),
|
| 610 | "harness_command_count": sum(
|
| 611 | oakbench_tokens.int_or_zero(row.get("harness_command_count")) for row in rows
|
| 612 | ),
|
| 613 | "retry_count": sum(oakbench_tokens.int_or_zero(row.get("retry_count")) for row in rows),
|
| 614 | "human_action_count": sum(
|
| 615 | oakbench_tokens.int_or_zero(row.get("human_action_count")) for row in rows
|
| 616 | ),
|
| 617 | "phase_elapsed_ms": {
|
| 618 | phase: round(elapsed, 3) for phase, elapsed in sorted(phase_elapsed_ms.items())
|
| 619 | },
|
| 620 | **oakbench_tokens.summed_token_fields(
|
| 621 | rows, "sum_of_task_loop_steps_command_plus_admitted_output_chars_div_4"
|
| 622 | ),
|
| 623 | "summarized_operations": [row["operation"] for row in rows],
|
| 624 | }
|
| 625 |
|
| 626 |
|
| 627 | def has_lane_failures(rows: list[dict[str, Any]]) -> bool:
|
| 628 | return any(row_returncode(row) not in (0, SKIP_RETURNCODE) for row in rows)
|
| 629 |
|
| 630 |
|
| 631 | def enforce_oracle_result(total: dict[str, Any], oracle_ok: bool | None) -> dict[str, Any]:
|
| 632 | if oracle_ok is False:
|
| 633 | total["returncode"] = 1
|
| 634 | total["stderr"] = "oracle_payload_missing_from_remote_branch"
|
| 635 | return total
|
| 636 |
|
| 637 |
|
| 638 | def interrupted_publication_fields(
|
| 639 | *,
|
| 640 | request_count: int,
|
| 641 | request_bytes: int | None,
|
| 642 | declared_request_bytes: int | None,
|
| 643 | request_framing_valid: bool,
|
| 644 | request_body_timed_out: bool,
|
| 645 | local_head_before: str | None,
|
| 646 | local_head_after: str | None,
|
| 647 | status_after: dict[str, Any] | None,
|
| 648 | worktree_payload_before: str | None,
|
| 649 | worktree_payload_after: str | None,
|
| 650 | cli_observation: dict[str, str],
|
| 651 | ) -> dict[str, Any]:
|
| 652 | """Evidence fields for a single interrupted mutation attempt.
|
| 653 |
|
| 654 | The fixture's receipt of a full request is ground truth for the harness,
|
| 655 | not a server durability receipt available to the CLI. A disconnect after
|
| 656 | that point is therefore unknown to the client and must never imply replay.
|
| 657 | """
|
| 658 | request_complete = bool(
|
| 659 | request_count == 1
|
| 660 | and request_framing_valid
|
| 661 | and declared_request_bytes is not None
|
| 662 | and request_bytes == declared_request_bytes
|
| 663 | )
|
| 664 | if request_count == 0:
|
| 665 | delivery_state = "no_request_observed"
|
| 666 | fixture_outcome = "not_sent_observed"
|
| 667 | elif not request_framing_valid:
|
| 668 | delivery_state = "invalid_request_framing"
|
| 669 | fixture_outcome = "indeterminate_after_invalid_request_disconnect"
|
| 670 | elif request_complete:
|
| 671 | delivery_state = "complete_request_observed"
|
| 672 | fixture_outcome = "unknown_after_complete_request_disconnect"
|
| 673 | else:
|
| 674 | delivery_state = "partial_request_observed"
|
| 675 | fixture_outcome = "indeterminate_after_partial_request_disconnect"
|
| 676 | payload_preserved = bool(
|
| 677 | worktree_payload_before
|
| 678 | and worktree_payload_before == worktree_payload_after
|
| 679 | )
|
| 680 | unpublished_state_preserved = bool(status_after and status_after.get("needs_push"))
|
| 681 | return {
|
| 682 | "publication_outcome": cli_observation["reported_outcome"],
|
| 683 | "cli_observation": cli_observation["observation"],
|
| 684 | "fixture_delivery_state": delivery_state,
|
| 685 | "fixture_outcome_knowledge": fixture_outcome,
|
| 686 | "fixture_received_complete_request_before_disconnect": request_complete,
|
| 687 | "fixture_request_framing_valid": request_framing_valid,
|
| 688 | "fixture_request_body_timed_out": request_body_timed_out,
|
| 689 | "fixture_declared_request_bytes": declared_request_bytes,
|
| 690 | "mutation_requests_observed": request_count,
|
| 691 | "blind_replay_requests": max(0, request_count - 1),
|
| 692 | "network_request_bytes_observed": request_bytes,
|
| 693 | "local_head_preserved": bool(local_head_before) and local_head_before == local_head_after,
|
| 694 | "worktree_payload_preserved": payload_preserved,
|
| 695 | "unpublished_state_preserved": unpublished_state_preserved,
|
| 696 | "local_work_preserved": payload_preserved and unpublished_state_preserved,
|
| 697 | "working_tree_dirty_after": None if status_after is None else status_after.get("dirty"),
|
| 698 | }
|
| 699 |
|
| 700 |
|
| 701 | def cli_publication_observation(row: dict[str, Any], *, json_requested: bool) -> dict[str, str]:
|
| 702 | """Classify only what the CLI reported, never what the fault fixture knows."""
|
| 703 | if row.get("timed_out"):
|
| 704 | return {"reported_outcome": "unreported", "observation": "process_timeout"}
|
| 705 | stdout = str(row.get("stdout") or "")
|
| 706 | stderr = str(row.get("stderr") or "")
|
| 707 | value = _parse_json_object(stdout) if json_requested else None
|
| 708 | if value is not None and row.get("process_returncode") == 0 and value.get("published") is True:
|
| 709 | return {"reported_outcome": "published", "observation": "json_success_receipt"}
|
| 710 | if value is not None and isinstance(value.get("error"), dict):
|
| 711 | message = str(value["error"].get("message") or "").lower()
|
| 712 | if any(marker in message for marker in ("unconfirmed", "unknown", "may have")):
|
| 713 | return {"reported_outcome": "unknown", "observation": "json_error_unknown"}
|
| 714 | return {
|
| 715 | "reported_outcome": "unreported",
|
| 716 | "observation": "json_error_without_publication_outcome",
|
| 717 | }
|
| 718 | message = f"{stdout}\n{stderr}".lower()
|
| 719 | if any(marker in message for marker in ("unconfirmed", "unknown", "may have")):
|
| 720 | return {"reported_outcome": "unknown", "observation": "text_error_unknown"}
|
| 721 | return {"reported_outcome": "unreported", "observation": "text_error_without_publication_outcome"}
|
| 722 |
|
| 723 |
|
| 724 | def file_sha256(path: Path) -> str | None:
|
| 725 | try:
|
| 726 | return hashlib.sha256(path.read_bytes()).hexdigest()
|
| 727 | except OSError:
|
| 728 | return None
|
| 729 |
|
| 730 |
|
| 731 | class _DisconnectAfterPushHandler(BaseHTTPRequestHandler):
|
| 732 | """Minimal loopback fixture for a complete request with a lost reply."""
|
| 733 |
|
| 734 | server_version = "oakbench-interruption/1"
|
| 735 |
|
| 736 | def log_message(self, _format: str, *_args: object) -> None:
|
| 737 | return
|
| 738 |
|
| 739 | def _json(self, status: int, payload: dict[str, Any]) -> None:
|
| 740 | encoded = json.dumps(payload).encode()
|
| 741 | self.send_response(status)
|
| 742 | self.send_header("Content-Type", "application/json")
|
| 743 | self.send_header("Content-Length", str(len(encoded)))
|
| 744 | self.end_headers()
|
| 745 | self.wfile.write(encoded)
|
| 746 |
|
| 747 | def do_GET(self) -> None: # noqa: N802 - BaseHTTPRequestHandler API
|
| 748 | self._json(404, {"error": "absent synthetic repo"})
|
| 749 |
|
| 750 | def do_POST(self) -> None: # noqa: N802 - BaseHTTPRequestHandler API
|
| 751 | declared_size: int | None = None
|
| 752 | framing_valid = False
|
| 753 | body = b""
|
| 754 | body_timed_out = False
|
| 755 | raw_sizes = self.headers.get_all("Content-Length", failobj=[])
|
| 756 | try:
|
| 757 | raw_size = raw_sizes[0].strip(" \t") if len(raw_sizes) == 1 else ""
|
| 758 | parsed_size = int(raw_size) if raw_size.isascii() and raw_size.isdecimal() else -1
|
| 759 | framing_valid = (
|
| 760 | 0 <= parsed_size <= INTERRUPTED_PUBLICATION_MAX_REQUEST_BYTES
|
| 761 | and self.headers.get("Transfer-Encoding") is None
|
| 762 | )
|
| 763 | if framing_valid:
|
| 764 | declared_size = parsed_size
|
| 765 | self.connection.settimeout(INTERRUPTED_PUBLICATION_REQUEST_TIMEOUT_S)
|
| 766 | while len(body) < parsed_size:
|
| 767 | try:
|
| 768 | chunk = self.rfile.read1(parsed_size - len(body))
|
| 769 | except (socket.timeout, TimeoutError):
|
| 770 | body_timed_out = True
|
| 771 | break
|
| 772 | except OSError:
|
| 773 | break
|
| 774 | if not chunk:
|
| 775 | break
|
| 776 | body += chunk
|
| 777 | except ValueError:
|
| 778 | framing_valid = False
|
| 779 | server = self.server
|
| 780 | if self.path == "/api/repos":
|
| 781 | self._json(200, {"name": "repo"})
|
| 782 | return
|
| 783 | if self.path == "/api/oak/repo/push":
|
| 784 | with server.receipt_lock: # type: ignore[attr-defined]
|
| 785 | server.push_request_count += 1 # type: ignore[attr-defined]
|
| 786 | server.push_request_bytes += len(body) # type: ignore[attr-defined]
|
| 787 | server.push_declared_request_bytes = declared_size # type: ignore[attr-defined]
|
| 788 | server.push_request_framing_valid = framing_valid # type: ignore[attr-defined]
|
| 789 | server.push_request_body_timed_out = body_timed_out # type: ignore[attr-defined]
|
| 790 | server.request_complete_before_disconnect = ( # type: ignore[attr-defined]
|
| 791 | framing_valid and len(body) == declared_size
|
| 792 | )
|
| 793 | self.close_connection = True
|
| 794 | try:
|
| 795 | self.connection.shutdown(socket.SHUT_RDWR)
|
| 796 | except OSError:
|
| 797 | pass
|
| 798 | self.connection.close()
|
| 799 | return
|
| 800 | self._json(404, {"error": "unexpected synthetic path"})
|
| 801 |
|
| 802 |
|
| 803 | def start_interrupted_publication_server() -> tuple[ThreadingHTTPServer, threading.Thread]:
|
| 804 | server = ThreadingHTTPServer(("127.0.0.1", 0), _DisconnectAfterPushHandler)
|
| 805 | server.receipt_lock = threading.Lock() # type: ignore[attr-defined]
|
| 806 | server.push_request_count = 0 # type: ignore[attr-defined]
|
| 807 | server.push_request_bytes = 0 # type: ignore[attr-defined]
|
| 808 | server.push_declared_request_bytes = None # type: ignore[attr-defined]
|
| 809 | server.push_request_framing_valid = False # type: ignore[attr-defined]
|
| 810 | server.push_request_body_timed_out = False # type: ignore[attr-defined]
|
| 811 | server.request_complete_before_disconnect = False # type: ignore[attr-defined]
|
| 812 | thread = threading.Thread(target=server.serve_forever, name="oakbench-interrupted-push", daemon=True)
|
| 813 | thread.start()
|
| 814 | return server, thread
|
| 815 |
|
| 816 |
|
| 817 | def _parse_json_object(text: str | None) -> dict[str, Any] | None:
|
| 818 | try:
|
| 819 | value = json.loads(text or "")
|
| 820 | except (TypeError, json.JSONDecodeError):
|
| 821 | return None
|
| 822 | return value if isinstance(value, dict) else None
|
| 823 |
|
| 824 |
|
| 825 | def run_interrupted_publication(
|
| 826 | subject: Subject,
|
| 827 | run_root: Path,
|
| 828 | run_index: int,
|
| 829 | metadata: dict[str, Any],
|
| 830 | admitted: int,
|
| 831 | ) -> list[dict[str, Any]]:
|
| 832 | """Run one real Oak push against a loopback server that drops its reply.
|
| 833 |
|
| 834 | The fixture observes request delivery but deliberately has no durable
|
| 835 | remote state. That makes the client's outcome unknown and lets the lane
|
| 836 | prove work preservation and absence of blind replay without production
|
| 837 | mutations or a fake success claim.
