| 1 | #!/usr/bin/env python3
|
| 2 | """P4 robustness suite runner: crash points, fault injection, concurrency, env faults.
|
| 3 |
|
| 4 | Writes robustness-lane JSONL rows (one per case) through oakbench.ResultsStore
|
| 5 | into ``--results`` (default results/robustness, never committed, ADR-0003),
|
| 6 | flushing each row to a .partial.jsonl as it completes, and prints a Markdown
|
| 7 | classification table. Every case is classified held / violated / unsupported /
|
| 8 | harness_error (see README.md). Loopback servers and local fixtures only.
|
| 9 |
|
| 10 | Examples:
|
| 11 | OAK_BENCH_OAK_READER_PYTHON=/path/to/py python3 suites/robustness/run_robustness.py --quick
|
| 12 | python3 suites/robustness/run_robustness.py --subjects oak,git --families crash_time,proxy
|
| 13 | """
|
| 14 |
|
| 15 | from __future__ import annotations
|
| 16 |
|
| 17 | import argparse
|
| 18 | import json
|
| 19 | import platform
|
| 20 | import shutil
|
| 21 | import sys
|
| 22 | import tempfile
|
| 23 | import time
|
| 24 | from datetime import datetime, timezone
|
| 25 | from pathlib import Path
|
| 26 | from typing import Any, Callable, Optional
|
| 27 |
|
| 28 | sys.path.insert(0, str(Path(__file__).resolve().parent))
|
| 29 |
|
| 30 | from robustness import (CLASSIFICATIONS, FAULT_NOT_FIRED, HARNESS_ERROR, HELD, SUITE_VERSION, UNSUPPORTED, # noqa: E402
|
| 31 | VIOLATED, known_bugs)
|
| 32 | from robustness import crashpoints # noqa: E402
|
| 33 | from robustness.crash import (TIME_FRACTIONS, calibrate, run_time_point, run_trial, syscall_fault, # noqa: E402
|
| 34 | syscall_points)
|
| 35 | from robustness.fixtures import FIXTURE_VERSION, build_git, build_oak # noqa: E402
|
| 36 | from robustness.procs import free_port # noqa: E402
|
| 37 |
|
| 38 | from oakbench.results import ResultsStore # noqa: E402
|
| 39 | from oakbench.rows import SKIP_RETURNCODE # noqa: E402
|
| 40 | from oakbench.subjects import binary_sha256, which_measured_git # noqa: E402
|
| 41 |
|
| 42 | LANE = "robustness"
|
| 43 | FAMILIES = ("crash_time", "crash_syscall", "proxy", "concurrency", "env", "hosted")
|
| 44 | CRASH_OPS = ("commit", "push", "pull_ff", "pull_diverged", "clone", "switch", "merge_local", "merge_hosted",
|
| 45 | "mount_finish")
|
| 46 | PROXY_OPS = ("push", "pull_ff", "pull_diverged", "clone")
|
| 47 | SCENARIO = {
|
| 48 | "crash_time": "crash_time_v1",
|
| 49 | "crash_syscall": "crash_syscall_v1",
|
| 50 | "proxy": "fault_proxy_v1",
|
| 51 | }
|
| 52 | RETURNCODE = {HELD: 0, VIOLATED: 1, HARNESS_ERROR: 2, UNSUPPORTED: SKIP_RETURNCODE, FAULT_NOT_FIRED: SKIP_RETURNCODE}
|
| 53 | MEASUREMENT_SOURCE = "independent_reader_invariants_v1"
|
| 54 |
|
| 55 | UNSUPPORTED_OPS = {
|
| 56 | ("oak", "merge_local"): "oak has no local branch-to-branch merge; `oak merge` is the hosted squash-merge (see merge_hosted)",
|
| 57 | ("oak", "merge_hosted"): "loopback oak serve 0.105.0 answers merge publication with 404; hosted merge needs a local oakspace (hosted_merge_invariants_v1)",
|
| 58 | ("git", "merge_hosted"): "git has no hosted merge; the git differential is merge_local",
|