|
| 838 | """
|
| 839 | scenario = INTERRUPTED_PUBLICATION_SCENARIO
|
| 840 | metadata = {
|
| 841 | **metadata,
|
| 842 | "benchmark_track": "failure-evidence",
|
| 843 | "remote_transport": "loopback_fault_fixture",
|
| 844 | "remote_server": "127.0.0.1",
|
| 845 | }
|
| 846 | if subject.kind != "oak":
|
| 847 | return [
|
| 848 | skip_row(
|
| 849 | subject,
|
| 850 | scenario,
|
| 851 | run_index,
|
| 852 | metadata,
|
| 853 | "interrupted publication fixture currently exercises the actual Oak HTTP client only",
|
| 854 | )
|
| 855 | ]
|
| 856 |
|
| 857 | workspace = run_root / f"interrupted-{subject.name}-r{run_index}"
|
| 858 | workspace.mkdir(parents=True)
|
| 859 | rows: list[dict[str, Any]] = []
|
| 860 | vcs = str(subject.bin)
|
| 861 | rows.append(
|
| 862 | step_row(subject, scenario, "setup.init", [vcs, "init", "."], workspace, run_index, metadata, admitted)
|
| 863 | )
|
| 864 | if row_returncode(rows[-1]) != 0:
|
| 865 | return rows
|
| 866 | rows.append(
|
| 867 | edit_step(
|
| 868 | subject,
|
| 869 | scenario,
|
| 870 | "setup.edit",
|
| 871 | workspace,
|
| 872 | f"interrupted publication run={run_index}",
|
| 873 | run_index,
|
| 874 | metadata,
|
| 875 | admitted,
|
| 876 | )
|
| 877 | )
|
| 878 | rows.append(
|
| 879 | step_row(subject, scenario, "setup.checkpoint", [vcs, "commit"], workspace, run_index, metadata, admitted)
|
| 880 | )
|
| 881 | if row_returncode(rows[-1]) != 0:
|
| 882 | return rows
|
| 883 |
|
| 884 | before = run_untimed([vcs, "hash"], workspace)
|
| 885 | head_before = before.stdout.strip() if before.returncode == 0 else None
|
| 886 | payload_path = workspace / "task-notes.md"
|
| 887 | payload_before = file_sha256(payload_path)
|
| 888 | before_usage = bounded_tree_usage(workspace)
|
| 889 | push_help = run_untimed([vcs, "push", "--help"], workspace)
|
| 890 | push_supports_json = "--json" in (push_help.stdout + push_help.stderr)
|
| 891 | server, thread = start_interrupted_publication_server()
|
| 892 | host, port = server.server_address[:2]
|
| 893 | remote = f"http://{host}:{port}"
|
| 894 | fixture_env = base_env()
|
| 895 | fixture_env.pop("OAK_API_KEY", None)
|
| 896 | fixture_env.pop("OAK_REMOTE", None)
|
| 897 | fixture_env["OAK_NO_UPDATE_CHECK"] = "1"
|
| 898 | try:
|
| 899 | push_command = [vcs, "push", "--repo", "oak/repo", "--remote", remote]
|
| 900 | if push_supports_json:
|
| 901 | push_command.append("--json")
|
| 902 | publication_attempt = step_row(
|
| 903 | subject,
|
| 904 | scenario,
|
| 905 | "publication.attempt",
|
| 906 | push_command,
|
| 907 | workspace,
|
| 908 | run_index,
|
| 909 | metadata,
|
| 910 | admitted,
|
| 911 | expected_returncodes=tuple(range(1, 128)),
|
| 912 | preserve_stdout=True,
|
| 913 | preserve_stderr=True,
|
| 914 | env=fixture_env,
|
| 915 | timeout_seconds=INTERRUPTED_PUBLICATION_PROCESS_TIMEOUT_S,
|
| 916 | )
|
| 917 | rows.append(publication_attempt)
|
| 918 | finally:
|
| 919 | server.shutdown()
|
| 920 | server.server_close()
|
| 921 | thread.join(timeout=2.0)
|
| 922 |
|
| 923 | state_row = step_row(
|
| 924 | subject,
|
| 925 | scenario,
|
| 926 | "reconcile.local_state",
|
| 927 | [vcs, "agent", "state", "--json", "--compact"],
|
| 928 | workspace,
|
| 929 | run_index,
|
| 930 | metadata,
|
| 931 | admitted,
|
| 932 | preserve_stdout=True,
|
| 933 | env=fixture_env,
|
| 934 | )
|
| 935 | rows.append(state_row)
|
| 936 | after = run_untimed([vcs, "hash"], workspace)
|
| 937 | head_after = after.stdout.strip() if after.returncode == 0 else None
|
| 938 | payload_after = file_sha256(payload_path)
|
| 939 | status_after = _parse_json_object(state_row.get("stdout"))
|
| 940 | with server.receipt_lock: # type: ignore[attr-defined]
|
| 941 | request_count = int(server.push_request_count) # type: ignore[attr-defined]
|
| 942 | request_bytes = int(server.push_request_bytes) # type: ignore[attr-defined]
|
| 943 | declared_request_bytes = server.push_declared_request_bytes # type: ignore[attr-defined]
|
| 944 | request_framing_valid = bool(server.push_request_framing_valid) # type: ignore[attr-defined]
|
| 945 | request_body_timed_out = bool(server.push_request_body_timed_out) # type: ignore[attr-defined]
|
| 946 | cli_observation = cli_publication_observation(
|
| 947 | publication_attempt,
|
| 948 | json_requested=push_supports_json,
|
| 949 | )
|
| 950 | evidence = interrupted_publication_fields(
|
| 951 | request_count=request_count,
|
| 952 | request_bytes=request_bytes if request_count else None,
|
| 953 | declared_request_bytes=declared_request_bytes,
|
| 954 | request_framing_valid=request_framing_valid,
|
| 955 | request_body_timed_out=request_body_timed_out,
|
| 956 | local_head_before=head_before,
|
| 957 | local_head_after=head_after,
|
| 958 | status_after=status_after,
|
| 959 | worktree_payload_before=payload_before,
|
| 960 | worktree_payload_after=payload_after,
|
| 961 | cli_observation=cli_observation,
|
| 962 | )
|
| 963 | evidence["workspace_usage_before"] = before_usage
|
| 964 | evidence["workspace_usage_after"] = bounded_tree_usage(workspace)
|
| 965 | evidence["remote_transport"] = "loopback_fault_fixture"
|
| 966 | evidence["remote_server"] = str(host)
|
| 967 | evidence["human_action_count"] = 0
|
| 968 | evidence["compatibility_probe_calls"] = 1
|
| 969 | evidence["harness_probe_calls"] = 3
|
| 970 | evidence["push_json_supported"] = push_supports_json
|
| 971 | total = phase_total_row(rows, subject, scenario, run_index, metadata, "interrupted_publication.total")
|
| 972 | evidence["harness_command_count"] = total["harness_command_count"] + 3
|
| 973 | total.update(evidence)
|
| 974 | if not (
|
| 975 | evidence["fixture_received_complete_request_before_disconnect"]
|
| 976 | and request_count == 1
|
| 977 | and evidence["local_head_preserved"]
|
| 978 | and evidence["local_work_preserved"]
|
| 979 | and not publication_attempt["timed_out"]
|
| 980 | and row_returncode(rows[-1]) == 0
|
| 981 | ):
|
| 982 | total["returncode"] = 1
|
| 983 | total["stderr"] = "interrupted_publication_evidence_incomplete"
|
| 984 | rows.append(total)
|
| 985 | return rows
|
| 986 |
|
| 987 |
|
| 988 | def edit_step(
|
| 989 | subject: Subject,
|
| 990 | scenario: str,
|
| 991 | operation: str,
|
| 992 | workspace: Path,
|
| 993 | payload: str,
|
| 994 | run_index: int,
|
| 995 | metadata: dict[str, Any],
|
| 996 | admitted: int,
|
| 997 | ) -> dict[str, Any]:
|
| 998 | command = ["/bin/zsh", "-c", f"echo {payload!r} >> task-notes.md"]
|
| 999 | return step_row(
|
| 1000 | subject, scenario, operation, command, workspace, run_index, metadata, admitted, kind="edit"
|
| 1001 | )
|
| 1002 |
|
| 1003 |
|
| 1004 | def oak_virtual_branch(subject: Subject, mount_dir: Path) -> str | None:
|
| 1005 | proc = run_untimed([str(subject.bin), "status"], mount_dir)
|
| 1006 | match = VIRTUAL_BRANCH_RE.search(proc.stdout + proc.stderr)
|
| 1007 | return match.group(1) if match else None
|
| 1008 |
|
| 1009 |
|
| 1010 | def run_oak_loop(
|
| 1011 | subject: Subject,
|
| 1012 | remote: oakbench_remotes.RemoteResolution,
|
| 1013 | run_root: Path,
|
| 1014 | run_index: int,
|
| 1015 | metadata: dict[str, Any],
|
| 1016 | admitted: int,
|
| 1017 | ) -> list[dict[str, Any]]:
|
| 1018 | scenario = "task_loop"
|
| 1019 | meta = {**metadata, **remote.row_fields()}
|
| 1020 | rows: list[dict[str, Any]] = []
|
| 1021 | branches: dict[str, str | None] = {}
|
| 1022 | payloads: dict[str, str] = {}
|
| 1023 | mounts: list[Path] = []
|
| 1024 | vcs = str(subject.bin)
|
| 1025 | try:
|
| 1026 | for phase in PHASES:
|
| 1027 | workspace = run_root / f"tl-{subject.name}-r{run_index}-{phase}"
|
| 1028 | payload = f"payload {phase} run={run_index} bench={metadata['bench_id']}"
|
| 1029 | payloads[phase] = payload
|
| 1030 | acquire = step_row(
|
| 1031 | subject, scenario, f"{phase}.acquire", [vcs, "mount", remote.repo or "", str(workspace)],
|
| 1032 | run_root, run_index, meta, admitted,
|
| 1033 | )
|
| 1034 | rows.append(acquire)
|
| 1035 | if int(acquire["returncode"]) != 0:
|
| 1036 | return rows
|
| 1037 | mounts.append(workspace)
|
| 1038 | branches[phase] = oak_virtual_branch(subject, workspace)
|
| 1039 | rows.append(edit_step(subject, scenario, f"{phase}.edit", workspace, payload, run_index, meta, admitted))
|
| 1040 | rows.append(step_row(subject, scenario, f"{phase}.snapshot", [vcs, "commit", "--no-verify"], workspace, run_index, meta, admitted))
|
| 1041 | rows.append(step_row(subject, scenario, f"{phase}.publish", [vcs, "push"], workspace, run_index, meta, admitted))
|
| 1042 | if phase == "task":
|
| 1043 | rows.append(
|
| 1044 | step_row(
|
| 1045 | subject, scenario, "task.describe",
|
| 1046 | [vcs, "desc", f"bench task loop {metadata['bench_id']} r{run_index}"],
|
| 1047 | workspace, run_index, meta, admitted,
|
| 1048 | )
|
| 1049 | )
|
| 1050 | end = step_row(
|
| 1051 | subject, scenario, f"{phase}.end", [vcs, "mount", "end", str(workspace)],
|
| 1052 | run_root, run_index, meta, admitted,
|
| 1053 | )
|
| 1054 | rows.append(end)
|
| 1055 | if int(end["returncode"]) == 0 and workspace in mounts:
|
| 1056 | mounts.remove(workspace)
|
| 1057 |
|
| 1058 | # Untimed oracle: each phase's payload must be on its pushed branch.
|
| 1059 | verifier = run_root / f"tl-verify-{subject.name}-r{run_index}"
|
| 1060 | clone = run_untimed([vcs, "clone", remote.repo or "", str(verifier)], run_root)
|
| 1061 | oracle_ok: bool | None = None
|
| 1062 | if clone.returncode == 0:
|
| 1063 | oracle_ok = True
|
| 1064 | for phase in PHASES:
|
| 1065 | branch = branches.get(phase)
|
| 1066 | if not branch:
|
| 1067 | oracle_ok = None
|
| 1068 | break
|
| 1069 | switch = run_untimed([vcs, "switch", branch], verifier)
|
| 1070 | content = (verifier / "task-notes.md")
|
| 1071 | seen = content.read_text() if content.exists() else ""
|
| 1072 | if switch.returncode != 0 or payloads[phase] not in seen:
|
| 1073 | oracle_ok = False
|
| 1074 | total = {
|
| 1075 | **phase_total_row(rows, subject, scenario, run_index, meta, "task_loop.total"),
|
| 1076 | "oracle_payload_on_remote_branch": oracle_ok,
|
| 1077 | "workspace_branches": branches,
|
| 1078 | "measurement_source": "verifier_clone_switch_grep; null oracle means branch name unparsed",
|
| 1079 | }
|
| 1080 | rows.append(enforce_oracle_result(total, oracle_ok))
|
| 1081 | return rows
|
| 1082 | finally:
|
| 1083 | for mount_dir in mounts:
|
| 1084 | run_untimed([vcs, "mount", "end", str(mount_dir), "-f"], run_root)
|
| 1085 |
|
| 1086 |
|
| 1087 | def ensure_git_remote_seeded(subject: Subject, remote: oakbench_remotes.RemoteResolution, scratch: Path) -> str | None:
|
| 1088 | """Untimed: make sure the git remote has a main with a seed commit."""
|
| 1089 | vcs = str(subject.bin)
|
| 1090 | probe = run_untimed([vcs, "ls-remote", str(remote.repo), "refs/heads/main"], None)
|
| 1091 | if probe.returncode == 0 and probe.stdout.strip():
|
| 1092 | if remote.transport == oakbench_remotes.TRANSPORT_LOCAL_FILE:
|
| 1093 | head = run_untimed([vcs, "symbolic-ref", "HEAD", "refs/heads/main"], Path(str(remote.repo)))
|
| 1094 | if head.returncode != 0:
|
| 1095 | return f"git remote HEAD setup failed: {head.stderr.strip()[:200]}"
|
| 1096 | return None
|
| 1097 | seed = scratch / "git-seed"
|
| 1098 | if seed.exists():
|
| 1099 | shutil.rmtree(seed)
|
| 1100 | seed.mkdir(parents=True)
|
| 1101 | (seed / "README.md").write_text("Task-loop benchmark repo (disposable).\n")
|
| 1102 | (seed / "task-notes.md").write_text("# task notes\n")
|
| 1103 | for command in (
|
| 1104 | [vcs, "init", "-q", "-b", "main"],
|
| 1105 | [vcs, "add", "."],
|
| 1106 | [vcs, "commit", "-q", "-m", "seed"],
|
| 1107 | [vcs, "push", "-q", str(remote.repo), "main:main"],
|
| 1108 | ):
|
| 1109 | proc = run_untimed(command, seed)
|
| 1110 | if proc.returncode != 0:
|
| 1111 | return f"git remote seed `{' '.join(command[1:3])}` failed: {proc.stderr.strip()[:200]}"
|
| 1112 | if remote.transport == oakbench_remotes.TRANSPORT_LOCAL_FILE:
|
| 1113 | head = run_untimed([vcs, "symbolic-ref", "HEAD", "refs/heads/main"], Path(str(remote.repo)))
|
| 1114 | if head.returncode != 0:
|
| 1115 | return f"git remote HEAD setup failed: {head.stderr.strip()[:200]}"
|
| 1116 | return None
|
| 1117 |
|
| 1118 |
|
| 1119 | def run_git_agent_task_baseline(
|
| 1120 | subject: Subject,
|
| 1121 | remote: oakbench_remotes.RemoteResolution,
|
| 1122 | run_root: Path,
|
| 1123 | run_index: int,
|
| 1124 | metadata: dict[str, Any],
|
| 1125 | admitted: int,
|
| 1126 | remote_create_ms: float = 0.0,
|
| 1127 | ) -> list[dict[str, Any]]:
|
| 1128 | """Measure a cold clone and a warm isolated worktree on one local remote.
|
| 1129 |
|
| 1130 | This is the core-equivalent, local-file cell. Hosted CI is deliberately a
|
| 1131 | skip row: a local commit oracle is not relabelled as provider CI.
|
| 1132 | """
|
| 1133 | scenario = AGENT_TASK_SCENARIO
|
| 1134 | meta = {
|
| 1135 | **metadata,
|
| 1136 | **remote.row_fields(),
|
| 1137 | "benchmark_track": (
|
| 1138 | "core-equivalent"
|
| 1139 | if remote.transport == oakbench_remotes.TRANSPORT_LOCAL_FILE
|
| 1140 | else "agent-default"
|
| 1141 | ),
|
| 1142 | }
|
| 1143 | vcs = str(subject.bin)
|
| 1144 | rows: list[dict[str, Any]] = [
|
| 1145 | harness_step_row(
|
| 1146 | subject,
|
| 1147 | scenario,
|
| 1148 | "setup.remote_container",
|
| 1149 | run_index,
|
| 1150 | meta,
|
| 1151 | remote_create_ms,
|
| 1152 | harness_command_count=1,
|
| 1153 | )
|
| 1154 | ]
|
| 1155 | setup_start = time.perf_counter()
|
| 1156 | seed_error = ensure_git_remote_seeded(subject, remote, run_root)
|
| 1157 | rows.append(
|
| 1158 | harness_step_row(
|
| 1159 | subject,
|
| 1160 | scenario,
|
| 1161 | "setup.remote_seed",
|
| 1162 | run_index,
|
| 1163 | meta,
|
| 1164 | (time.perf_counter() - setup_start) * 1000,
|
| 1165 | returncode=1 if seed_error else 0,
|
| 1166 | harness_command_count=6,
|
| 1167 | )
|
| 1168 | )
|
| 1169 | if seed_error:
|
| 1170 | rows[-1]["stderr"] = seed_error
|
| 1171 | return rows
|
| 1172 |
|
| 1173 | base = run_root / f"agent-task-{subject.name}-r{run_index}-cold"
|
| 1174 | cold_branch = oakbench_remotes.disposable_branch(
|
| 1175 | str(metadata["bench_id"]), "agent-task-cold", subject.name, run_index
|
| 1176 | )
|
| 1177 | warm_branch = oakbench_remotes.disposable_branch(
|
| 1178 | str(metadata["bench_id"]), "agent-task-warm", subject.name, run_index
|
| 1179 | )
|
| 1180 | cold_payload = f"cold task run={run_index} bench={metadata['bench_id']}"
|
| 1181 | warm_payload = f"warm task run={run_index} bench={metadata['bench_id']}"
|
| 1182 |
|
| 1183 | rows.append(
|
| 1184 | step_row(
|
| 1185 | subject,
|
| 1186 | scenario,
|
| 1187 | "cold.acquire",
|
| 1188 | [vcs, "clone", "-q", str(remote.repo), str(base)],
|
| 1189 | run_root,
|
| 1190 | run_index,
|
| 1191 | meta,
|
| 1192 | admitted,
|
| 1193 | )
|
| 1194 | )
|
| 1195 | if row_returncode(rows[-1]) != 0:
|
| 1196 | return rows
|
| 1197 | rows.append(
|
| 1198 | step_row(
|
| 1199 | subject, scenario, "cold.branch", [vcs, "switch", "-q", "-c", cold_branch],
|
| 1200 | base, run_index, meta, admitted,
|
| 1201 | )
|
| 1202 | )
|
| 1203 | rows.append(
|
| 1204 | step_row(
|
| 1205 | subject, scenario, "cold.inspect", [vcs, "status", "--porcelain=v1"],
|
| 1206 | base, run_index, meta, admitted,
|
| 1207 | )
|
| 1208 | )
|
| 1209 | rows.append(edit_step(subject, scenario, "cold.edit", base, cold_payload, run_index, meta, admitted))
|
| 1210 | rows.append(step_row(subject, scenario, "cold.checkpoint.add", [vcs, "add", "."], base, run_index, meta, admitted))
|
| 1211 | rows.append(
|
| 1212 | step_row(
|
| 1213 | subject, scenario, "cold.checkpoint.commit",
|
| 1214 | [vcs, "commit", "-q", "-m", f"agent task cold {metadata['bench_id']} r{run_index}"],
|
| 1215 | base, run_index, meta, admitted,
|
| 1216 | )
|
| 1217 | )
|
| 1218 | rows.append(
|
| 1219 | step_row(
|
| 1220 | subject, scenario, "cold.publish", [vcs, "push", "-q", "-u", "origin", cold_branch],
|
| 1221 | base, run_index, meta, admitted,
|
| 1222 | )
|
| 1223 | )
|
| 1224 | rows.append(
|
| 1225 | unmeasured_step_row(
|
| 1226 | subject, scenario, "cold.ci.observe", run_index, meta,
|
| 1227 | "local-file transport has no hosted CI provider; exact CI observation is unmeasured",
|
| 1228 | )
|
| 1229 | )
|
| 1230 | rows.append(
|
| 1231 | step_row(
|
| 1232 | subject, scenario, "cold.review_handoff",
|
| 1233 | [vcs, "show", "--format=%H%n%s", "--stat", "--oneline", "HEAD"],
|
| 1234 | base, run_index, meta, admitted,
|
| 1235 | )
|
| 1236 | )
|
| 1237 |
|
| 1238 | warm = run_root / f"agent-task-{subject.name}-r{run_index}-warm"
|
| 1239 | rows.append(
|
| 1240 | step_row(
|
| 1241 | subject, scenario, "warm.acquire",
|
| 1242 | [vcs, "worktree", "add", "-q", str(warm), "-b", warm_branch, "origin/main"],
|
| 1243 | base, run_index, meta, admitted,
|
| 1244 | )
|
| 1245 | )
|
| 1246 | if row_returncode(rows[-1]) == 0:
|
| 1247 | rows.append(
|
| 1248 | step_row(
|
| 1249 | subject, scenario, "warm.inspect", [vcs, "status", "--porcelain=v1"],