| 59 | ("oak", "mount_finish"): "oak mount needs macOS FSKit (signed Oak Mount app) or Linux FUSE; not exercised against loopback serve in suite v1",
|
| 60 | ("git", "mount_finish"): "git has no mount",
|
| 61 | }
|
| 62 |
|
| 63 |
|
| 64 | class Runner:
|
| 65 | def __init__(self, args: argparse.Namespace):
|
| 66 | self.args = args
|
| 67 | self.bench_id = args.bench_id or datetime.now(timezone.utc).strftime("%Y%m%dT%H%M%SZ")
|
| 68 | self.work = Path(args.workdir or tempfile.mkdtemp(prefix="p4-robustness-")).resolve()
|
| 69 | self.work.mkdir(parents=True, exist_ok=True)
|
| 70 | self.results = Path(args.results).resolve()
|
| 71 | self.evidence = self.results / f"{self.bench_id}.robustness-evidence"
|
| 72 | self.store = ResultsStore(self.results, lane=LANE, filename_suffix="robustness", flush_partial=True)
|
| 73 | self.rows: list[dict] = []
|
| 74 | self.run_index: dict[tuple, int] = {}
|
| 75 | self.retained = 0
|
| 76 | self.bins = {"oak": args.oak_bin, "git": args.git_bin}
|
| 77 | self.instrument = None
|
| 78 | self.instrument_reason = ""
|
| 79 |
|
| 80 | # rows ----------------------------------------------------------------------
|
| 81 | def emit(self, subject_kind: str, scenario: str, operation: str, family: str, data: dict) -> dict:
|
| 82 | key = (subject_kind, scenario, operation)
|
| 83 | run = self.run_index.get(key, 0)
|
| 84 | self.run_index[key] = run + 1
|
| 85 | classification = data.get("classification", UNSUPPORTED if "unsupported" in data else HARNESS_ERROR)
|
| 86 | binary = self.bins.get(subject_kind)
|
| 87 | row: dict[str, Any] = {
|
| 88 | "bench_id": self.bench_id,
|
| 89 | "suite_version": SUITE_VERSION,
|
| 90 | "scenario": scenario,
|
| 91 | "operation": operation,
|
| 92 | "run": run,
|
| 93 | "subject": "oak_installed" if subject_kind == "oak" else "git",
|
| 94 | "subject_kind": subject_kind,
|
| 95 | "subject_bin": binary,
|
| 96 | "binary_sha256": binary_sha256(Path(binary)) if binary and Path(binary).exists() else None,
|
| 97 | "subject_version": self.versions.get(subject_kind),
|
| 98 | "fault_family": family,
|
| 99 | "fault_point": data.get("fault_point", "n/a"),
|
| 100 | "fault_fired": data.get("fault_fired", data.get("observations", {}).get("fault_fired")),
|
| 101 | "crash_model": "process_kill" if family.startswith("crash") or family == "proxy" else None,
|
| 102 | "classification": classification,
|
| 103 | "returncode": RETURNCODE[classification],
|
| 104 | "elapsed_ms": float(data.get("elapsed_ms", 0.0) or 0.0),
|
| 105 | "invariants": data.get("invariants", {}),
|
| 106 | "violations": data.get("violations", sorted(k for k, v in data.get("invariants", {}).items() if v is False)),
|
| 107 | "observations": data.get("observations", {}),
|
| 108 | "command": data.get("command", []),
|
| 109 | "fixture_version": FIXTURE_VERSION,
|
| 110 | "measurement_source": MEASUREMENT_SOURCE,
|
| 111 | "host_platform": platform.platform(),
|
| 112 | }
|
| 113 | if classification in (UNSUPPORTED, FAULT_NOT_FIRED):
|
| 114 | # Skip rows: a coverage gap (unsupported) or a clean run whose fault never
|
| 115 | # took effect (fault_not_fired). Neither is a measured result.