|
| 1250 | warm, run_index, meta, admitted,
|
| 1251 | )
|
| 1252 | )
|
| 1253 | rows.append(edit_step(subject, scenario, "warm.edit", warm, warm_payload, run_index, meta, admitted))
|
| 1254 | rows.append(step_row(subject, scenario, "warm.checkpoint.add", [vcs, "add", "."], warm, run_index, meta, admitted))
|
| 1255 | rows.append(
|
| 1256 | step_row(
|
| 1257 | subject, scenario, "warm.checkpoint.commit",
|
| 1258 | [vcs, "commit", "-q", "-m", f"agent task warm {metadata['bench_id']} r{run_index}"],
|
| 1259 | warm, run_index, meta, admitted,
|
| 1260 | )
|
| 1261 | )
|
| 1262 | rows.append(
|
| 1263 | step_row(
|
| 1264 | subject, scenario, "warm.publish", [vcs, "push", "-q", "-u", "origin", warm_branch],
|
| 1265 | warm, run_index, meta, admitted,
|
| 1266 | )
|
| 1267 | )
|
| 1268 | rows.append(
|
| 1269 | unmeasured_step_row(
|
| 1270 | subject, scenario, "warm.ci.observe", run_index, meta,
|
| 1271 | "local-file transport has no hosted CI provider; exact CI observation is unmeasured",
|
| 1272 | )
|
| 1273 | )
|
| 1274 | rows.append(
|
| 1275 | step_row(
|
| 1276 | subject, scenario, "warm.review_handoff",
|
| 1277 | [vcs, "show", "--format=%H%n%s", "--stat", "--oneline", "HEAD"],
|
| 1278 | warm, run_index, meta, admitted,
|
| 1279 | )
|
| 1280 | )
|
| 1281 |
|
| 1282 | usage_before_cleanup = {
|
| 1283 | "cold": bounded_tree_usage(base),
|
| 1284 | "warm": bounded_tree_usage(warm),
|
| 1285 | }
|
| 1286 | oracle_ok = True
|
| 1287 | for branch, payload in ((cold_branch, cold_payload), (warm_branch, warm_payload)):
|
| 1288 | show = run_untimed([vcs, "--git-dir", str(remote.repo), "show", f"{branch}:task-notes.md"], run_root)
|
| 1289 | if show.returncode != 0 or payload not in show.stdout:
|
| 1290 | oracle_ok = False
|
| 1291 |
|
| 1292 | if warm.exists():
|
| 1293 | rows.append(
|
| 1294 | step_row(
|
| 1295 | subject, scenario, "cleanup.warm",
|
| 1296 | [vcs, "worktree", "remove", str(warm)], base, run_index, meta, admitted,
|
| 1297 | )
|
| 1298 | )
|
| 1299 | cleanup_start = time.perf_counter()
|
| 1300 | shutil.rmtree(base, ignore_errors=True)
|
| 1301 | rows.append(
|
| 1302 | harness_step_row(
|
| 1303 | subject, scenario, "cleanup.cold", run_index, meta,
|
| 1304 | (time.perf_counter() - cleanup_start) * 1000,
|
| 1305 | )
|
| 1306 | )
|
| 1307 |
|
| 1308 | total = phase_total_row(rows, subject, scenario, run_index, meta, "agent_task.total")
|
| 1309 | total.update(
|
| 1310 | {
|
| 1311 | "oracle_payload_on_remote_branch": oracle_ok,
|
| 1312 | "acquisition_models": {
|
| 1313 | "cold": "fresh_full_clone",
|
| 1314 | "warm": "shared_object_worktree",
|
| 1315 | },
|
| 1316 | "repository_cache_state": {
|
| 1317 | "cold": "fresh_destination_no_shared_object_store",
|
| 1318 | "warm": "explicit_shared_object_store",
|
| 1319 | },
|
| 1320 | "os_cache_state": "unknown",
|
| 1321 | "os_cache_state_reason": "page_cache_not_purged_or_instrumented",
|
| 1322 | "workspace_usage_before_cleanup": usage_before_cleanup,
|
| 1323 | "network_bytes": None,
|
| 1324 | "network_bytes_reason": "local_file_transport_not_observed_by_byte_proxy",
|
| 1325 | "partial_clone_comparator": None,
|
| 1326 | "partial_clone_reason": "local file transport does not establish a network partial-clone comparison",
|
| 1327 | "human_action_count": 0,
|
| 1328 | "parallel_evidence_lane": "parallel_contention.py/workspace_per_task",
|
| 1329 | "oracle_probe_calls": 2,
|
| 1330 | "harness_command_count": total["harness_command_count"] + 2,
|
| 1331 | }
|
| 1332 | )
|
| 1333 | if not oracle_ok:
|
| 1334 | total["returncode"] = 1
|
| 1335 | total["stderr"] = "agent_task_remote_payload_oracle_failed"
|
| 1336 | rows.append(total)
|
| 1337 | return rows
|
| 1338 |
|
| 1339 |
|
| 1340 | def reserve_loopback_port() -> int:
|
| 1341 | with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as listener:
|
| 1342 | listener.bind(("127.0.0.1", 0))
|
| 1343 | return int(listener.getsockname()[1])
|
| 1344 |
|
| 1345 |
|
| 1346 | def wait_for_loopback_port(
|
| 1347 | process: subprocess.Popen[Any], port: int, timeout_s: float = LOCAL_OAK_SERVER_READY_TIMEOUT_S
|
| 1348 | ) -> bool:
|
| 1349 | deadline = time.monotonic() + timeout_s
|
| 1350 | while time.monotonic() < deadline:
|
| 1351 | if process.poll() is not None:
|
| 1352 | return False
|
| 1353 | try:
|
| 1354 | with socket.create_connection(("127.0.0.1", port), timeout=0.2):
|
| 1355 | return True
|
| 1356 | except OSError:
|
| 1357 | time.sleep(0.05)
|
| 1358 | return False
|
| 1359 |
|
| 1360 |
|
| 1361 | def harness_command_row(
|
| 1362 | subject: Subject,
|
| 1363 | scenario: str,
|
| 1364 | operation: str,
|
| 1365 | command: list[str],
|
| 1366 | cwd: Path,
|
| 1367 | run_index: int,
|
| 1368 | metadata: dict[str, Any],
|
| 1369 | ) -> tuple[dict[str, Any], subprocess.CompletedProcess[str]]:
|
| 1370 | started = time.perf_counter()
|
| 1371 | result = run_bounded_harness_command(command, cwd)
|
| 1372 | row = harness_step_row(
|
| 1373 | subject,
|
| 1374 | scenario,
|
| 1375 | operation,
|
| 1376 | run_index,
|
| 1377 | metadata,
|
| 1378 | (time.perf_counter() - started) * 1000,
|
| 1379 | returncode=0 if result.returncode == 0 else result.returncode,
|
| 1380 | harness_command_count=1,
|
| 1381 | )
|
| 1382 | if result.returncode != 0:
|
| 1383 | row["stderr"] = str(result.stderr)[-4000:]
|
| 1384 | return row, result
|
| 1385 |
|
| 1386 |
|
| 1387 | def run_bounded_harness_command(
|
| 1388 | command: list[str], cwd: Path, *, env: dict[str, str] | None = None
|
| 1389 | ) -> subprocess.CompletedProcess[str]:
|
| 1390 | capture = run_timed(
|
| 1391 | command,
|
| 1392 | cwd,
|
| 1393 | env or base_env(),
|
| 1394 | admitted_output_chars=4_000,
|
| 1395 | timeout_seconds=INTERRUPTED_PUBLICATION_PROCESS_TIMEOUT_S,
|
| 1396 | )
|
| 1397 | return subprocess.CompletedProcess(
|
| 1398 | command,
|
| 1399 | capture.returncode,
|
| 1400 | capture.stdout_text,
|
| 1401 | capture.stderr_text,
|
| 1402 | )
|
| 1403 |
|
| 1404 |
|
| 1405 | def remote_branch_head(remote: str, repo_spec: str, branch: str) -> str | None:
|
| 1406 | owner, repo = repo_spec.split("/", 1)
|
| 1407 | branch_segment = urllib.parse.quote(branch, safe="")
|
| 1408 | url = f"{remote}/api/{owner}/{repo}/branches/{branch_segment}"
|
| 1409 | try:
|
| 1410 | with urllib.request.urlopen(url, timeout=2.0) as response:
|
| 1411 | value = json.loads(response.read())
|
| 1412 | except (OSError, ValueError, urllib.error.URLError):
|
| 1413 | return None
|
| 1414 | head = value.get("head") if isinstance(value, dict) else None
|
| 1415 | return head if isinstance(head, str) else None
|
| 1416 |
|
| 1417 |
|
| 1418 | def run_oak_local_agent_task_baseline(
|
| 1419 | subject: Subject,
|
| 1420 | run_root: Path,
|
| 1421 | run_index: int,
|
| 1422 | metadata: dict[str, Any],
|
| 1423 | admitted: int,
|
| 1424 | ) -> list[dict[str, Any]]:
|
| 1425 | """Exercise two fresh Oak checkouts against an owned loopback Oak server."""
|
| 1426 | run_root = run_root.resolve()
|
| 1427 | scenario = AGENT_TASK_OAK_LOCAL_SCENARIO
|
| 1428 | meta = {
|
| 1429 | **metadata,
|
| 1430 | "benchmark_track": "oak-local-loopback",
|
| 1431 | "remote_transport": "loopback_http",
|
| 1432 | "remote_server": "127.0.0.1",
|
| 1433 | }
|
| 1434 | rows: list[dict[str, Any]] = []
|
| 1435 | vcs = str(subject.bin)
|
| 1436 | server_root = run_root / f"oak-local-server-{subject.name}-r{run_index}"
|
| 1437 | seed = run_root / f"oak-local-seed-{subject.name}-r{run_index}"
|
| 1438 | cold = run_root / f"oak-local-cold-{subject.name}-r{run_index}"
|
| 1439 | followup = run_root / f"oak-local-followup-{subject.name}-r{run_index}"
|
| 1440 | verifier = run_root / f"oak-local-verifier-{subject.name}-r{run_index}"
|
| 1441 | repo_spec = f"qa/task-loop-{subject.name}-r{run_index}"
|
| 1442 | port = reserve_loopback_port()
|
| 1443 | remote = f"http://127.0.0.1:{port}"
|
| 1444 | server_root.mkdir(parents=True)
|
| 1445 | server_started = time.perf_counter()
|
| 1446 | server: subprocess.Popen[Any] | None = None
|
| 1447 | server_start_error: str | None = None
|
| 1448 | try:
|
| 1449 | server = subprocess.Popen(
|
| 1450 | [vcs, "serve", "--dir", str(server_root), "--host", "127.0.0.1", "--port", str(port)],
|
| 1451 | cwd=run_root,
|
| 1452 | env=base_env(),
|
| 1453 | stdout=subprocess.DEVNULL,
|
| 1454 | stderr=subprocess.DEVNULL,
|
| 1455 | start_new_session=os.name == "posix",
|
| 1456 | )
|
| 1457 | except OSError as error:
|
| 1458 | server_start_error = type(error).__name__
|
| 1459 | ready = server is not None and wait_for_loopback_port(server, port)
|
| 1460 | rows.append(
|
| 1461 | harness_step_row(
|
| 1462 | subject,
|
| 1463 | scenario,
|
| 1464 | "setup.server",
|
| 1465 | run_index,
|
| 1466 | meta,
|
| 1467 | (time.perf_counter() - server_started) * 1000,
|
| 1468 | returncode=0 if ready else 1,
|
| 1469 | harness_command_count=1,
|
| 1470 | )
|
| 1471 | )
|
| 1472 | branch: str | None = None
|
| 1473 | expected_head: str | None = None
|
| 1474 | expected_payload = "# task notes\n"
|
| 1475 | failure: str | None = None if ready else (
|
| 1476 | f"local_oak_server_start_failed:{server_start_error}"
|
| 1477 | if server_start_error else "local_oak_server_not_ready"
|
| 1478 | )
|
| 1479 | try:
|
| 1480 | if failure is None:
|
| 1481 | seed.mkdir(parents=True)
|
| 1482 | (seed / "task-notes.md").write_text(expected_payload)
|
| 1483 | for operation, command in (
|
| 1484 | ("setup.seed_init", [vcs, "init", "."]),
|
| 1485 | ("setup.seed_commit", [vcs, "commit"]),
|
| 1486 | ("setup.seed_push", [vcs, "push", "--repo", repo_spec, "--remote", remote]),
|
| 1487 | ):
|
| 1488 | row, result = harness_command_row(
|
| 1489 | subject, scenario, operation, command, seed, run_index, meta
|
| 1490 | )
|
| 1491 | rows.append(row)
|
| 1492 | if result.returncode != 0:
|
| 1493 | failure = f"{operation}_failed"
|
| 1494 | break
|
| 1495 |
|
| 1496 | def agent_step(operation: str, command: list[str], cwd: Path, *, preserve: bool = False) -> dict[str, Any]:
|
| 1497 | row = step_row(
|
| 1498 | subject,
|
| 1499 | scenario,
|
| 1500 | operation,
|
| 1501 | command,
|
| 1502 | cwd,
|
| 1503 | run_index,
|
| 1504 | meta,
|
| 1505 | admitted,
|
| 1506 | preserve_stdout=preserve,
|
| 1507 | timeout_seconds=INTERRUPTED_PUBLICATION_PROCESS_TIMEOUT_S,
|
| 1508 | )
|
| 1509 | rows.append(row)
|
| 1510 | return row
|
| 1511 |
|
| 1512 | if failure is None:
|
| 1513 | row = agent_step(
|
| 1514 | "cold.acquire",
|
| 1515 | [vcs, "clone", repo_spec, str(cold), "--remote", remote, "--allow-legacy-scope"],
|
| 1516 | run_root,
|
| 1517 | )
|
| 1518 | if row_returncode(row) != 0:
|
| 1519 | failure = "cold.acquire_failed"
|
| 1520 | if failure is None:
|
| 1521 | status = agent_step("cold.inspect", [vcs, "status", "--json"], cold, preserve=True)
|
| 1522 | status_value = _parse_json_object(status.get("stdout"))
|
| 1523 | branch_value = status_value.get("branch") if status_value else None
|
| 1524 | branch = branch_value if isinstance(branch_value, str) else None
|
| 1525 | if not branch:
|
| 1526 | failure = "cold.branch_identity_missing"
|
| 1527 | if failure is None:
|
| 1528 | cold_payload = f"cold task run={run_index} bench={metadata['bench_id']}\n"
|
| 1529 | expected_payload += cold_payload
|
| 1530 | rows.append(edit_step(subject, scenario, "cold.edit", cold, cold_payload.strip(), run_index, meta, admitted))
|
| 1531 | if row_returncode(agent_step("cold.checkpoint", [vcs, "commit"], cold)) != 0:
|
| 1532 | failure = "cold.checkpoint_failed"
|
| 1533 | elif row_returncode(
|
| 1534 | agent_step("cold.publish", [vcs, "push", "--repo", repo_spec, "--remote", remote], cold)
|
| 1535 | ) != 0:
|
| 1536 | failure = "cold.publish_failed"
|
| 1537 | rows.append(
|
| 1538 | unmeasured_step_row(
|
| 1539 | subject, scenario, "cold.ci.observe", run_index, meta,
|
| 1540 | "owned local oak serve has no configured CI provider",
|
| 1541 | )
|
| 1542 | )
|
| 1543 | rows.append(
|
| 1544 | unmeasured_step_row(
|
| 1545 | subject, scenario, "cold.review_handoff", run_index, meta,
|
| 1546 | "owned local oak serve does not provide hosted review semantics",
|
| 1547 | )
|
| 1548 | )
|
| 1549 | if failure is None and branch is not None:
|
| 1550 | row = agent_step(
|
| 1551 | "followup.acquire",
|
| 1552 | [
|
| 1553 | vcs, "clone", repo_spec, str(followup), "--remote", remote,
|
| 1554 | "--branch", branch, "--allow-legacy-scope",
|
| 1555 | ],
|
| 1556 | run_root,
|
| 1557 | )
|
| 1558 | if row_returncode(row) != 0:
|
| 1559 | failure = "followup.acquire_failed"
|
| 1560 | if failure is None:
|
| 1561 | agent_step("followup.inspect", [vcs, "status", "--json"], followup)
|
| 1562 | followup_payload = f"followup task run={run_index} bench={metadata['bench_id']}\n"
|
| 1563 | expected_payload += followup_payload
|
| 1564 | rows.append(
|
| 1565 | edit_step(
|
| 1566 | subject, scenario, "followup.edit", followup,
|
| 1567 | followup_payload.strip(), run_index, meta, admitted,
|
| 1568 | )
|
| 1569 | )
|
| 1570 | if row_returncode(agent_step("followup.checkpoint", [vcs, "commit"], followup)) != 0:
|
| 1571 | failure = "followup.checkpoint_failed"
|
| 1572 | else:
|
| 1573 | head_result = run_bounded_harness_command([vcs, "hash"], followup)
|
| 1574 | expected_head = head_result.stdout.strip() if head_result.returncode == 0 else None
|
| 1575 | if row_returncode(
|
| 1576 | agent_step(
|
| 1577 | "followup.publish",
|
| 1578 | [vcs, "push", "--repo", repo_spec, "--remote", remote],
|
| 1579 | followup,
|
| 1580 | )
|
| 1581 | ) != 0:
|
| 1582 | failure = "followup.publish_failed"
|
| 1583 | rows.append(
|
| 1584 | unmeasured_step_row(
|
| 1585 | subject, scenario, "followup.ci.observe", run_index, meta,
|
| 1586 | "owned local oak serve has no configured CI provider",
|
| 1587 | )
|
| 1588 | )
|
| 1589 | rows.append(
|
| 1590 | unmeasured_step_row(
|
| 1591 | subject, scenario, "followup.review_handoff", run_index, meta,
|
| 1592 | "owned local oak serve does not provide hosted review semantics",
|
| 1593 | )
|
| 1594 | )
|
| 1595 |
|
| 1596 | payload_ok = False
|
| 1597 | verifier_head: str | None = None
|
| 1598 | advertised_head: str | None = None
|
| 1599 | if failure is None and branch is not None:
|
| 1600 | clone = run_bounded_harness_command(
|
| 1601 | [
|
| 1602 | vcs, "clone", repo_spec, str(verifier), "--remote", remote,
|
| 1603 | "--branch", branch, "--allow-legacy-scope",
|
| 1604 | ],
|
| 1605 | run_root,
|
| 1606 | )
|
| 1607 | if clone.returncode == 0:
|
| 1608 | payload_path = verifier / "task-notes.md"
|
| 1609 | payload_ok = payload_path.exists() and payload_path.read_text() == expected_payload
|
| 1610 | verifier_result = run_bounded_harness_command([vcs, "hash"], verifier)
|
| 1611 | verifier_head = verifier_result.stdout.strip() if verifier_result.returncode == 0 else None
|
| 1612 | advertised_head = remote_branch_head(remote, repo_spec, branch)
|
| 1613 | else:
|
| 1614 | failure = "verifier_clone_failed"
|
| 1615 | head_ok = bool(expected_head and expected_head == verifier_head == advertised_head)
|
| 1616 | if not payload_ok:
|
| 1617 | failure = failure or "local_oak_payload_oracle_failed"
|
| 1618 | if not head_ok:
|
| 1619 | failure = failure or "local_oak_head_oracle_failed"
|
| 1620 | finally:
|
| 1621 | if server is not None:
|
| 1622 | stop_owned_process(server)
|
| 1623 | server_stopped = server is None or server.poll() is not None
|
| 1624 | cleanup_started = time.perf_counter()
|
| 1625 | for path in (seed, cold, followup, verifier, server_root):
|
| 1626 | shutil.rmtree(path, ignore_errors=True)
|
| 1627 | rows.append(
|
| 1628 | harness_step_row(
|
| 1629 | subject,
|
| 1630 | scenario,
|
| 1631 | "cleanup.local_server_and_checkouts",
|
| 1632 | run_index,
|
| 1633 | meta,
|
| 1634 | (time.perf_counter() - cleanup_started) * 1000,
|
| 1635 | )
|
| 1636 | )
|
| 1637 |
|
| 1638 | total = phase_total_row(rows, subject, scenario, run_index, meta, "agent_task_oak_local.total")
|
| 1639 | total.update(
|
| 1640 | {
|
| 1641 | "oracle_payload_on_remote_branch": payload_ok and head_ok,
|
| 1642 | "oracle_payload_matches": payload_ok,
|
| 1643 | "oracle_branch_head_matches": head_ok,
|
| 1644 | "server_process_stopped": server_stopped,
|
| 1645 | "acquisition_models": {
|
| 1646 | "cold": "fresh_full_clone",
|
| 1647 | "followup": "second_fresh_full_clone",
|
| 1648 | },
|
| 1649 | "repository_cache_state": {
|
| 1650 | "cold": "fresh_destination_no_shared_object_store",
|
| 1651 | "followup": "no_shared_object_store",
|
| 1652 | },
|
| 1653 | "os_cache_state": "unknown",
|
| 1654 | "os_cache_state_reason": "page_cache_not_purged_or_instrumented",
|
| 1655 | "network_bytes": None,
|
| 1656 | "network_bytes_reason": "loopback transport not observed by byte proxy",
|
| 1657 | "hosted_ci_configured": False,
|
| 1658 | "oracle_probe_calls": 4,
|
| 1659 | # The expected-head hash plus verifier clone/hash/HTTP head probe
|
| 1660 | # are oracle/setup commands, not agent-facing tool calls.
|
| 1661 | "harness_command_count": total["harness_command_count"] + 4,
|
| 1662 | }
|
| 1663 | )
|
| 1664 | if failure or not server_stopped:
|
| 1665 | total["returncode"] = 1
|
| 1666 | total["stderr"] = failure or "local_oak_server_cleanup_failed"
|
| 1667 | rows.append(total)
|
| 1668 | return rows
|
| 1669 |
|
| 1670 |
|
| 1671 | def run_legacy_selected_materialization_campaign(
|
| 1672 | subjects: list[Subject],
|
| 1673 | server_bin: Path | None,
|
| 1674 | server_source: Path | None,
|
| 1675 | server_source_head: str | None,
|
| 1676 | baseline_subject_name: str | None,
|
| 1677 | run_root: Path,
|
| 1678 | runs: int,
|
| 1679 | metadata: dict[str, Any],
|
| 1680 | admitted: int,
|
| 1681 | randomize_order: bool,
|
| 1682 | random_seed: int,
|
| 1683 | timing_controls_requested: bool,
|
| 1684 | ) -> list[dict[str, Any]]:
|
| 1685 | """Measure one legacy selected-branch clone against one fixed Serve fixture."""
|
| 1686 | scenario = LEGACY_SELECTED_MATERIALIZATION_SCENARIO
|
| 1687 | oak_subjects = [subject for subject in subjects if subject.kind == "oak"]
|
| 1688 | if not oak_subjects:
|
| 1689 | return []
|
| 1690 | if (
|
| 1691 | server_bin is None
|
| 1692 | or not server_bin.is_absolute()
|
| 1693 | or not server_bin.is_file()
|
| 1694 | or not os.access(server_bin, os.X_OK)
|
| 1695 | ):
|
| 1696 | return [
|
| 1697 | unmeasured_step_row(
|
| 1698 | subject,
|
| 1699 | scenario,
|
| 1700 | "scenario.skip",
|
| 1701 | 0,
|
| 1702 | metadata,
|
| 1703 | "--oak-serve-bin must name an absolute executable used only as the fixed Serve fixture",
|
| 1704 | )
|
| 1705 | for subject in oak_subjects
|
| 1706 | ]
|
| 1707 | if server_source_head is None or re.fullmatch(r"[0-9a-f]{64}", server_source_head) is None:
|
| 1708 | return [
|
| 1709 | unmeasured_step_row(
|
| 1710 | subject,
|
| 1711 | scenario,
|
| 1712 | "scenario.skip",
|
| 1713 | 0,
|
| 1714 | metadata,
|
| 1715 | "the fixed Serve binary's exact 64-hex --oak-serve-source-head is required",
|
| 1716 | )
|
| 1717 | for subject in oak_subjects
|
| 1718 | ]
|
| 1719 | baseline_subject = next(
|
| 1720 | (subject for subject in oak_subjects if subject.name == baseline_subject_name), None
|
| 1721 | )
|
| 1722 | if baseline_subject is None:
|
| 1723 | return [
|
| 1724 | unmeasured_step_row(
|
| 1725 | subject,
|
| 1726 | scenario,
|
| 1727 | "scenario.skip",
|
| 1728 | 0,
|
| 1729 | metadata,
|
| 1730 | "--materialization-baseline-subject must name one enabled Oak subject",
|
| 1731 | )
|
| 1732 | for subject in oak_subjects
|
| 1733 | ]
|
| 1734 | git_bin = shutil.which("git")
|
| 1735 | if git_bin is None:
|
| 1736 | return [
|
| 1737 | unmeasured_step_row(
|
| 1738 | subject,
|
| 1739 | scenario,
|
| 1740 | "scenario.skip",
|
| 1741 | 0,
|
| 1742 | metadata,
|
| 1743 | "git is required only to create the deterministic main-history fixture",
|
| 1744 | )
|
| 1745 | for subject in oak_subjects
|
| 1746 | ]
|
| 1747 |
|
| 1748 | server_bin = server_bin.resolve()
|
| 1749 | resolved_server_source: Path | None = None
|
| 1750 | if server_source is not None:
|
| 1751 | if not server_source.is_absolute() or not server_source.is_dir():
|
| 1752 | return [
|
| 1753 | unmeasured_step_row(
|
| 1754 | subject,
|
| 1755 | scenario,
|
| 1756 | "scenario.skip",
|
| 1757 | 0,
|
| 1758 | metadata,
|
| 1759 | "--oak-serve-source must name an absolute read-only source checkout",
|
| 1760 | )
|
| 1761 | for subject in oak_subjects
|
| 1762 | ]
|
| 1763 | resolved_server_source = server_source.resolve()
|
| 1764 | source_head_result = run_bounded_harness_command(
|
| 1765 | [str(server_bin), "hash"],
|
| 1766 | resolved_server_source,
|
| 1767 | env=isolated_oak_fixture_env(run_root / "source-check"),
|
| 1768 | )
|
| 1769 | if source_head_result.returncode != 0 or source_head_result.stdout.strip() != server_source_head:
|
| 1770 | return [
|
| 1771 | unmeasured_step_row(
|
| 1772 | subject,
|
| 1773 | scenario,
|
| 1774 | "scenario.skip",
|
| 1775 | 0,
|
| 1776 | metadata,
|
| 1777 | "fixed Serve source checkout does not match --oak-serve-source-head",
|
| 1778 | )
|
| 1779 | for subject in oak_subjects
|
| 1780 | ]
|
| 1781 | baseline_binary_sha256 = binary_sha256(baseline_subject.bin)
|
| 1782 | timing_controls_met = materialization_timing_controls_met(
|
| 1783 | metadata, runs, randomize_order, timing_controls_requested
|
| 1784 | )
|
| 1785 | fixture_root = run_root / "legacy-selected-materialization-fixture"
|
| 1786 | server_root = fixture_root / "server"
|
| 1787 | git_seed = fixture_root / "git-seed.git"
|
| 1788 | seed = fixture_root / "seed"
|
| 1789 | task_config_root = fixture_root / "config"
|
| 1790 | for path in (server_root, git_seed, task_config_root):
|
| 1791 | path.mkdir(parents=True, exist_ok=True)
|
| 1792 | fixture_env = isolated_oak_fixture_env(task_config_root)
|
| 1793 | fixture_meta = {
|
| 1794 | **metadata,
|
| 1795 | "benchmark_track": "oak-local-loopback-materialization",
|
| 1796 | "remote_transport": "loopback_http",
|
| 1797 | "remote_server": "127.0.0.1",
|
| 1798 | "serve_binary": str(server_bin),
|
| 1799 | "serve_binary_sha256": binary_sha256(server_bin),
|
| 1800 | "serve_source": str(resolved_server_source) if resolved_server_source else None,
|
| 1801 | "serve_source_checkout_verification": (
|
| 1802 | "exact_head_verified" if resolved_server_source else "unavailable_binary_hash_and_head_supplied"
|
| 1803 | ),
|
| 1804 | "serve_source_head": server_source_head,
|
| 1805 | "serve_version": subject_version(Subject("serve_fixture", "oak", "Serve fixture", server_bin)),
|
| 1806 | "baseline_subject": baseline_subject.name,
|
| 1807 | "baseline_binary_sha256": baseline_binary_sha256,
|
| 1808 | "fixture_id": "legacy-selected-materialization-192x16k-v1",
|
| 1809 | "fixture_regular_files": 192,
|
| 1810 | "fixture_regular_file_bytes": 16 * 1024,
|
| 1811 | "fixture_empty_files": 1,
|
| 1812 | "fixture_executable_files": 8,
|
| 1813 | "fixture_selected_only_files": 1,
|
| 1814 | "os_cache_state": "warm_or_unknown_after_required_fixture_probe",
|
| 1815 | "materialization_timing_controls_met": timing_controls_met,
|
| 1816 | "materialization_timing_controls_reason": (
|
| 1817 | "harness_controls_met_aa_and_quiet_host_evidence_still_required"
|
| 1818 | if timing_controls_met
|
| 1819 | else "correctness_only_or_harness_controls_not_met"
|
| 1820 | ),
|
| 1821 | }
|
| 1822 | rows: list[dict[str, Any]] = []
|
| 1823 | setup_subject = baseline_subject
|
| 1824 |
|
| 1825 | def setup_command(operation: str, command: list[str], cwd: Path) -> bool:
|
| 1826 | started = time.perf_counter()
|
| 1827 | result = run_bounded_harness_command(command, cwd, env=fixture_env)
|
| 1828 | rows.append(
|
| 1829 | harness_step_row(
|
| 1830 | setup_subject,
|
| 1831 | scenario,
|
| 1832 | operation,
|
| 1833 | 0,
|
| 1834 | fixture_meta,
|
| 1835 | (time.perf_counter() - started) * 1000,
|
| 1836 | returncode=0 if result.returncode == 0 else result.returncode,
|
| 1837 | harness_command_count=1,
|
| 1838 | )
|
| 1839 | )
|
| 1840 | if result.returncode != 0:
|
| 1841 | rows[-1]["stderr"] = result.stderr[-4000:]
|
| 1842 | return result.returncode == 0
|
| 1843 |
|
| 1844 | server: subprocess.Popen[Any] | None = None
|
| 1845 | proxy: CountingProxy | None = None
|
| 1846 | failure: str | None = None
|
| 1847 | expected_artifacts: list[str] | None = None
|
| 1848 | baseline_cache: dict[str, Any] | None = None
|
| 1849 | baseline_receipt_contract: dict[str, Any] | None = None
|
| 1850 | baseline_network_contract: dict[str, Any] | None = None
|
| 1851 | expected_status_contract: dict[str, Any] | None = None
|
| 1852 | selected_branch = "selected-materialization"
|
| 1853 | repo_spec = "qa/materialization-bench"
|
| 1854 | try:
|
| 1855 | for index in range(192):
|
| 1856 | directory = git_seed / f"group-{index % 12:02d}"
|
| 1857 | directory.mkdir(parents=True, exist_ok=True)
|
| 1858 | digest = hashlib.sha256(f"oak-materialization-{index}".encode()).digest()
|
| 1859 | content = (digest * ((16 * 1024 + len(digest) - 1) // len(digest)))[: 16 * 1024]
|
| 1860 | path = directory / f"file-{index:03d}.bin"
|
| 1861 | path.write_bytes(content)
|
| 1862 | if index < 8:
|
| 1863 | path.chmod(0o755)
|
| 1864 | (git_seed / "empty.txt").write_bytes(b"")
|
| 1865 | for operation, command in (
|
| 1866 | ("setup.git_init", [git_bin, "init", "-q", "-b", "main"]),
|
| 1867 | ("setup.git_add", [git_bin, "add", "."]),
|
| 1868 | ("setup.git_commit", [git_bin, "commit", "-q", "-m", "fixture main"]),
|
| 1869 | ):
|
| 1870 | if not setup_command(operation, command, git_seed):
|
| 1871 | failure = f"{operation}_failed"
|
| 1872 | break
|
| 1873 |
|
| 1874 | port = reserve_loopback_port()
|
| 1875 | direct_remote = f"http://127.0.0.1:{port}"
|
| 1876 | if failure is None:
|
| 1877 | server = subprocess.Popen(
|
| 1878 | [
|
| 1879 | str(server_bin),
|
| 1880 | "serve",
|
| 1881 | "--dir",
|
| 1882 | str(server_root),
|
| 1883 | "--host",
|
| 1884 | "127.0.0.1",
|
| 1885 | "--port",
|
| 1886 | str(port),
|
| 1887 | ],
|
| 1888 | cwd=fixture_root,
|
| 1889 | env=fixture_env,
|
| 1890 | stdout=subprocess.DEVNULL,
|
| 1891 | stderr=subprocess.DEVNULL,
|
| 1892 | start_new_session=os.name == "posix",
|
| 1893 | )
|
| 1894 | if not wait_for_loopback_port(server, port):
|
| 1895 | failure = "setup.server_not_ready"
|
| 1896 | if failure is None:
|
| 1897 | for operation, command, cwd in (
|
| 1898 | ("setup.oak_import", [str(server_bin), "clone", str(git_seed), str(seed)], fixture_root),
|
| 1899 | ("setup.branch", [str(server_bin), "switch", "-c", selected_branch], seed),
|
| 1900 | ):
|
| 1901 | if not setup_command(operation, command, cwd):
|
| 1902 | failure = f"{operation}_failed"
|
| 1903 | break
|
| 1904 | if failure is None:
|
| 1905 | (seed / "selected.txt").write_text("selected branch\n")
|
| 1906 | for operation, command in (
|
| 1907 | ("setup.oak_commit", [str(server_bin), "commit"]),
|
| 1908 | (
|
| 1909 | "setup.oak_push",
|
| 1910 | [str(server_bin), "push", "--repo", repo_spec, "--remote", direct_remote],
|
| 1911 | ),
|
| 1912 | ):
|
| 1913 | if not setup_command(operation, command, seed):
|
| 1914 | failure = f"{operation}_failed"
|
| 1915 | break
|
| 1916 |
|
| 1917 | expected_head_result = (
|
| 1918 | run_bounded_harness_command([str(server_bin), "hash"], seed, env=fixture_env)
|
| 1919 | if failure is None
|
| 1920 | else None
|
| 1921 | )
|
| 1922 | expected_head = (
|
| 1923 | expected_head_result.stdout.strip()
|
| 1924 | if expected_head_result is not None and expected_head_result.returncode == 0
|
| 1925 | else None
|
| 1926 | )
|
| 1927 | expected_history = (
|
| 1928 | normalized_history(json_command([str(server_bin), "log", "--json"], seed, fixture_env))
|
| 1929 | if failure is None
|
| 1930 | else None
|
| 1931 | )
|
| 1932 | expected_worktree = normalized_worktree_snapshot(seed) if failure is None else None
|
| 1933 | if not expected_head or expected_history is None or expected_worktree is None:
|
| 1934 | failure = failure or "setup.fixture_oracle_failed"
|
| 1935 |
|
| 1936 | if failure is None:
|
| 1937 | proxy = CountingProxy("127.0.0.1", port).start()
|
| 1938 | measured_remote = f"http://127.0.0.1:{proxy.address[1]}"
|
| 1939 | probe = fixture_root / "baseline-probe"
|
| 1940 | probe_env = isolated_oak_fixture_env(fixture_root / "baseline-probe-config")
|
| 1941 | proxy.reset_counts()
|
| 1942 | result = run_bounded_harness_command(
|
| 1943 | [
|
| 1944 | str(baseline_subject.bin),
|
| 1945 | "clone",
|
| 1946 | "--json",
|
| 1947 | "--allow-legacy-scope",
|
| 1948 | "--remote",
|
| 1949 | measured_remote,
|
| 1950 | "--branch",
|
| 1951 | selected_branch,
|
| 1952 | repo_spec,
|
| 1953 | str(probe),
|
| 1954 | ],
|
| 1955 | fixture_root,
|
| 1956 | env=probe_env,
|
| 1957 | )
|
| 1958 | receipt = _parse_json_object(result.stdout)
|
| 1959 | evidence = legacy_selected_receipt_evidence(receipt or {})
|
| 1960 | baseline_proxy_stats = proxy.snapshot()
|
| 1961 | baseline_cache = stat_cache_snapshot(probe)
|
| 1962 | expected_artifacts = normalized_oak_artifacts(probe)
|
| 1963 | baseline_receipt_contract = legacy_selected_receipt_contract(evidence)
|
| 1964 | baseline_network_contract = {
|
| 1965 | "connections": baseline_proxy_stats.connections,
|
| 1966 | "receipt": baseline_receipt_contract,
|
| 1967 | }
|
| 1968 | expected_status_contract = normalized_status(
|
| 1969 | json_command([str(baseline_subject.bin), "status", "--json"], probe, probe_env)
|
| 1970 | )
|
| 1971 | probe_checks = {
|
| 1972 | "returncode": result.returncode == 0,
|
| 1973 | "legacy_route": evidence["legacy_route_observed"],