|
| 116 | row.update(skipped=True, skip_reason=data.get("unsupported") or data.get("skip_reason") or classification)
|
| 117 | row.pop("measurement_source")
|
| 118 | bugs = known_bugs(subject_kind, scenario, operation, row["violations"]) if classification == VIOLATED else []
|
| 119 | if bugs:
|
| 120 | row["known_bugs"] = bugs
|
| 121 | if data.get("free_bytes_measured") is not None:
|
| 122 | row["free_bytes_measured"] = data["free_bytes_measured"]
|
| 123 | if data.get("harness_error"):
|
| 124 | row["harness_error"] = data["harness_error"]
|
| 125 | transport = data.get("remote_transport")
|
| 126 | if transport:
|
| 127 | row.update(remote_transport=transport, remote_server=data.get("remote_server"),
|
| 128 | remote_repo=data.get("remote_repo"))
|
| 129 | if self.instrument is not None and family == "crash_syscall":
|
| 130 | row["crash_instrument"] = self.instrument.provenance
|
| 131 | self.rows.append(row)
|
| 132 | self.store.append_partial(self.bench_id, [row])
|
| 133 | mark = {HELD: ".", VIOLATED: "V", HARNESS_ERROR: "E", UNSUPPORTED: "u", FAULT_NOT_FIRED: "n"}[classification]
|
| 134 | print(mark, end="", flush=True)
|
| 135 | if classification in (VIOLATED, HARNESS_ERROR):
|
| 136 | print(f"\n [{classification}] {subject_kind} {scenario} {operation} {row['fault_point']} "
|
| 137 | f"{row['violations']} {row.get('harness_error', '')}", flush=True)
|
| 138 | return row
|
| 139 |
|
| 140 | def keep(self, trial, classification: str) -> None:
|
| 141 | trial.stop_server()
|
| 142 | if classification in (VIOLATED, HARNESS_ERROR) and self.retained < self.args.max_evidence:
|
| 143 | self.retained += 1
|
| 144 | dest = self.evidence / f"{self.retained:03d}-{trial.dir.name}"
|
| 145 | dest.parent.mkdir(parents=True, exist_ok=True)
|
| 146 | shutil.move(str(trial.dir), str(dest))
|
| 147 | else:
|
| 148 | shutil.rmtree(trial.dir, ignore_errors=True)
|
| 149 |
|
| 150 | def transport(self, kind: str, family: str) -> dict:
|
| 151 | if kind == "oak":
|
| 152 | return {"remote_transport": "network", "remote_server": "loopback-oak-serve", "remote_repo": "p4/robust"}
|
| 153 | if family == "proxy":
|
| 154 | return {"remote_transport": "network", "remote_server": "loopback-git-http-backend",
|
| 155 | "remote_repo": "remote.git"}
|
| 156 | return {"remote_transport": "local_file", "remote_server": "local", "remote_repo": "../srv/remote.git"}
|
| 157 |
|
| 158 | # families --------------------------------------------------------------------
|
| 159 | def fixture(self, kind: str, op: str, tag: str):
|
| 160 | root = self.work / "fixtures" / f"{tag}-{kind}-{op}"
|
| 161 | if kind == "oak":
|
| 162 | return build_oak(self.bins["oak"], op, root, free_port())
|
| 163 | return build_git(self.bins["git"], op, root, free_port())
|
| 164 |
|
| 165 | def crash_family(self, kind: str, families: list[str]) -> None:
|
| 166 | ops = [op for op in self.args.ops if op in CRASH_OPS]
|
| 167 | for op in ops:
|
| 168 | if (kind, op) in UNSUPPORTED_OPS:
|
| 169 | for fam in families:
|
| 170 | self.emit(kind, SCENARIO[fam], op, fam, {"unsupported": UNSUPPORTED_OPS[(kind, op)]})
|
| 171 | continue
|
| 172 | try:
|
| 173 | fx = self.fixture(kind, op, "crash")
|
| 174 | calib = calibrate(fx, self.bins[kind], self.work / "trials", self.instrument)
|
| 175 | except Exception as exc: # noqa: BLE001
|
| 176 | for fam in families:
|
| 177 | self.emit(kind, SCENARIO[fam], op, fam, {"classification": HARNESS_ERROR,
|
| 178 | "harness_error": f"fixture/calibration: {exc}"})
|
| 179 | continue
|
| 180 | if not calib.ok:
|
| 181 | for fam in families:
|
| 182 | self.emit(kind, SCENARIO[fam], op, fam, {"classification": HARNESS_ERROR,
|
| 183 | "harness_error": "calibration: " + calib.detail})
|
| 184 | continue
|
| 185 | extra = self.transport(kind, "crash")
|
| 186 | if "crash_time" in families:
|
| 187 | fractions = TIME_FRACTIONS if not self.args.quick else (0.1, 0.5, 0.9)
|
| 188 | for frac in fractions:
|
| 189 | row = run_time_point(fx, self.bins[kind], self.work / "trials", f"t-{kind}-{op}-{frac}", calib,
|
| 190 | frac, self.keep)
|
| 191 | self.emit(kind, "crash_time_v1", op, "crash_time", {**row, **extra})
|
| 192 | if "crash_syscall" in families:
|
| 193 | reason = crashpoints.unavailable_reason(self.instrument, self.instrument_reason, kind)
|
| 194 | if reason:
|
| 195 | self.emit(kind, "crash_syscall_v1", op, "crash_syscall", {"unsupported": reason})
|
| 196 | elif calib.boundaries == 0:
|
| 197 | self.emit(kind, "crash_syscall_v1", op, "crash_syscall", {
|
| 198 | "unsupported": crashpoints.NO_BOUNDARY_REASON[kind]})
|
| 199 | else:
|
| 200 | points = syscall_points(calib.boundaries)
|
| 201 | if self.args.quick:
|
| 202 | points = points[:: max(1, len(points) // 4)][:4]
|
| 203 | for n in points:
|
| 204 | row = run_trial(fx, self.bins[kind], self.work / "trials", f"s-{kind}-{op}-{n}", calib,
|
| 205 | syscall_fault(self.instrument, n, calib.boundary_kinds), self.keep)
|
| 206 | row.setdefault("observations", {})["boundaries_total"] = calib.boundaries
|
| 207 | self.emit(kind, "crash_syscall_v1", op, "crash_syscall", {**row, **extra})
|
| 208 | shutil.rmtree(fx.root, ignore_errors=True)
|
| 209 |
|
| 210 | def proxy_family(self, kind: str) -> None:
|
| 211 | from robustness.proxycases import PER_REQUEST, fault_points, plan, record, run_proxy_trial
|
| 212 |
|
| 213 | for op in [op for op in self.args.ops if op in PROXY_OPS]:
|
| 214 | try:
|
| 215 | fx = self.fixture(kind, op, "proxy")
|
| 216 | calib, keys, methods = record(fx, self.bins[kind], self.work / "trials")
|
| 217 | except Exception as exc: # noqa: BLE001
|
| 218 | self.emit(kind, "fault_proxy_v1", op, "proxy", {"classification": HARNESS_ERROR,
|
| 219 | "harness_error": f"record: {exc}"})
|
| 220 | continue
|
| 221 | if not calib.ok:
|
| 222 | self.emit(kind, "fault_proxy_v1", op, "proxy", {"classification": HARNESS_ERROR,
|
| 223 | "harness_error": "record: " + calib.detail})
|
| 224 | continue
|
| 225 | faults = plan(keys, methods)
|
| 226 | if self.args.quick:
|
| 227 | pts = fault_points(keys)
|
| 228 | keep_keys = {pts[0], pts[-1]} | {p for p in pts if p[0].startswith("POST")}
|
| 229 | faults = [f for f in faults if f.key is None or (f.key, f.occurrence) in keep_keys][:14]
|
| 230 | for index, fault in enumerate(faults):
|
| 231 | row = run_proxy_trial(fx, self.bins[kind], self.work / "trials", f"p-{kind}-{op}-{index}", calib,
|
| 232 | fault, self.keep)
|
| 233 | row.setdefault("observations", {})["recorded_requests"] = calib.requests[:60]
|
| 234 | self.emit(kind, "fault_proxy_v1", op, "proxy", {**row, **self.transport(kind, "proxy")})
|
| 235 | shutil.rmtree(fx.root, ignore_errors=True)
|
| 236 |
|
| 237 | def concurrency_family(self, kind: str) -> None:
|
| 238 | from robustness.concurrency import same_branch, shared_checkout
|
| 239 | from robustness.envfaults import FsyncLoad
|
| 240 |
|
| 241 | sizes = self.args.clients
|
| 242 | extra = self.transport(kind, "concurrency")
|
| 243 | for n in sizes:
|
| 244 | row = self._guard(lambda: same_branch(kind, self.bins[kind], self.work / f"conc-same-{kind}-{n}", n))
|
| 245 | self.emit(kind, "conc_same_branch_v1", "concurrent.push", "concurrency",