|
| 1974 | "passes_two": evidence["materialization_passes_observed"] == 2,
|
| 1975 | "head": evidence["head"] == expected_head,
|
| 1976 | "manifest": bool(evidence["manifest"]),
|
| 1977 | "artifacts": expected_artifacts is not None,
|
| 1978 | "cache": normalized_cache_contract(baseline_cache) is not None,
|
| 1979 | "status": expected_status_contract is not None,
|
| 1980 | "network": baseline_proxy_stats.connection_errors == 0,
|
| 1981 | }
|
| 1982 | if not all(probe_checks.values()):
|
| 1983 | failed_checks = ",".join(key for key, passed in probe_checks.items() if not passed)
|
| 1984 | failure = f"baseline_probe_failed:{failed_checks}"
|
| 1985 | if isinstance(baseline_cache, dict) and baseline_cache.get("error"):
|
| 1986 | failure += f":{baseline_cache['error']}"
|
| 1987 | shutil.rmtree(probe, ignore_errors=True)
|
| 1988 |
|
| 1989 | if failure is None and proxy is not None:
|
| 1990 | measured_remote = f"http://127.0.0.1:{proxy.address[1]}"
|
| 1991 | order_rng = random.Random(random_seed)
|
| 1992 | for run_index in range(runs):
|
| 1993 | run_subjects = list(oak_subjects)
|
| 1994 | if randomize_order:
|
| 1995 | order_rng.shuffle(run_subjects)
|
| 1996 | for subject in run_subjects:
|
| 1997 | destination = run_root / f"materialization-{subject.name}-r{run_index}"
|
| 1998 | subject_env = isolated_oak_fixture_env(
|
| 1999 | run_root / f"materialization-config-{subject.name}-r{run_index}"
|
| 2000 | )
|
| 2001 | proxy.reset_counts()
|
| 2002 | row = step_row(
|
| 2003 | subject,
|
| 2004 | scenario,
|
| 2005 | "legacy_selected.acquire",
|
| 2006 | [
|
| 2007 | str(subject.bin),
|
| 2008 | "clone",
|
| 2009 | "--json",
|
| 2010 | "--allow-legacy-scope",
|
| 2011 | "--remote",
|
| 2012 | measured_remote,
|
| 2013 | "--branch",
|
| 2014 | selected_branch,
|
| 2015 | repo_spec,
|
| 2016 | str(destination),
|
| 2017 | ],
|
| 2018 | fixture_root,
|
| 2019 | run_index,
|
| 2020 | {
|
| 2021 | **fixture_meta,
|
| 2022 | "run_subject_order": [item.name for item in run_subjects],
|
| 2023 | },
|
| 2024 | admitted,
|
| 2025 | preserve_stdout=True,
|
| 2026 | env=subject_env,
|
| 2027 | timeout_seconds=INTERRUPTED_PUBLICATION_PROCESS_TIMEOUT_S,
|
| 2028 | )
|
| 2029 | receipt = _parse_json_object(row.get("stdout"))
|
| 2030 | evidence = legacy_selected_receipt_evidence(receipt or {})
|
| 2031 | proxy_stats = proxy.snapshot()
|
| 2032 | cache_before_status = stat_cache_snapshot(destination)
|
| 2033 | artifacts = normalized_oak_artifacts(destination)
|
| 2034 | trial_meta = {
|
| 2035 | **fixture_meta,
|
| 2036 | "run_subject_order": [item.name for item in run_subjects],
|
| 2037 | }
|
| 2038 | status_row = step_row(
|
| 2039 | subject,
|
| 2040 | scenario,
|
| 2041 | "legacy_selected.first_status",
|
| 2042 | [str(subject.bin), "status", "--json"],
|
| 2043 | destination,
|
| 2044 | run_index,
|
| 2045 | trial_meta,
|
| 2046 | admitted,
|
| 2047 | preserve_stdout=True,
|
| 2048 | env=subject_env,
|
| 2049 | timeout_seconds=INTERRUPTED_PUBLICATION_PROCESS_TIMEOUT_S,
|
| 2050 | )
|
| 2051 | status = normalized_status(_parse_json_object(status_row.get("stdout")) or {})
|
| 2052 | commit_row = step_row(
|
| 2053 | subject,
|
| 2054 | scenario,
|
| 2055 | "legacy_selected.first_noop_commit",
|
| 2056 | [str(subject.bin), "commit"],
|
| 2057 | destination,
|
| 2058 | run_index,
|
| 2059 | trial_meta,
|
| 2060 | admitted,
|
| 2061 | env=subject_env,
|
| 2062 | timeout_seconds=INTERRUPTED_PUBLICATION_PROCESS_TIMEOUT_S,
|
| 2063 | )
|
| 2064 | history = normalized_history(
|
| 2065 | json_command([str(subject.bin), "log", "--json"], destination, subject_env)
|
| 2066 | )
|
| 2067 | head_result = run_bounded_harness_command(
|
| 2068 | [str(subject.bin), "hash"], destination, env=subject_env
|
| 2069 | )
|
| 2070 | local_head = head_result.stdout.strip() if head_result.returncode == 0 else None
|
| 2071 | worktree = normalized_worktree_snapshot(destination) if destination.exists() else None
|
| 2072 | advertised_head = remote_branch_head(direct_remote, repo_spec, selected_branch)
|
| 2073 | subject_sha256 = binary_sha256(subject.bin)
|
| 2074 | expected_passes = 2 if subject_sha256 == baseline_binary_sha256 else 1
|
| 2075 | receipt_contract = legacy_selected_receipt_contract(evidence)
|
| 2076 | network_contract = {
|
| 2077 | "connections": proxy_stats.connections,
|
| 2078 | "receipt": receipt_contract,
|
| 2079 | }
|
| 2080 | cache_contract = normalized_cache_contract(cache_before_status)
|
| 2081 | baseline_cache_contract = normalized_cache_contract(baseline_cache)
|
| 2082 | expected_cache_valid = subject_sha256 != baseline_binary_sha256
|
| 2083 | cache_valid = bool(
|
| 2084 | isinstance(cache_before_status, dict)
|
| 2085 | and cache_before_status.get("all_metadata_valid") is expected_cache_valid
|
| 2086 | )
|
| 2087 | status_ok = status == expected_status_contract
|
| 2088 | artifacts_ok = artifacts == expected_artifacts
|
| 2089 | network_ok = network_contract == baseline_network_contract
|
| 2090 | oracle_ok = bool(
|
| 2091 | row_returncode(row) == 0
|
| 2092 | and row_returncode(status_row) == 0
|
| 2093 | and row_returncode(commit_row) == 0
|
| 2094 | and evidence["legacy_route_observed"]
|
| 2095 | and evidence["snapshot_bound"] is False
|
| 2096 | and evidence["materialization_passes_observed"] == expected_passes
|
| 2097 | and evidence["head"] == expected_head == local_head == advertised_head
|
| 2098 | and evidence["manifest"] == baseline_receipt_contract["manifest"]
|
| 2099 | and evidence["written_paths"] == len(expected_worktree)
|
| 2100 | and receipt_contract == baseline_receipt_contract
|
| 2101 | and history == expected_history
|
| 2102 | and worktree == expected_worktree
|
| 2103 | and status_ok
|
| 2104 | and artifacts_ok
|
| 2105 | and cache_contract == baseline_cache_contract
|
| 2106 | and cache_valid
|
| 2107 | and network_ok
|
| 2108 | and proxy_stats.connection_errors == 0
|
| 2109 | )
|
| 2110 | row.update(
|
| 2111 | {
|
| 2112 | **evidence,
|
| 2113 | "network_connections_observed": proxy_stats.connections,
|
| 2114 | "network_client_to_server_bytes_observed": proxy_stats.client_to_upstream_bytes,
|
| 2115 | "network_server_to_client_bytes_observed": proxy_stats.upstream_to_client_bytes,
|
| 2116 | "network_connection_errors": proxy_stats.connection_errors,
|
| 2117 | "normalized_network_work": network_contract,
|
| 2118 | "oracle_network_work_matches_baseline": network_ok,
|
| 2119 | "oracle_head_matches": expected_head == local_head == advertised_head,
|
| 2120 | "oracle_manifest_matches": evidence["manifest"]
|
| 2121 | == baseline_receipt_contract["manifest"],
|
| 2122 | "oracle_receipt_contract_matches": receipt_contract
|
| 2123 | == baseline_receipt_contract,
|
| 2124 | "oracle_history_matches": history == expected_history,
|
| 2125 | "oracle_worktree_bytes_and_modes_match": worktree == expected_worktree,
|
| 2126 | "oracle_clean_status": status_ok,
|
| 2127 | "oracle_local_artifact_names_match": artifacts_ok,
|
| 2128 | "oracle_stat_cache_contract_matches": cache_contract
|
| 2129 | == baseline_cache_contract,
|
| 2130 | "stat_cache_metadata_valid_before_status": (
|
| 2131 | cache_before_status.get("all_metadata_valid")
|
| 2132 | if isinstance(cache_before_status, dict)
|
| 2133 | else None
|
| 2134 | ),
|
| 2135 | "stat_cache_rows_before_status": (
|
| 2136 | len(cache_before_status.get("rows", []))
|
| 2137 | if isinstance(cache_before_status, dict)
|
| 2138 | and isinstance(cache_before_status.get("rows"), list)
|
| 2139 | else None
|
| 2140 | ),
|
| 2141 | "stat_cache_expected_valid_before_status": expected_cache_valid,
|
| 2142 | "oracle_clone_residue_absent": artifacts_ok,
|
| 2143 | "oracle_passed": oracle_ok,
|
| 2144 | "repository_cache_state": "fresh_destination_no_shared_object_store",
|
| 2145 | }
|
| 2146 | )
|
| 2147 | for followup_row in (status_row, commit_row):
|
| 2148 | followup_row.update(
|
| 2149 | {
|
| 2150 | "materialization_timing_controls_met": timing_controls_met,
|
| 2151 | "stat_cache_expected_valid_before_first_status": expected_cache_valid,
|
| 2152 | "oracle_passed": oracle_ok,
|
| 2153 | }
|
| 2154 | )
|
| 2155 | if not oracle_ok:
|
| 2156 | row["returncode"] = 1
|
| 2157 | row["stderr"] = "legacy_selected_materialization_oracle_failed"
|
| 2158 | status_row["returncode"] = 1
|
| 2159 | commit_row["returncode"] = 1
|
| 2160 | total = phase_total_row(
|
| 2161 | [row, status_row, commit_row],
|
| 2162 | subject,
|
| 2163 | scenario,
|
| 2164 | run_index,
|
| 2165 | trial_meta,
|
| 2166 | "legacy_selected_clone_materialization.total",
|
| 2167 | )
|
| 2168 | total.update(
|
| 2169 | {
|
| 2170 | key: row[key]
|
| 2171 | for key in (
|
| 2172 | "materialization_passes_observed",
|
| 2173 | "network_connections_observed",
|
| 2174 | "network_client_to_server_bytes_observed",
|
| 2175 | "network_server_to_client_bytes_observed",
|
| 2176 | "oracle_passed",
|
| 2177 | "oracle_head_matches",
|
| 2178 | "oracle_manifest_matches",
|
| 2179 | "oracle_receipt_contract_matches",
|
| 2180 | "oracle_history_matches",
|
| 2181 | "oracle_worktree_bytes_and_modes_match",
|
| 2182 | "oracle_clean_status",
|
| 2183 | "oracle_local_artifact_names_match",
|
| 2184 | "oracle_stat_cache_contract_matches",
|
| 2185 | "stat_cache_metadata_valid_before_status",
|
| 2186 | "stat_cache_rows_before_status",
|
| 2187 | "oracle_clone_residue_absent",
|
| 2188 | "oracle_network_work_matches_baseline",
|
| 2189 | "serve_binary_sha256",
|
| 2190 | "serve_source_head",
|
| 2191 | "fixture_id",
|
| 2192 | "os_cache_state",
|
| 2193 | "materialization_timing_controls_met",
|
| 2194 | )
|
| 2195 | }
|
| 2196 | )
|
| 2197 | total.update(
|
| 2198 | {
|
| 2199 | "clone_elapsed_ms": row.get("elapsed_ms"),
|
| 2200 | "first_status_elapsed_ms": status_row.get("elapsed_ms"),
|
| 2201 | "first_noop_commit_elapsed_ms": commit_row.get("elapsed_ms"),
|
| 2202 | }
|
| 2203 | )
|
| 2204 | rows.extend([row, status_row, commit_row, total])
|
| 2205 | shutil.rmtree(destination, ignore_errors=True)
|
| 2206 | finally:
|
| 2207 | if proxy is not None:
|
| 2208 | proxy.stop()
|
| 2209 | if server is not None:
|
| 2210 | stop_owned_process(server)
|
| 2211 | shutil.rmtree(fixture_root, ignore_errors=True)
|
| 2212 |
|
| 2213 | if failure is not None:
|
| 2214 | rows.append(
|
| 2215 | harness_step_row(
|
| 2216 | setup_subject,
|
| 2217 | scenario,
|
| 2218 | "scenario.fixture_failure",
|
| 2219 | 0,
|
| 2220 | fixture_meta,
|
| 2221 | 0.0,
|
| 2222 | returncode=1,
|
| 2223 | )
|
| 2224 | )
|
| 2225 | rows[-1]["stderr"] = failure
|
| 2226 | return rows
|
| 2227 |
|
| 2228 |
|
| 2229 | def parse_shallow_refresh_fixtures(text: str) -> list[tuple[int, int]]:
|
| 2230 | """Parse ``COMMITS:SEED[,COMMITS:SEED...]``; the first is primary, the rest held out."""
|
| 2231 | specs: list[tuple[int, int]] = []
|
| 2232 | for item in text.split(","):
|
| 2233 | item = item.strip()
|
| 2234 | if not item:
|
| 2235 | continue
|
| 2236 | commits, sep, seed = item.partition(":")
|
| 2237 | if not sep or not commits.isdigit() or not seed.isdigit() or int(commits) < 1:
|
| 2238 | raise ValueError(f"invalid fixture spec {item!r}; expected COMMITS:SEED")
|
| 2239 | specs.append((int(commits), int(seed)))
|
| 2240 | if not specs:
|
| 2241 | raise ValueError("at least one COMMITS:SEED fixture is required")
|
| 2242 | return specs
|
| 2243 |
|
| 2244 |
|
| 2245 | def _summed_or_none(rows: list[dict[str, Any]], key: str) -> int | None:
|
| 2246 | values = [row.get(key) for row in rows]
|
| 2247 | if any(value is None for value in values):
|
| 2248 | return None
|
| 2249 | return sum(int(value) for value in values)
|
| 2250 |
|
| 2251 |
|
| 2252 | def run_shallow_refresh_campaign(
|
| 2253 | subjects: list[Subject],
|
| 2254 | server_bin: Path | None,
|
| 2255 | server_source_head: str | None,
|
| 2256 | fixture_specs: list[tuple[int, int]],
|
| 2257 | run_root: Path,
|
| 2258 | runs: int,
|
| 2259 | metadata: dict[str, Any],
|
| 2260 | admitted: int,
|
| 2261 | randomize_order: bool,
|
| 2262 | random_seed: int,
|
| 2263 | injected_request_delay_ms: float,
|
| 2264 | ) -> list[dict[str, Any]]:
|
| 2265 | """Fresh shallow worker -> named task branch at main -> first parent refresh.
|
| 2266 |
|
| 2267 | One deterministic merge-history fixture per spec is imported into an owned
|
| 2268 | loopback ``oak serve`` (fixed ``--oak-serve-bin``) and a bare Git repo served
|
| 2269 | by ``GitHTTPServer``, both behind request-framing proxies, during untimed
|
| 2270 | setup. Each subject then runs its cells in (optionally shuffled) order.
|
| 2271 | """
|
| 2272 | scenario = SHALLOW_REFRESH_SCENARIO
|
| 2273 | rows: list[dict[str, Any]] = []
|
| 2274 | if not subjects:
|
| 2275 | return rows
|
| 2276 |
|
| 2277 | def skip_all(reason: str) -> list[dict[str, Any]]:
|
| 2278 | return [
|
| 2279 | unmeasured_step_row(subject, scenario, "scenario.skip", 0, metadata, reason)
|
| 2280 | for subject in subjects
|
| 2281 | ]
|
| 2282 |
|
| 2283 | if (
|
| 2284 | server_bin is None
|
| 2285 | or not server_bin.is_absolute()
|
| 2286 | or not server_bin.is_file()
|
| 2287 | or not os.access(server_bin, os.X_OK)
|
| 2288 | ):
|
| 2289 | return skip_all("--oak-serve-bin must name an absolute executable used only as the fixed Serve fixture")
|
| 2290 | git_bin = shutil.which("git")
|
| 2291 | if git_bin is None:
|
| 2292 | return skip_all("git is required to create the deterministic merge-history fixture")
|
| 2293 | server_bin = server_bin.resolve()
|
| 2294 | serve_meta = {
|
| 2295 | "benchmark_track": "loopback-shallow-refresh",
|
| 2296 | "remote_transport": "loopback_http",
|
| 2297 | "remote_server": "127.0.0.1",
|
| 2298 | "serve_binary": str(server_bin),
|
| 2299 | "serve_binary_sha256": binary_sha256(server_bin),
|
| 2300 | "serve_version": subject_version(Subject("serve_fixture", "oak", "Serve fixture", server_bin)),
|
| 2301 | "serve_source_head": server_source_head,
|
| 2302 | "injected_request_delay_ms": injected_request_delay_ms,
|
| 2303 | "injected_request_delay_kind": "fixed_per_request_sleep_in_proxy_not_netem",
|
| 2304 | "hosted_equivalent": False,
|
| 2305 | "hosted_equivalence_reason": (
|
| 2306 | "loopback oak serve has no bounded integrity preflight (clone needs --allow-legacy-scope) "
|
| 2307 | "and is not the hosted server; see serve_depth_behavior for whether --shallow was honored"
|
| 2308 | ),
|
| 2309 | "os_cache_state": "warm_or_unknown",
|
| 2310 | "repository_cache_state": "fresh_destination_per_trial",
|
| 2311 | }
|
| 2312 | for commits, seed in fixture_specs:
|
| 2313 | rows.extend(
|
| 2314 | _run_shallow_refresh_fixture(
|
| 2315 | subjects, server_bin, git_bin, commits, seed, run_root, runs,
|
| 2316 | {**metadata, **serve_meta}, admitted, randomize_order, random_seed,
|
| 2317 | injected_request_delay_ms,
|
| 2318 | )
|
| 2319 | )
|
| 2320 | return rows
|
| 2321 |
|
| 2322 |
|
| 2323 | def _run_shallow_refresh_fixture(
|
| 2324 | subjects: list[Subject],
|
| 2325 | server_bin: Path,
|
| 2326 | git_bin: str,
|
| 2327 | commits: int,
|
| 2328 | seed: int,
|
| 2329 | run_root: Path,
|
| 2330 | runs: int,
|
| 2331 | metadata: dict[str, Any],
|
| 2332 | admitted: int,
|
| 2333 | randomize_order: bool,
|
| 2334 | random_seed: int,
|
| 2335 | injected_request_delay_ms: float,
|
| 2336 | ) -> list[dict[str, Any]]:
|
| 2337 | scenario = SHALLOW_REFRESH_SCENARIO
|
| 2338 | fixture_root = (run_root / f"shallow-refresh-c{commits}-s{seed}").resolve()
|
| 2339 | served_git = fixture_root / "git-served"
|
| 2340 | for path in (fixture_root, served_git):
|
| 2341 | path.mkdir(parents=True, exist_ok=True)
|
| 2342 | fixture_env = isolated_oak_fixture_env(fixture_root / "config")
|
| 2343 | rows: list[dict[str, Any]] = []
|
| 2344 | setup_subject = subjects[0]
|
| 2345 | fixture_meta: dict[str, Any] = {**metadata, "fixture_commits": commits, "fixture_seed": str(seed)}
|
| 2346 | repo_spec = "qa/shallow-refresh"
|
| 2347 | task_branch = SHALLOW_REFRESH_TASK_BRANCH
|
| 2348 | server: subprocess.Popen[Any] | None = None
|
| 2349 | oak_proxy: RequestCountingProxy | None = None
|
| 2350 | git_proxy: RequestCountingProxy | None = None
|
| 2351 | git_server: GitHTTPServer | None = None
|
| 2352 | failure: str | None = None
|
| 2353 |
|
| 2354 | def setup_command(operation: str, command: list[str], cwd: Path) -> bool:
|
| 2355 | started = time.perf_counter()
|
| 2356 | result = run_bounded_harness_command(command, cwd, env=fixture_env)
|
| 2357 | rows.append(
|
| 2358 | harness_step_row(
|
| 2359 | setup_subject, scenario, operation, 0, fixture_meta,
|
| 2360 | (time.perf_counter() - started) * 1000,
|
| 2361 | returncode=0 if result.returncode == 0 else result.returncode,
|
| 2362 | harness_command_count=1,
|
| 2363 | )
|
| 2364 | )
|
| 2365 | if result.returncode != 0:
|
| 2366 | rows[-1]["stderr"] = result.stderr[-4000:]
|
| 2367 | return result.returncode == 0
|
| 2368 |
|
| 2369 | try:
|
| 2370 | manifest = merge_history_fixture.generate(
|
| 2371 | fixture_root / "src.git", commits, seed, git_bin=git_bin, env=fixture_env
|
| 2372 | )
|
| 2373 | fixture_meta.update(manifest.as_dict())
|
| 2374 | expected_tree = manifest.tree_content_hash
|
| 2375 | bare = served_git / "fixture.git"
|
| 2376 | if not setup_command(
|
| 2377 | "setup.git_bare", [git_bin, "clone", "-q", "--bare", str(fixture_root / "src.git"), str(bare)],
|
| 2378 | fixture_root,
|
| 2379 | ):
|
| 2380 | failure = "setup.git_bare_failed"
|
| 2381 | port = reserve_loopback_port()
|
| 2382 | direct_remote = f"http://127.0.0.1:{port}"
|
| 2383 | if failure is None:
|
| 2384 | server = subprocess.Popen(
|
| 2385 | [str(server_bin), "serve", "--dir", str(fixture_root / "oak-served"),
|
| 2386 | "--host", "127.0.0.1", "--port", str(port)],
|
| 2387 | cwd=fixture_root, env=fixture_env,
|
| 2388 | stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL,
|
| 2389 | start_new_session=os.name == "posix",
|
| 2390 | )
|
| 2391 | if not wait_for_loopback_port(server, port):
|
| 2392 | failure = "setup.server_not_ready"
|
| 2393 | seed_dir = fixture_root / "oak-seed"
|
| 2394 | if failure is None:
|
| 2395 | for operation, command, cwd in (
|
| 2396 | ("setup.oak_import", [str(server_bin), "clone", str(fixture_root / "src.git"), str(seed_dir)], fixture_root),
|
| 2397 | ("setup.oak_push", [str(server_bin), "push", "--repo", repo_spec, "--remote", direct_remote], seed_dir),
|
| 2398 | ):
|
| 2399 | if not setup_command(operation, command, cwd):
|
| 2400 | failure = f"{operation}_failed"
|
| 2401 | break
|
| 2402 | oak_main_head = remote_branch_head(direct_remote, repo_spec, "main") if failure is None else None
|
| 2403 | git_main = run_bounded_harness_command([git_bin, "rev-parse", "refs/heads/main"], bare, env=fixture_env)
|
| 2404 | git_main_head = git_main.stdout.strip() if git_main.returncode == 0 else None
|
| 2405 | import_tree = shallow_refresh.worktree_tree_hash(seed_dir) if failure is None else None
|
| 2406 | fixture_meta.update(
|
| 2407 | {
|
| 2408 | "server_oak_main_head": oak_main_head,
|
| 2409 | "server_git_main_head": git_main_head,
|
| 2410 | "oracle_import_tree_matches_fixture": import_tree == expected_tree,
|
| 2411 | }
|
| 2412 | )
|
| 2413 | if failure is None and not oak_main_head:
|
| 2414 | failure = "setup.oak_server_main_head_unobserved"
|
| 2415 | if failure is None and git_main_head != manifest.head:
|
| 2416 | failure = "setup.git_server_head_mismatch"
|
| 2417 | if failure is None and import_tree != expected_tree:
|
| 2418 | failure = "setup.oak_import_tree_mismatch"
|
| 2419 | shutil.rmtree(seed_dir, ignore_errors=True)
|
| 2420 |
|
| 2421 | if failure is None:
|
| 2422 | git_server = GitHTTPServer(served_git, git_bin=git_bin).start()
|
| 2423 | oak_proxy = RequestCountingProxy(
|
| 2424 | "127.0.0.1", port, path_suffixes=(COMMITS_INFO_SUFFIX,),
|
| 2425 | injected_request_delay_ms=injected_request_delay_ms,
|
| 2426 | ).start()
|
| 2427 | git_proxy = RequestCountingProxy(
|
| 2428 | "127.0.0.1", git_server.server_port, path_suffixes=(COMMITS_INFO_SUFFIX,),
|
| 2429 | injected_request_delay_ms=injected_request_delay_ms,
|
| 2430 | ).start()
|
| 2431 | oak_remote = f"http://127.0.0.1:{oak_proxy.address[1]}"
|
| 2432 | git_url = f"http://127.0.0.1:{git_proxy.address[1]}/fixture.git"
|
| 2433 | order_rng = random.Random(f"{random_seed}:{commits}:{seed}")
|
| 2434 | for run_index in range(runs):
|
| 2435 | run_subjects = list(subjects)
|
| 2436 | if randomize_order:
|
| 2437 | order_rng.shuffle(run_subjects)
|
| 2438 | trial_meta = {**fixture_meta, "run_subject_order": [s.name for s in run_subjects]}
|
| 2439 | for subject in run_subjects:
|
| 2440 | print(
|
| 2441 | f"[run] {scenario} {manifest.fixture_id} run={run_index} subject={subject.name}",
|
| 2442 | flush=True,
|
| 2443 | )
|
| 2444 | tag = f"{subject.name}-c{commits}-s{seed}-r{run_index}"
|
| 2445 | subject_env = isolated_oak_fixture_env(run_root / f"sr-config-{tag}")
|
| 2446 | if subject.kind == "oak":
|
| 2447 | for cell in ("shallow", "shallow_pull"):
|
| 2448 | rows.extend(
|
| 2449 | _shallow_refresh_oak_trial(
|
| 2450 | subject, oak_proxy, oak_remote, repo_spec, oak_main_head,
|
| 2451 | expected_tree, run_root / f"sr-oak-{cell}-{tag}", run_index,
|
| 2452 | trial_meta, admitted,
|
| 2453 | isolated_oak_fixture_env(run_root / f"sr-config-{cell}-{tag}"),
|
| 2454 | cell=cell,
|
| 2455 | )
|
| 2456 | )
|
| 2457 | elif subject.kind == "git":
|
| 2458 | rows.extend(
|
| 2459 | _shallow_refresh_git_trials(
|
| 2460 | subject, git_proxy, git_url, git_main_head, expected_tree,
|
| 2461 | run_root, tag, run_index, trial_meta, admitted, subject_env,
|
| 2462 | )
|
| 2463 | )
|
| 2464 | else:
|
| 2465 | rows.append(
|
| 2466 | unmeasured_step_row(
|
| 2467 | subject, scenario, "scenario.skip", run_index, trial_meta,
|
| 2468 | f"subject kind {subject.kind!r} has no shallow-refresh cell",
|
| 2469 | )
|
| 2470 | )
|
| 2471 | except (OSError, RuntimeError, ValueError) as error:
|
| 2472 | failure = failure or f"setup_exception:{type(error).__name__}:{error}"
|
| 2473 | finally:
|
| 2474 | for proxy in (oak_proxy, git_proxy):
|
| 2475 | if proxy is not None:
|
| 2476 | proxy.stop()
|
| 2477 | if git_server is not None:
|
| 2478 | git_server.stop()
|
| 2479 | if server is not None:
|
| 2480 | stop_owned_process(server)
|
| 2481 | shutil.rmtree(fixture_root, ignore_errors=True)
|
| 2482 |
|
| 2483 | if failure is not None:
|
| 2484 | rows.append(
|
| 2485 | harness_step_row(setup_subject, scenario, "scenario.fixture_failure", 0, fixture_meta, 0.0, returncode=1)
|
| 2486 | )
|
| 2487 | rows[-1]["stderr"] = failure
|
| 2488 | return rows
|
| 2489 |
|
| 2490 |
|
| 2491 | def _shallow_refresh_step(
|
| 2492 | subject: Subject,
|
| 2493 | proxy: RequestCountingProxy,
|
| 2494 | operation: str,
|
| 2495 | command: list[str],
|
| 2496 | cwd: Path,
|
| 2497 | run_index: int,
|
| 2498 | meta: dict[str, Any],
|
| 2499 | admitted: int,
|
| 2500 | env: dict[str, str],
|
| 2501 | ) -> dict[str, Any]:
|
| 2502 | proxy.reset_counts()
|
| 2503 | row = step_row(
|
| 2504 | subject, SHALLOW_REFRESH_SCENARIO, operation, command, cwd, run_index, meta, admitted,
|
| 2505 | env=env, timeout_seconds=SHALLOW_REFRESH_STEP_TIMEOUT_S,
|
| 2506 | )
|
| 2507 | row.update(
|
| 2508 | shallow_refresh.request_metric_fields(proxy.request_stats(), proxy.snapshot(), COMMITS_INFO_SUFFIX)
|
| 2509 | )
|
| 2510 | return row
|
| 2511 |
|
| 2512 |
|
| 2513 | def _shallow_refresh_total(
|
| 2514 | steps: list[dict[str, Any]],
|
| 2515 | subject: Subject,
|
| 2516 | run_index: int,
|
| 2517 | meta: dict[str, Any],
|
| 2518 | cell: str,
|
| 2519 | oracle: dict[str, Any],
|
| 2520 | extra: dict[str, Any],
|
| 2521 | ) -> dict[str, Any]:
|
| 2522 | total = phase_total_row(steps, subject, SHALLOW_REFRESH_SCENARIO, run_index, meta, f"{cell}.total")
|
| 2523 | total.update(oracle)
|
| 2524 | total.update(extra)
|
| 2525 | for key in (
|
| 2526 | "http_request_count",
|
| 2527 | "http_commits_info_request_count",
|
| 2528 | "network_wire_bytes_observed",
|
| 2529 | "network_connections_observed",
|
| 2530 | ):
|
| 2531 | total[key] = _summed_or_none(steps, key)
|
| 2532 | total["step_elapsed_ms"] = {row["operation"]: row.get("elapsed_ms") for row in steps}
|
| 2533 | if not oracle.get("oracle_passed"):
|
| 2534 | total["returncode"] = 1
|
| 2535 | total["stderr"] = total.get("stderr") or f"{cell}_final_state_oracle_failed"
|
| 2536 | for row in steps:
|
| 2537 | row["oracle_passed"] = oracle.get("oracle_passed")
|
| 2538 | return total
|
| 2539 |
|
| 2540 |
|
| 2541 | def _shallow_refresh_oak_trial(
|
| 2542 | subject: Subject,
|
| 2543 | proxy: RequestCountingProxy,
|
| 2544 | remote: str,
|
| 2545 | repo_spec: str,
|
| 2546 | expected_head: str | None,
|
| 2547 | expected_tree: str,
|
| 2548 | worker: Path,
|
| 2549 | run_index: int,
|
| 2550 | meta: dict[str, Any],
|
| 2551 | admitted: int,
|
| 2552 | env: dict[str, str],
|
| 2553 | *,
|
| 2554 | cell: str = "shallow",
|
| 2555 | ) -> list[dict[str, Any]]:
|
| 2556 | """Oak cells: ``shallow`` refreshes with ``fetch``; ``shallow_pull`` with ``pull``.
|
| 2557 |
|
| 2558 | Both start from a fresh shallow clone and create the task branch first, so
|
| 2559 | comparing them shows which command pays first-refresh ancestry work.