|
| 246 | {**row, "fault_point": f"clients={n}", **extra})
|
| 247 | row = self._guard(lambda: same_branch(kind, self.bins[kind], self.work / f"conc-dist-{kind}-{n}", n,
|
| 248 | distinct=True))
|
| 249 | self.emit(kind, "conc_distinct_branches_v1", "concurrent.push", "concurrency",
|
| 250 | {**row, "fault_point": f"clients={n}", **extra})
|
| 251 | n = sizes[0]
|
| 252 | with FsyncLoad(self.work / f"fsyncload-{kind}"):
|
| 253 | row = self._guard(lambda: same_branch(kind, self.bins[kind], self.work / f"conc-load-{kind}", n))
|
| 254 | self.emit(kind, "env_fsync_load_v1", "concurrent.push", "env",
|
| 255 | {**row, "fault_point": f"clients={n},fsync_writers=4", **extra})
|
| 256 | for what, scenario, op in (("commit", "lock_shared_commit_v1", "concurrent.commit"),
|
| 257 | ("push", "lock_shared_push_v1", "concurrent.push"),
|
| 258 | ("stale_owner", "lock_stale_owner_v1", "concurrent.commit")):
|
| 259 | row = self._guard(lambda: shared_checkout(kind, self.bins[kind], self.work / f"lock-{what}-{kind}", 8, what))
|
| 260 | self.emit(kind, scenario, op, "concurrency", {**row, "fault_point": "contenders=8",
|
| 261 | **(extra if what == "push" else {})})
|
| 262 | self.emit(kind, "lock_credentials_v1", "credentials.save", "concurrency", {
|
| 263 | "unsupported": ("no non-interactive credential writer in the CLI (oak login is interactive; OAK_API_KEY "
|
| 264 | "bypasses the store); credential-lock stress stays in oak unit tests (L8 credstress)")
|
| 265 | if kind == "oak" else "git has no credential store under test"})
|
| 266 |
|
| 267 | def env_family(self, kind: str) -> None:
|
| 268 | from robustness.envfaults import FREE_KIB, disk_full_trial, op_need_bytes, readonly_exercised, readonly_trial
|
| 269 |
|
| 270 | for op in [op for op in ("commit", "pull_ff", "clone") if op in self.args.ops]:
|
| 271 | try:
|
| 272 | fx = self.fixture(kind, op, "env")
|
| 273 | calib = calibrate(fx, self.bins[kind], self.work / "trials", None)
|
| 274 | need = op_need_bytes(fx, self.bins[kind], self.work / "trials")
|
| 275 | except Exception as exc: # noqa: BLE001
|
| 276 | self.emit(kind, "env_disk_full_v1", op, "env", {"classification": HARNESS_ERROR,
|
| 277 | "harness_error": f"fixture: {exc}"})
|
| 278 | continue
|
| 279 | extra = self.transport(kind, "env")
|
| 280 | for free in (FREE_KIB if not self.args.quick else (0,)):
|
| 281 | row = disk_full_trial(fx, self.bins[kind], self.work / "trials", f"df-{kind}-{op}-{free}", calib, free,
|
| 282 | need)
|
| 283 | self.emit(kind, "env_disk_full_v1", op, "env", {**row, **extra})
|
| 284 | if op != "clone":
|
| 285 | for target in ("store", "subdir"):
|
| 286 | if target == "subdir" and op == "commit":
|
| 287 | continue
|
| 288 | try:
|
| 289 | exercised, ex_detail = readonly_exercised(fx, self.bins[kind], self.work / "trials",
|
| 290 | f"ro-probe-{kind}-{op}-{target}", target)
|
| 291 | except Exception as exc: # noqa: BLE001
|
| 292 | exercised, ex_detail = None, {"probe_error": f"{type(exc).__name__}: {exc}"}
|
| 293 | row = readonly_trial(fx, self.bins[kind], self.work / "trials", f"ro-{kind}-{op}-{target}", calib,
|
| 294 | target, exercised, ex_detail)
|
| 295 | self.emit(kind, "env_readonly_v1", op, "env", {**row, **extra})
|
| 296 | shutil.rmtree(fx.root, ignore_errors=True)
|
| 297 |
|
| 298 | def hosted_family(self) -> None:
|
| 299 | from robustness.hosted import rows_spec
|
| 300 |
|
| 301 | for spec in rows_spec():
|
| 302 | self.emit("oak", "hosted_merge_invariants_v1", spec["operation"], "hosted",
|
| 303 | {"unsupported": spec["skip_reason"], "fault_point": spec["invariant_id"]})