|
| 2560 | """
|
| 2561 | vcs = str(subject.bin)
|
| 2562 | refresh_command = {"shallow": [vcs, "fetch"], "shallow_pull": [vcs, "pull"]}[cell]
|
| 2563 | cell_meta = {
|
| 2564 | **meta,
|
| 2565 | "cell": "shallow_clone" if cell == "shallow" else "shallow_clone_pull",
|
| 2566 | "starting_state": "no_local_repository_fresh_shallow_clone",
|
| 2567 | }
|
| 2568 | steps: list[dict[str, Any]] = []
|
| 2569 | extra: dict[str, Any] = {}
|
| 2570 | try:
|
| 2571 | acquire = _shallow_refresh_step(
|
| 2572 | subject, proxy, f"{cell}.acquire",
|
| 2573 | [vcs, "clone", "--shallow", "--allow-legacy-scope", "--remote", remote, repo_spec, str(worker)],
|
| 2574 | worker.parent, run_index, cell_meta, admitted, env,
|
| 2575 | )
|
| 2576 | steps.append(acquire)
|
| 2577 | after_clone = shallow_refresh.oak_disk_state(worker, SHALLOW_REFRESH_TASK_BRANCH)
|
| 2578 | depth_behavior = shallow_refresh.serve_depth_behavior(after_clone["commit_rows"])
|
| 2579 | extra.update(
|
| 2580 | {
|
| 2581 | "oak_commit_rows_after_clone": after_clone["commit_rows"],
|
| 2582 | "serve_depth_behavior": depth_behavior,
|
| 2583 | }
|
| 2584 | )
|
| 2585 | if row_returncode(acquire) == 0:
|
| 2586 | for operation, command in (
|
| 2587 | (f"{cell}.branch_create", [vcs, "switch", "-c", SHALLOW_REFRESH_TASK_BRANCH]),
|
| 2588 | (f"{cell}.parent_refresh", refresh_command),
|
| 2589 | ):
|
| 2590 | row = _shallow_refresh_step(
|
| 2591 | subject, proxy, operation, command, worker, run_index, cell_meta, admitted, env
|
| 2592 | )
|
| 2593 | steps.append(row)
|
| 2594 | if row_returncode(row) != 0:
|
| 2595 | break
|
| 2596 | disk = shallow_refresh.oak_disk_state(worker, SHALLOW_REFRESH_TASK_BRANCH)
|
| 2597 | oracle = shallow_refresh.evaluate_oracle(
|
| 2598 | expected_branch=SHALLOW_REFRESH_TASK_BRANCH,
|
| 2599 | expected_head=expected_head,
|
| 2600 | expected_tree_hash=expected_tree,
|
| 2601 | observed_branch=disk["current_branch"],
|
| 2602 | observed_branch_head=disk["branch_head"],
|
| 2603 | observed_head=disk["head"],
|
| 2604 | worktree_tree_hash=shallow_refresh.worktree_tree_hash(worker),
|
| 2605 | )
|
| 2606 | extra.update(
|
| 2607 | {
|
| 2608 | "oracle_source": "oak_db_query_only_plus_server_branch_api_plus_fixture_model",
|
| 2609 | "oak_commit_rows_after_refresh": disk["commit_rows"],
|
| 2610 | "oak_ancestry_missing_edges_after_refresh": disk["ancestry_missing_edges"],
|
| 2611 | "oak_disk_state_error": disk["error"],
|
| 2612 | }
|
| 2613 | )
|
| 2614 | for row in steps:
|
| 2615 | row["serve_depth_behavior"] = depth_behavior
|
| 2616 | total = _shallow_refresh_total(steps, subject, run_index, cell_meta, cell, oracle, extra)
|
| 2617 | rows = [*steps, total]
|
| 2618 | if cell == "shallow" and oracle["oracle_passed"]:
|
| 2619 | # Separate measurement: a second branch creation in the now-refreshed
|
| 2620 | # worker. Isolates local switch cost from the first-refresh network work.
|
| 2621 | repeat = _shallow_refresh_step(
|
| 2622 | subject, proxy, "shallow.branch_create_repeat_local",
|
| 2623 | [vcs, "switch", "-c", f"{SHALLOW_REFRESH_TASK_BRANCH}-repeat"],
|
| 2624 | worker, run_index,
|
| 2625 | {**cell_meta, "starting_state": "refreshed_worker_after_first_parent_refresh"},
|
| 2626 | admitted, env,
|
| 2627 | )
|
| 2628 | repeat_disk = shallow_refresh.oak_disk_state(worker, f"{SHALLOW_REFRESH_TASK_BRANCH}-repeat")
|
| 2629 | repeat_ok = bool(
|
| 2630 | row_returncode(repeat) == 0
|
| 2631 | and repeat_disk["current_branch"] == f"{SHALLOW_REFRESH_TASK_BRANCH}-repeat"
|
| 2632 | and repeat_disk["branch_head"] == expected_head
|
| 2633 | )
|
| 2634 | repeat.update({"serve_depth_behavior": depth_behavior, "oracle_passed": repeat_ok})
|
| 2635 | if not repeat_ok:
|
| 2636 | repeat["returncode"] = 1
|
| 2637 | repeat["stderr"] = repeat.get("stderr") or "branch_create_repeat_oracle_failed"
|
| 2638 | rows.append(repeat)
|
| 2639 | return rows
|
| 2640 | finally:
|
| 2641 | shutil.rmtree(worker, ignore_errors=True)
|
| 2642 |
|
| 2643 |
|
| 2644 | def _shallow_refresh_git_trials(
|
| 2645 | subject: Subject,
|
| 2646 | proxy: RequestCountingProxy,
|
| 2647 | url: str,
|
| 2648 | expected_head: str | None,
|
| 2649 | expected_tree: str,
|
| 2650 | run_root: Path,
|
| 2651 | tag: str,
|
| 2652 | run_index: int,
|
| 2653 | meta: dict[str, Any],
|
| 2654 | admitted: int,
|
| 2655 | env: dict[str, str],
|
| 2656 | ) -> list[dict[str, Any]]:
|
| 2657 | git = str(subject.bin)
|
| 2658 | rows: list[dict[str, Any]] = []
|
| 2659 |
|
| 2660 | def finish(cell: str, cell_meta: dict[str, Any], steps: list[dict[str, Any]], worktree: Path) -> None:
|
| 2661 | disk = shallow_refresh.git_disk_state(worktree, git, env)
|
| 2662 | oracle = shallow_refresh.evaluate_oracle(
|
| 2663 | expected_branch=SHALLOW_REFRESH_TASK_BRANCH,
|
| 2664 | expected_head=expected_head,
|
| 2665 | expected_tree_hash=expected_tree,
|
| 2666 | observed_branch=disk["current_branch"],
|
| 2667 | observed_branch_head=disk["head"] if disk["current_branch"] == SHALLOW_REFRESH_TASK_BRANCH else None,
|
| 2668 | observed_head=disk["head"],
|
| 2669 | worktree_tree_hash=shallow_refresh.worktree_tree_hash(worktree),
|
| 2670 | )
|
| 2671 | extra = {
|
| 2672 | "oracle_source": "git_rev_parse_plus_server_bare_repo_plus_fixture_model",
|
| 2673 | "git_commit_rows_after_refresh": disk["commit_rows"],
|
| 2674 | "git_is_shallow_after_refresh": disk["shallow"],
|
| 2675 | "serve_depth_behavior": None,
|
| 2676 | }
|
| 2677 | rows.extend([*steps, _shallow_refresh_total(steps, subject, run_index, cell_meta, cell, oracle, extra)])
|
| 2678 |
|
| 2679 | shallow = run_root / f"sr-git-shallow-{tag}"
|
| 2680 | shallow_meta = {**meta, "cell": "shallow_clone", "starting_state": "no_local_repository_fresh_depth1_clone"}
|
| 2681 | try:
|
| 2682 | steps = [
|
| 2683 | _shallow_refresh_step(
|
| 2684 | subject, proxy, "shallow.acquire", [git, "clone", "-q", "--depth", "1", url, str(shallow)],
|
| 2685 | run_root, run_index, shallow_meta, admitted, env,
|
| 2686 | )
|
| 2687 | ]
|
| 2688 | if row_returncode(steps[0]) == 0:
|
| 2689 | for operation, command in (
|
| 2690 | ("shallow.branch_create", [git, "switch", "-q", "-c", SHALLOW_REFRESH_TASK_BRANCH]),
|
| 2691 | ("shallow.parent_refresh", [git, "fetch", "-q"]),
|
| 2692 | ):
|
| 2693 | steps.append(
|
| 2694 | _shallow_refresh_step(subject, proxy, operation, command, shallow, run_index, shallow_meta, admitted, env)
|
| 2695 | )
|
| 2696 | if row_returncode(steps[-1]) != 0:
|
| 2697 | break
|
| 2698 | finish("shallow", shallow_meta, steps, shallow)
|
| 2699 | finally:
|
| 2700 | shutil.rmtree(shallow, ignore_errors=True)
|
| 2701 |
|
| 2702 | full = run_root / f"sr-git-full-{tag}"
|
| 2703 | worktree = run_root / f"sr-git-worktree-{tag}"
|
| 2704 | warm_meta = {
|
| 2705 | **meta,
|
| 2706 | "cell": "warm_full_clone_worktree",
|
| 2707 | "starting_state": "warm_full_clone_created_untimed_not_equivalent_to_fresh_shallow_worker",
|
| 2708 | }
|
| 2709 | try:
|
| 2710 | setup_started = time.perf_counter()
|
| 2711 | clone = run_bounded_harness_command([git, "clone", "-q", url, str(full)], run_root, env=env)
|
| 2712 | setup_row = harness_step_row(
|
| 2713 | subject, SHALLOW_REFRESH_SCENARIO, "warm_worktree.setup_full_clone", run_index, warm_meta,
|
| 2714 | (time.perf_counter() - setup_started) * 1000,
|
| 2715 | returncode=0 if clone.returncode == 0 else clone.returncode, harness_command_count=1,
|
| 2716 | )
|
| 2717 | rows.append(setup_row)
|
| 2718 | steps = []
|
| 2719 | if clone.returncode == 0:
|
| 2720 | for operation, command, cwd in (
|
| 2721 | ("warm_worktree.branch_create",
|
| 2722 | [git, "worktree", "add", "-q", "-b", SHALLOW_REFRESH_TASK_BRANCH, str(worktree), "origin/main"], full),
|
| 2723 | ("warm_worktree.parent_refresh", [git, "fetch", "-q"], worktree),
|
| 2724 | ):
|
| 2725 | steps.append(
|
| 2726 | _shallow_refresh_step(subject, proxy, operation, command, cwd, run_index, warm_meta, admitted, env)
|
| 2727 | )
|
| 2728 | if row_returncode(steps[-1]) != 0:
|
| 2729 | break
|
| 2730 | finish("warm_worktree", warm_meta, steps, worktree)
|
| 2731 | finally:
|
| 2732 | shutil.rmtree(worktree, ignore_errors=True)
|
| 2733 | shutil.rmtree(full, ignore_errors=True)
|
| 2734 | return rows
|
| 2735 |
|
| 2736 |
|
| 2737 | def run_git_loop(
|
| 2738 | subject: Subject,
|
| 2739 | remote: oakbench_remotes.RemoteResolution,
|
| 2740 | run_root: Path,
|
| 2741 | run_index: int,
|
| 2742 | metadata: dict[str, Any],
|
| 2743 | admitted: int,
|
| 2744 | ) -> list[dict[str, Any]]:
|
| 2745 | scenario = "task_loop"
|
| 2746 | meta = {
|
| 2747 | **metadata,
|
| 2748 | **remote.row_fields(),
|
| 2749 | "benchmark_track": (
|
| 2750 | "core-equivalent"
|
| 2751 | if remote.transport == oakbench_remotes.TRANSPORT_LOCAL_FILE
|
| 2752 | else "agent-default"
|
| 2753 | ),
|
| 2754 | }
|
| 2755 | vcs = str(subject.bin)
|
| 2756 | rows: list[dict[str, Any]] = []
|
| 2757 |
|
| 2758 | seed_error = ensure_git_remote_seeded(subject, remote, run_root)
|
| 2759 | if seed_error:
|
| 2760 | return [skip_row(subject, scenario, run_index, meta, seed_error)]
|
| 2761 |
|
| 2762 | # Untimed: the base clone is the long-lived checkout worktrees hang off;
|
| 2763 | # an agent fleet pays it once, not per task.
|
| 2764 | base = run_root / f"tl-git-base-r{run_index}"
|
| 2765 | clone = run_untimed([vcs, "clone", "-q", str(remote.repo), str(base)], run_root)
|
| 2766 | if clone.returncode != 0:
|
| 2767 | return [skip_row(subject, scenario, run_index, meta, f"base clone failed: {clone.stderr.strip()[:200]}")]
|
| 2768 |
|
| 2769 | branches: dict[str, str] = {}
|
| 2770 | payloads: dict[str, str] = {}
|
| 2771 | for phase in PHASES:
|
| 2772 | workspace = run_root / f"tl-{subject.name}-r{run_index}-{phase}"
|
| 2773 | branch = oakbench_remotes.disposable_branch(
|
| 2774 | str(metadata["bench_id"]), f"taskloop-{phase}", subject.name, run_index
|
| 2775 | )
|
| 2776 | branches[phase] = branch
|
| 2777 | payload = f"payload {phase} run={run_index} bench={metadata['bench_id']}"
|
| 2778 | payloads[phase] = payload
|
| 2779 | acquire = step_row(
|
| 2780 | subject, scenario, f"{phase}.acquire",
|
| 2781 | [vcs, "worktree", "add", str(workspace), "-b", branch], base, run_index, meta, admitted,
|
| 2782 | )
|
| 2783 | rows.append(acquire)
|
| 2784 | if int(acquire["returncode"]) != 0:
|
| 2785 | return rows
|
| 2786 | rows.append(edit_step(subject, scenario, f"{phase}.edit", workspace, payload, run_index, meta, admitted))
|
| 2787 | # Agent-default snapshot: git pays stage+commit as two calls.
|
| 2788 | rows.append(step_row(subject, scenario, f"{phase}.snapshot.add", [vcs, "add", "."], workspace, run_index, meta, admitted))
|
| 2789 | rows.append(
|
| 2790 | step_row(
|
| 2791 | subject, scenario, f"{phase}.snapshot.commit",
|
| 2792 | [vcs, "commit", "-q", "-m", f"bench task loop {metadata['bench_id']} r{run_index} {phase}"],
|
| 2793 | workspace, run_index, meta, admitted,
|
| 2794 | )
|
| 2795 | )
|
| 2796 | rows.append(
|
| 2797 | step_row(
|
| 2798 | subject, scenario, f"{phase}.publish",
|
| 2799 | [vcs, "push", "-q", "-u", "origin", branch], workspace, run_index, meta, admitted,
|
| 2800 | )
|
| 2801 | )
|
| 2802 | # No task.describe: git's description is the commit message already
|
| 2803 | # paid in {phase}.snapshot.commit (asymmetry documented in the lane doc).
|
| 2804 | rows.append(
|
| 2805 | step_row(
|
| 2806 | subject, scenario, f"{phase}.end",
|
| 2807 | [vcs, "worktree", "remove", str(workspace)], base, run_index, meta, admitted,
|
| 2808 | )
|
| 2809 | )
|
| 2810 |
|
| 2811 | # Untimed oracle: payloads visible on the remote branches.
|
| 2812 | oracle_ok: bool | None = True
|
| 2813 | for phase in PHASES:
|
| 2814 | fetch = run_untimed([vcs, "fetch", "-q", "origin", branches[phase]], base)
|
| 2815 | show = run_untimed([vcs, "show", f"FETCH_HEAD:task-notes.md"], base)
|
| 2816 | if fetch.returncode != 0 or payloads[phase] not in show.stdout:
|
| 2817 | oracle_ok = False
|
| 2818 | rows.append(
|
| 2819 | {
|
| 2820 | **phase_total_row(rows, subject, scenario, run_index, meta, "task_loop.total"),
|
| 2821 | "oracle_payload_on_remote_branch": oracle_ok,
|
| 2822 | "workspace_branches": branches,
|
| 2823 | "measurement_source": "fetch_plus_show_grep",
|
| 2824 | }
|
| 2825 | )
|
| 2826 | return rows
|
| 2827 |
|
| 2828 |
|
| 2829 | def loop_summary_fields(rows: list[dict[str, Any]]) -> dict[str, Any]:
|
| 2830 | """followup-vs-task warm deltas computed from this run's step rows."""
|
| 2831 | def op_ms(operation: str) -> float | None:
|
| 2832 | for row in rows:
|
| 2833 | if row.get("operation") == operation:
|
| 2834 | return float(row.get("elapsed_ms", 0.0))
|
| 2835 | return None
|
| 2836 |
|
| 2837 | def phase_ms(prefix: str) -> float:
|
| 2838 | return sum(
|
| 2839 | float(row.get("elapsed_ms", 0.0))
|
| 2840 | for row in rows
|
| 2841 | if str(row.get("operation", "")).startswith(prefix + ".")