|
| 304 |
|
| 305 | @staticmethod
|
| 306 | def _guard(fn: Callable[[], dict]) -> dict:
|
| 307 | try:
|
| 308 | return fn()
|
| 309 | except Exception as exc: # noqa: BLE001
|
| 310 | return {"classification": HARNESS_ERROR, "harness_error": f"{type(exc).__name__}: {exc}"}
|
| 311 |
|
| 312 | # main ------------------------------------------------------------------------
|
| 313 | def run(self) -> int:
|
| 314 | import subprocess
|
| 315 |
|
| 316 | self.versions = {}
|
| 317 | for kind in self.args.subjects:
|
| 318 | binary = self.bins.get(kind)
|
| 319 | if not binary or not Path(binary).exists():
|
| 320 | raise SystemExit(f"{kind} binary not found: {binary}")
|
| 321 | self.versions[kind] = subprocess.run([binary, "--version"], capture_output=True, text=True).stdout.strip()
|
| 322 | if "crash_syscall" in self.args.families:
|
| 323 | self.instrument, self.instrument_reason = crashpoints.build(
|
| 324 | self.work / "instrument", self.bins.get("oak") if "oak" in self.args.subjects else None,
|
| 325 | self.bins.get("git") if "git" in self.args.subjects else None)
|
| 326 | started = time.time()
|
| 327 | for kind in self.args.subjects:
|
| 328 | crash = [f for f in ("crash_time", "crash_syscall") if f in self.args.families]
|
| 329 | if crash:
|
| 330 | self.crash_family(kind, crash)
|
| 331 | if "proxy" in self.args.families:
|
| 332 | self.proxy_family(kind)
|
| 333 | if "concurrency" in self.args.families:
|
| 334 | self.concurrency_family(kind)
|
| 335 | if "env" in self.args.families:
|
| 336 | self.env_family(kind)
|
| 337 | if "hosted" in self.args.families:
|
| 338 | self.hosted_family()
|
| 339 | print()
|
| 340 | summary = summarize(self.rows, self.bench_id, time.time() - started)
|
| 341 | raw, summary_path = self.store.write(self.bench_id, self.rows, summary)
|
| 342 | print(summary)
|
| 343 | print(f"rows: {raw}\nsummary: {summary_path}")
|
| 344 | if not self.args.keep_workdir:
|
| 345 | shutil.rmtree(self.work / "trials", ignore_errors=True)
|
| 346 | return 1 if any(r["classification"] == VIOLATED for r in self.rows) and self.args.fail_on_violation else 0
|
| 347 |
|
| 348 |
|
| 349 | def summarize(rows: list[dict], bench_id: str, seconds: float) -> str:
|
| 350 | table: dict[tuple, dict[str, int]] = {}
|
| 351 | for row in rows:
|
| 352 | key = (row["subject_kind"], row["scenario"], row["operation"])
|
| 353 | cell = table.setdefault(key, {c: 0 for c in CLASSIFICATIONS})
|
| 354 | cell[row["classification"]] += 1
|
| 355 | lines = [f"# Robustness suite {SUITE_VERSION} — {bench_id}", "",
|
| 356 | f"{len(rows)} cases in {seconds / 60:.1f} min.", "",
|
| 357 | "| subject | scenario | operation | held | violated | fault_not_fired | unsupported | harness_error |",
|
| 358 | "|---|---|---|---:|---:|---:|---:|---:|"]
|
| 359 | for (subject, scenario, operation), cell in sorted(table.items()):
|
| 360 | lines.append(f"| {subject} | {scenario} | {operation} | {cell[HELD]} | {cell[VIOLATED]} | "
|
| 361 | f"{cell[FAULT_NOT_FIRED]} | {cell[UNSUPPORTED]} | {cell[HARNESS_ERROR]} |")
|
| 362 | totals = {c: sum(1 for r in rows if r["classification"] == c) for c in CLASSIFICATIONS}
|
| 363 | lines += ["", "Totals: " + ", ".join(f"{k}={v}" for k, v in totals.items()), ""]
|
| 364 | for subject in sorted({r["subject_kind"] for r in rows}):
|
| 365 | sub = {c: sum(1 for r in rows if r["subject_kind"] == subject and r["classification"] == c)
|
| 366 | for c in CLASSIFICATIONS}
|
| 367 | lines.append(f"- {subject}: " + ", ".join(f"{k}={v}" for k, v in sub.items()))
|
| 368 | lines.append("")
|
| 369 | violations: dict[tuple, list[str]] = {}
|
| 370 | for row in rows:
|
| 371 | if row["classification"] == VIOLATED:
|
| 372 | key = (row["subject_kind"], row["scenario"], row["operation"], ",".join(row["violations"]),
|
| 373 | ",".join(row.get("known_bugs") or []))
|
| 374 | violations.setdefault(key, []).append(row["fault_point"])
|
| 375 | if violations:
|
| 376 | lines += ["## Violations", ""]
|
| 377 | for (subject, scenario, operation, names, bugs), points in sorted(violations.items()):
|
| 378 | tag = f" [{bugs}]" if bugs else ""
|
| 379 | lines.append(f"- {subject} {scenario} {operation}: {names}{tag} ({len(points)}x; e.g. "
|
| 380 | f"{', '.join(points[:3])})")
|
| 381 | return "\n".join(lines) + "\n"
|
| 382 |
|
| 383 |
|
| 384 | def parse_args(argv: Optional[list[str]] = None) -> argparse.Namespace:
|
| 385 | parser = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
|
| 386 | parser.add_argument("--subjects", default="oak,git")
|
| 387 | parser.add_argument("--families", default=",".join(FAMILIES))
|
| 388 | parser.add_argument("--ops", default=",".join(CRASH_OPS))
|
| 389 | parser.add_argument("--clients", default="8,16", help="concurrency sizes (comma-separated)")
|
| 390 | parser.add_argument("--oak-bin", default=shutil.which("oak"))
|
| 391 | parser.add_argument("--git-bin", default=None, help="default: real git (xcrun shim resolved)")
|
| 392 | parser.add_argument("--results", default=str(Path(__file__).resolve().parents[2] / "results" / "robustness"))
|
| 393 | parser.add_argument("--workdir", default=None)
|
| 394 | parser.add_argument("--bench-id", default=None)
|
| 395 | parser.add_argument("--quick", action="store_true", help="few points per case (smoke)")
|
| 396 | parser.add_argument("--max-evidence", type=int, default=40, help="violated/harness_error trial dirs retained")
|
| 397 | parser.add_argument("--keep-workdir", action="store_true")
|
| 398 | parser.add_argument("--fail-on-violation", action="store_true")
|
| 399 | parser.add_argument("--measurement-lock", action="store_true",
|
| 400 | help="hold the benchmark measurement lock for the whole run (serialize with timing campaigns)")
|
| 401 | args = parser.parse_args(argv)
|
| 402 | args.subjects = [s for s in args.subjects.split(",") if s]
|
| 403 | args.families = [f for f in args.families.split(",") if f]
|
| 404 | args.ops = [o for o in args.ops.split(",") if o]
|
| 405 | args.clients = [int(c) for c in str(args.clients).split(",") if c]
|
| 406 | unknown = set(args.families) - set(FAMILIES)
|
| 407 | if unknown or not set(args.subjects) <= {"oak", "git"} or not set(args.ops) <= set(CRASH_OPS):
|
| 408 | parser.error(f"unknown family/subject/op: {sorted(unknown)} {args.subjects} {args.ops}")
|
| 409 | if args.git_bin is None:
|
| 410 | args.git_bin = which_measured_git()
|
| 411 | return args
|
| 412 |
|
| 413 |
|
| 414 | def main(argv: Optional[list[str]] = None) -> int:
|
| 415 | args = parse_args(argv)
|
| 416 | if not args.measurement_lock:
|
| 417 | # Robustness cases are not timing measurements, and holding the shared
|
| 418 | # lock for a long run would block every other lane (and their tests).
|
| 419 | # Do not run the full suite beside a timing campaign; pass
|
| 420 | # --measurement-lock to serialize behind one instead.
|
| 421 | return Runner(args).run()
|
| 422 | from oakbench.runlock import measurement_lock
|
| 423 |
|
| 424 | with measurement_lock("robustness-suite"):
|
| 425 | return Runner(args).run()
|
| 426 |
|
| 427 |
|
| 428 | if __name__ == "__main__":
|
| 429 | raise SystemExit(main())
|