|
| 2842 | )
|
| 2843 |
|
| 2844 | task_acquire = op_ms("task.acquire")
|
| 2845 | followup_acquire = op_ms("followup.acquire")
|
| 2846 | fields: dict[str, Any] = {
|
| 2847 | "task_phase_ms": round(phase_ms("task"), 3),
|
| 2848 | "followup_phase_ms": round(phase_ms("followup"), 3),
|
| 2849 | }
|
| 2850 | if task_acquire is not None and followup_acquire is not None:
|
| 2851 | fields["followup_warm_delta_ms"] = round(followup_acquire - task_acquire, 3)
|
| 2852 | fields["acquire_ms_task"] = round(task_acquire, 3)
|
| 2853 | fields["acquire_ms_followup"] = round(followup_acquire, 3)
|
| 2854 | else:
|
| 2855 | fields["followup_warm_delta_ms"] = None
|
| 2856 | return fields
|
| 2857 |
|
| 2858 |
|
| 2859 | def summary_text(rows: list[dict[str, Any]]) -> str:
|
| 2860 | totals = [row for row in rows if row.get("operation") == "task_loop.total"]
|
| 2861 | lines = [
|
| 2862 | "# Task Loop Summary",
|
| 2863 | "",
|
| 2864 | "Whole-task cost of the signature motion: acquire workspace -> edit -> snapshot ->",
|
| 2865 | "publish -> describe -> end, then a follow-up task in a fresh workspace.",
|
| 2866 | "Command sequences are NOT step-equivalent across subjects",
|
| 2867 | "(oak pays desc; git's description is its commit message). Latency across unlike",
|
| 2868 | "transports is never comparable; scripted call equivalents and estimated token proxies are.",
|
| 2869 | "",
|
| 2870 | "| Subject | Run | Transport | Wall ms | Scripted call equivalents | Cost-wt tokens (env) | Warm acquire delta ms | Oracle |",
|
| 2871 | "| --- | ---: | --- | ---: | ---: | ---: | ---: | --- |",
|
| 2872 | ]
|
| 2873 | for row in totals:
|
| 2874 | oracle = row.get("oracle_payload_on_remote_branch")
|
| 2875 | lines.append(
|
| 2876 | "| `{}` | {} | {} | {} | {} | {} | {} | {} |".format(
|
| 2877 | row["subject"],
|
| 2878 | row["run"],
|
| 2879 | row.get("remote_transport", "unmeasured"),
|
| 2880 | fmt_ms(row.get("elapsed_ms")),
|
| 2881 | row.get("tool_call_count", "unmeasured"),
|
| 2882 | row.get("estimated_cost_weighted_tokens_with_envelope", "unmeasured"),
|
| 2883 | row.get("followup_warm_delta_ms", "unmeasured"),
|
| 2884 | "pass" if oracle else ("FAIL" if oracle is False else "unmeasured"),
|
| 2885 | )
|
| 2886 | )
|
| 2887 | skips = [
|
| 2888 | row for row in rows
|
| 2889 | if row.get("skipped") and row.get("scenario") == TASK_LOOP_SCENARIO
|
| 2890 | ]
|
| 2891 | for row in skips:
|
| 2892 | lines.append(f"| `{row['subject']}` | {row['run']} | skipped | | | | | {row['skip_reason'][:70]} |")
|
| 2893 |
|
| 2894 | agent_totals = [
|
| 2895 | row
|
| 2896 | for row in rows
|
| 2897 | if row.get("operation") in ("agent_task.total", "agent_task_oak_local.total")
|
| 2898 | ]
|
| 2899 | if agent_totals:
|
| 2900 | lines.extend(
|
| 2901 | [
|
| 2902 | "",
|
| 2903 | "## Cold clone and warm isolated task",
|
| 2904 | "",
|
| 2905 | "Local-file Git rows compare a fresh full clone with a worktree sharing",
|
| 2906 | "the first clone's object store. Oak local rows use two independent fresh",
|
| 2907 | "clones against an owned loopback server; they do not claim a shared cache.",
|
| 2908 | "OS cache and network bytes remain unknown. Hosted CI/review are skips.",
|
| 2909 | "",
|
| 2910 | "| Subject | Run | Wall ms | Scripted call equivalents | Output bytes | Coverage | Oracle |",
|
| 2911 | "| --- | ---: | ---: | ---: | ---: | --- | --- |",
|
| 2912 | ]
|
| 2913 | )
|
| 2914 | for row in agent_totals:
|
| 2915 | lines.append(
|
| 2916 | "| `{}` | {} | {} | {} | {} | {} | {} |".format(
|
| 2917 | row["subject"],
|
| 2918 | row["run"],
|
| 2919 | fmt_ms(row.get("elapsed_ms")),
|
| 2920 | row.get("tool_call_count", "unmeasured"),
|
| 2921 | row.get("raw_output_bytes", "unmeasured"),
|
| 2922 | "complete" if row.get("coverage_complete") else "partial",
|
| 2923 | "pass" if row.get("oracle_payload_on_remote_branch") else "FAIL",
|
| 2924 | )
|
| 2925 | )
|
| 2926 |
|
| 2927 | materialization_totals = [
|
| 2928 | row
|
| 2929 | for row in rows
|
| 2930 | if row.get("operation") == "legacy_selected_clone_materialization.total"
|
| 2931 | ]
|
| 2932 | if materialization_totals:
|
| 2933 | lines.extend(
|
| 2934 | [
|
| 2935 | "",
|
| 2936 | "## Legacy selected-branch materialization",
|
| 2937 | "",
|
| 2938 | "One fixed loopback Serve and one immutable fixture are shared across subjects.",
|
| 2939 | "Fresh destinations do not imply a cold OS page cache. Timing claims require",
|
| 2940 | "separate interleaved A/A noise and A/B campaigns.",
|
| 2941 | "",
|
| 2942 | "| Subject | Run | Clone ms | First status ms | First no-op commit ms | Passes | Harness controls met | Oracle |",
|
| 2943 | "| --- | ---: | ---: | ---: | ---: | ---: | --- | --- |",
|
| 2944 | ]
|
| 2945 | )
|
| 2946 | for row in materialization_totals:
|
| 2947 | lines.append(
|
| 2948 | "| `{}` | {} | {} | {} | {} | {} | {} | {} |".format(
|
| 2949 | row["subject"],
|
| 2950 | row["run"],
|
| 2951 | fmt_ms(row.get("clone_elapsed_ms")),
|
| 2952 | fmt_ms(row.get("first_status_elapsed_ms")),
|
| 2953 | fmt_ms(row.get("first_noop_commit_elapsed_ms")),
|
| 2954 | row.get("materialization_passes_observed", "unmeasured"),
|
| 2955 | "yes" if row.get("materialization_timing_controls_met") else "no",
|
| 2956 | "pass" if row.get("oracle_passed") else "FAIL",
|
| 2957 | )
|
| 2958 | )
|
| 2959 |
|
| 2960 | interrupted_totals = [
|
| 2961 | row for row in rows if row.get("operation") == "interrupted_publication.total"
|
| 2962 | ]
|
| 2963 | if interrupted_totals:
|
| 2964 | lines.extend(
|
| 2965 | [
|
| 2966 | "",
|
| 2967 | "## Interrupted publication",
|
| 2968 | "",
|
| 2969 | "A loopback fixture validates the request framing and drops the response.",
|
| 2970 | "Fixture delivery knowledge is reported separately from the CLI's own outcome",
|
| 2971 | "claim. Request receipt is not durability proof; the harness never retries.",
|
| 2972 | "",
|
| 2973 | "| Subject | Run | Wall ms | Scripted call equivalents | Output bytes | CLI outcome | Fixture delivery | POSTs | Request bytes | Head/work preserved |",
|
| 2974 | "| --- | ---: | ---: | ---: | ---: | --- | --- | ---: | ---: | --- |",
|
| 2975 | ]
|
| 2976 | )
|
| 2977 | for row in interrupted_totals:
|
| 2978 | lines.append(
|
| 2979 | "| `{}` | {} | {} | {} | {} | `{}` | `{}` | {} | {} | {}/{} |".format(
|
| 2980 | row["subject"],
|
| 2981 | row["run"],
|
| 2982 | fmt_ms(row.get("elapsed_ms")),
|
| 2983 | row.get("tool_call_count", "unmeasured"),
|
| 2984 | row.get("raw_output_bytes", "unmeasured"),
|
| 2985 | row.get("publication_outcome", "unmeasured"),
|
| 2986 | row.get("fixture_delivery_state", "unmeasured"),
|
| 2987 | row.get("mutation_requests_observed", "unmeasured"),
|
| 2988 | row.get("network_request_bytes_observed", "unmeasured"),
|
| 2989 | "yes" if row.get("local_head_preserved") else "no",
|
| 2990 | "yes" if row.get("local_work_preserved") else "no",
|
| 2991 | )
|
| 2992 | )
|
| 2993 | return "\n".join(lines) + "\n"
|
| 2994 |
|
| 2995 |
|
| 2996 | def fmt_ms(value: Any) -> str:
|
| 2997 | try:
|
| 2998 | return f"{float(value):.0f}"
|
| 2999 | except (TypeError, ValueError):
|
| 3000 | return "unmeasured"
|
| 3001 |
|
| 3002 |
|
| 3003 | def main() -> int:
|
| 3004 | args = parse_args()
|
| 3005 | scenarios = [item.strip() for item in args.scenarios.split(",") if item.strip()]
|
| 3006 | known_scenarios = {
|
| 3007 | TASK_LOOP_SCENARIO,
|
| 3008 | AGENT_TASK_SCENARIO,
|
| 3009 | AGENT_TASK_OAK_LOCAL_SCENARIO,
|
| 3010 | LEGACY_SELECTED_MATERIALIZATION_SCENARIO,
|
| 3011 | SHALLOW_REFRESH_SCENARIO,
|
| 3012 | INTERRUPTED_PUBLICATION_SCENARIO,
|
| 3013 | }
|
| 3014 | unknown_scenarios = sorted(set(scenarios) - known_scenarios)
|
| 3015 | if unknown_scenarios:
|
| 3016 | raise SystemExit(
|
| 3017 | "Unknown scenarios: " + ", ".join(unknown_scenarios)
|
| 3018 | + "; valid: " + ", ".join(sorted(known_scenarios))
|
| 3019 | )
|
| 3020 | subjects = load_subjects(args)
|
| 3021 | timestamp = datetime.now(timezone.utc).strftime("%Y%m%dT%H%M%SZ")
|
| 3022 | args.results.mkdir(parents=True, exist_ok=True)
|
| 3023 | args.workdir.mkdir(parents=True, exist_ok=True)
|
| 3024 | run_root = args.workdir / "runs" / timestamp
|
| 3025 | run_root.mkdir(parents=True, exist_ok=True)
|
| 3026 |
|
| 3027 | metadata = {
|
| 3028 | "bench_id": timestamp,
|
| 3029 | "profile": "task-loop",
|
| 3030 | "benchmark_track": "agent-default",
|
| 3031 | "timestamp_utc": timestamp,
|
| 3032 | "host": platform.node(),
|
| 3033 | "platform": platform.platform(),
|
| 3034 | "machine": platform.machine(),
|
| 3035 | "python": platform.python_version(),
|
| 3036 | "env_isolation_version": ENV_ISOLATION_VERSION,
|
| 3037 | "subject_versions": subject_versions(subjects),
|
| 3038 | "subject_details": subject_details(subjects),
|
| 3039 | "source": source_metadata(args.oak_repo),
|
| 3040 | "git_remote_mode": args.git_remote,
|
| 3041 | "trial_order_randomized": args.randomize_subject_order,
|
| 3042 | "trial_order_seed": args.random_seed if args.randomize_subject_order else None,
|
| 3043 | }
|
| 3044 |
|
| 3045 | rows: list[dict[str, Any]] = []
|
| 3046 | with measurement_lock("task_loop") as lock_info:
|
| 3047 | metadata["measurement_lock_wait_ms"] = lock_info.wait_ms
|
| 3048 | metadata["measurement_lock"] = "held" if lock_info.enabled else "disabled"
|
| 3049 | order_rng = random.Random(args.random_seed)
|
| 3050 | if LEGACY_SELECTED_MATERIALIZATION_SCENARIO in scenarios:
|
| 3051 | rows.extend(
|
| 3052 | run_legacy_selected_materialization_campaign(
|
| 3053 | subjects,
|
| 3054 | args.oak_serve_bin,
|
| 3055 | args.oak_serve_source,
|
| 3056 | args.oak_serve_source_head,
|
| 3057 | args.materialization_baseline_subject,
|
| 3058 | run_root,
|
| 3059 | args.runs,
|
| 3060 | metadata,
|
| 3061 | args.admitted_output_chars,
|
| 3062 | args.randomize_subject_order,
|
| 3063 | args.random_seed,
|
| 3064 | args.materialization_timing_controls_met,
|
| 3065 | )
|
| 3066 | )
|
| 3067 | if SHALLOW_REFRESH_SCENARIO in scenarios:
|
| 3068 | rows.extend(
|
| 3069 | run_shallow_refresh_campaign(
|
| 3070 | subjects,
|
| 3071 | args.oak_serve_bin,
|
| 3072 | args.oak_serve_source_head,
|
| 3073 | parse_shallow_refresh_fixtures(args.shallow_refresh_fixtures),
|
| 3074 | run_root,
|
| 3075 | args.runs,
|
| 3076 | metadata,
|
| 3077 | args.admitted_output_chars,
|
| 3078 | args.randomize_subject_order,
|
| 3079 | args.random_seed,
|
| 3080 | args.injected_request_delay_ms,
|
| 3081 | )
|
| 3082 | )
|
| 3083 | campaign_scenarios = {LEGACY_SELECTED_MATERIALIZATION_SCENARIO, SHALLOW_REFRESH_SCENARIO}
|
| 3084 | ordinary_scenarios = [scenario for scenario in scenarios if scenario not in campaign_scenarios]
|
| 3085 | for run_index in range(args.runs):
|
| 3086 | run_subjects = list(subjects)
|
| 3087 | if args.randomize_subject_order:
|
| 3088 | order_rng.shuffle(run_subjects)
|
| 3089 | run_metadata = {
|
| 3090 | **metadata,
|
| 3091 | "run_subject_order": [subject.name for subject in run_subjects],
|
| 3092 | }
|
| 3093 | for subject in run_subjects:
|
| 3094 | for scenario in ordinary_scenarios:
|
| 3095 | print(f"[run] {scenario} run={run_index} subject={subject.name}", flush=True)
|
| 3096 | if scenario == INTERRUPTED_PUBLICATION_SCENARIO:
|
| 3097 | rows.extend(
|
| 3098 | run_interrupted_publication(
|
| 3099 | subject, run_root, run_index, run_metadata, args.admitted_output_chars
|
| 3100 | )
|
| 3101 | )
|
| 3102 | continue
|
| 3103 | if scenario == AGENT_TASK_SCENARIO:
|
| 3104 | if subject.kind != "git":
|
| 3105 | rows.append(
|
| 3106 | unmeasured_step_row(
|
| 3107 | subject,
|
| 3108 | scenario,
|
| 3109 | "agent_task.skipped",
|
| 3110 | run_index,
|
| 3111 | run_metadata,
|
| 3112 | "Oak end-to-end publication and exact CI require an explicitly configured disposable staging lane; production mutation is forbidden",
|
| 3113 | )
|
| 3114 | )
|
| 3115 | continue
|
| 3116 | remote_create_start = time.perf_counter()
|
| 3117 | remote = oakbench_remotes.make_git_bare_remote(
|
| 3118 | subject.bin, run_root, f"agent-task-remote-r{run_index}.git"
|
| 3119 | )
|
| 3120 | remote_create_ms = (time.perf_counter() - remote_create_start) * 1000
|
| 3121 | rows.extend(
|
| 3122 | run_git_agent_task_baseline(
|
| 3123 | subject,
|
| 3124 | remote,
|
| 3125 | run_root,
|
| 3126 | run_index,
|
| 3127 | run_metadata,
|
| 3128 | args.admitted_output_chars,
|
| 3129 | remote_create_ms,
|
| 3130 | )
|
| 3131 | )
|
| 3132 | continue
|
| 3133 | if scenario == AGENT_TASK_OAK_LOCAL_SCENARIO:
|
| 3134 | if subject.kind != "oak":
|
| 3135 | rows.append(
|
| 3136 | unmeasured_step_row(
|
| 3137 | subject,
|
| 3138 | scenario,
|
| 3139 | "agent_task_oak_local.skipped",
|
| 3140 | run_index,
|
| 3141 | run_metadata,
|
| 3142 | "owned local oak serve scenario applies only to Oak subjects",
|
| 3143 | )
|
| 3144 | )
|
| 3145 | continue
|
| 3146 | rows.extend(
|
| 3147 | run_oak_local_agent_task_baseline(
|
| 3148 | subject,
|
| 3149 | run_root,
|
| 3150 | run_index,
|
| 3151 | run_metadata,
|
| 3152 | args.admitted_output_chars,
|
| 3153 | )
|
| 3154 | )
|
| 3155 | continue
|
| 3156 | if subject.kind == "git":
|
| 3157 | if args.git_remote == "github":
|
| 3158 | remote = oakbench_remotes.resolve_git_github_remote()
|
| 3159 | else:
|
| 3160 | remote = oakbench_remotes.make_git_bare_remote(
|
| 3161 | subject.bin, run_root, f"task-loop-remote-r{run_index}.git"
|
| 3162 | )
|
| 3163 | if not remote.resolved:
|
| 3164 | rows.append(skip_row(subject, TASK_LOOP_SCENARIO, run_index, run_metadata, remote.skip_reason or ""))
|
| 3165 | continue
|
| 3166 | subject_rows = run_git_loop(subject, remote, run_root, run_index, run_metadata, args.admitted_output_chars)
|
| 3167 | else:
|
| 3168 | remote = oakbench_remotes.resolve_oak_remote("sync")
|
| 3169 | if not remote.resolved:
|
| 3170 | rows.append(skip_row(subject, TASK_LOOP_SCENARIO, run_index, run_metadata, remote.skip_reason or ""))
|
| 3171 | continue
|
| 3172 | subject_rows = run_oak_loop(subject, remote, run_root, run_index, run_metadata, args.admitted_output_chars)
|
| 3173 | total = next((row for row in subject_rows if row.get("operation") == "task_loop.total"), None)
|
| 3174 | if total is not None:
|
| 3175 | total.update(loop_summary_fields(subject_rows))
|
| 3176 | rows.extend(subject_rows)
|
| 3177 |
|
| 3178 | store = ResultsStore(args.results, lane="task-loop")
|
| 3179 | raw_path, summary_path = store.write(timestamp, rows, summary_text(rows))
|
| 3180 | if not args.keep_workdirs:
|
| 3181 | shutil.rmtree(run_root, ignore_errors=True)
|
| 3182 | print(f"[result] {raw_path}")
|
| 3183 | print(f"[summary] {summary_path}")
|
| 3184 | return 1 if has_lane_failures(rows) else 0
|
| 3185 |
|
| 3186 |
|
| 3187 | if __name__ == "__main__":
|
| 3188 | raise SystemExit(main())
|