test_coord_write_remote_atomic.py
python
sha256:660fcac1df3ab28f61862e961890bd2ca8b754fa0242079d93ca1e25037ec8a6
chore(tests): add docstring to tests/__init__.py so rc14 tr…
Human
26 days ago
| 1 | """ |
| 2 | Tests for the bug: _write_remote_records uses write_text() which truncates |
| 3 | the target file to 0 bytes before writing. Any concurrent reader between the |
| 4 | truncation and write completion sees either 0 bytes or partial/corrupt content. |
| 5 | |
| 6 | Root cause (coord_sync.py::_write_remote_records, last line): |
| 7 | |
| 8 | target.write_text(json.dumps(rec) + "\n", encoding="utf-8") |
| 9 | |
| 10 | write_text() opens with mode 'w', which truncates to 0 bytes immediately. |
| 11 | The write then fills the file. Between truncation and final close() there is |
| 12 | a window where the file is partially written or empty. |
| 13 | |
| 14 | The fix: use write_text_atomic() which writes to a mkstemp temp file and then |
| 15 | os.rename()s into place. rename() is atomic on POSIX — readers always see |
| 16 | either the old content or the complete new content, never 0 bytes. |
| 17 | |
| 18 | write_text_atomic is already imported in coord_sync.py and used throughout |
| 19 | the codebase for exactly this purpose. |
| 20 | |
| 21 | Coverage: |
| 22 | Unit — _write_remote_records produces valid files post-write |
| 23 | Concurrency — concurrent writers never leave 0-byte or invalid-JSON files |
| 24 | Data integrity — written content exactly matches the source record |
| 25 | Idempotency — overwriting same UUID produces correct final state |
| 26 | Security — path validation still enforced after atomicity fix |
| 27 | Performance — atomic write is not dramatically slower than write_text |
| 28 | Regression — all existing _write_remote_records security tests still pass |
| 29 | """ |
| 30 | from __future__ import annotations |
| 31 | |
| 32 | import json |
| 33 | import pathlib |
| 34 | import threading |
| 35 | import time |
| 36 | from unittest.mock import patch |
| 37 | |
| 38 | import pytest |
| 39 | |
| 40 | from muse.core._types import MsgpackDict |
| 41 | |
| 42 | # --------------------------------------------------------------------------- |
| 43 | # Helpers |
| 44 | # --------------------------------------------------------------------------- |
| 45 | |
| 46 | _ALL_KINDS = ("reservation", "intent", "release", "heartbeat", "dependency", "task", "claim") |
| 47 | _FUTURE_TS = "2099-12-31T23:59:59+00:00" |
| 48 | |
| 49 | |
| 50 | def _make_repo(tmp_path: pathlib.Path) -> pathlib.Path: |
| 51 | (tmp_path / ".muse").mkdir(parents=True, exist_ok=True) |
| 52 | return tmp_path |
| 53 | |
| 54 | |
| 55 | def _record(kind: str, uuid: str, payload_size: int = 64) -> MsgpackDict: |
| 56 | return { |
| 57 | "kind": kind, |
| 58 | "record_uuid": uuid, |
| 59 | "run_id": "run-torvalds", |
| 60 | "payload": {"data": "k" * payload_size}, |
| 61 | "expires_at": _FUTURE_TS, |
| 62 | } |
| 63 | |
| 64 | |
| 65 | def _remote_path(root: pathlib.Path, kind: str, uuid: str) -> pathlib.Path: |
| 66 | return root / ".muse" / "coordination" / "remote" / kind / f"{uuid}.json" |
| 67 | |
| 68 | |
| 69 | def _write_remote(root: pathlib.Path, records: list[dict]) -> None: |
| 70 | from muse.cli.commands.coord_sync import _write_remote_records |
| 71 | _write_remote_records(root, records) |
| 72 | |
| 73 | |
| 74 | # ============================================================================= |
| 75 | # 1. UNIT — basic correctness after write |
| 76 | # ============================================================================= |
| 77 | |
| 78 | class TestWriteRemoteUnit: |
| 79 | |
| 80 | def test_written_file_is_valid_json(self, tmp_path: pathlib.Path) -> None: |
| 81 | root = _make_repo(tmp_path) |
| 82 | rec = _record("reservation", "res-000001") |
| 83 | _write_remote(root, [rec]) |
| 84 | path = _remote_path(root, "reservation", "res-000001") |
| 85 | assert path.exists() |
| 86 | data = json.loads(path.read_text()) |
| 87 | assert data["kind"] == "reservation" |
| 88 | assert data["record_uuid"] == "res-000001" |
| 89 | |
| 90 | def test_written_file_content_exactly_matches_record(self, tmp_path: pathlib.Path) -> None: |
| 91 | root = _make_repo(tmp_path) |
| 92 | rec = _record("task", "task-abc123", payload_size=256) |
| 93 | _write_remote(root, [rec]) |
| 94 | path = _remote_path(root, "task", "task-abc123") |
| 95 | data = json.loads(path.read_text()) |
| 96 | assert data == rec |
| 97 | |
| 98 | def test_file_is_never_empty_after_write(self, tmp_path: pathlib.Path) -> None: |
| 99 | root = _make_repo(tmp_path) |
| 100 | for i in range(20): |
| 101 | rec = _record("heartbeat", f"hb-{i:04d}") |
| 102 | _write_remote(root, [rec]) |
| 103 | path = _remote_path(root, "heartbeat", f"hb-{i:04d}") |
| 104 | content = path.read_text() |
| 105 | assert content, f"file {path} is empty after write" |
| 106 | |
| 107 | def test_overwrite_produces_new_content(self, tmp_path: pathlib.Path) -> None: |
| 108 | root = _make_repo(tmp_path) |
| 109 | rec_v1 = _record("claim", "claim-001") |
| 110 | rec_v1["payload"] = {"version": 1} |
| 111 | _write_remote(root, [rec_v1]) |
| 112 | |
| 113 | rec_v2 = _record("claim", "claim-001") |
| 114 | rec_v2["payload"] = {"version": 2} |
| 115 | _write_remote(root, [rec_v2]) |
| 116 | |
| 117 | path = _remote_path(root, "claim", "claim-001") |
| 118 | data = json.loads(path.read_text()) |
| 119 | assert data["payload"]["version"] == 2 |
| 120 | |
| 121 | def test_all_seven_kinds_written_correctly(self, tmp_path: pathlib.Path) -> None: |
| 122 | root = _make_repo(tmp_path) |
| 123 | records = [_record(kind, f"{kind}-001") for kind in _ALL_KINDS] |
| 124 | _write_remote(root, records) |
| 125 | for kind in _ALL_KINDS: |
| 126 | path = _remote_path(root, kind, f"{kind}-001") |
| 127 | assert path.exists(), f"{kind} file not written" |
| 128 | data = json.loads(path.read_text()) |
| 129 | assert data["kind"] == kind |
| 130 | |
| 131 | |
| 132 | # ============================================================================= |
| 133 | # 2. CONCURRENCY — concurrent writers must never produce 0-byte or corrupt files |
| 134 | # ============================================================================= |
| 135 | |
| 136 | class TestWriteRemoteConcurrency: |
| 137 | """ |
| 138 | These tests FAIL before the fix (write_text) and PASS after (write_text_atomic). |
| 139 | |
| 140 | The bug: write_text opens with 'w' which truncates to 0 bytes before writing. |
| 141 | Between truncation and close(), concurrent readers see empty or partial files. |
| 142 | """ |
| 143 | |
| 144 | def test_concurrent_writes_same_uuid_no_zero_byte_reads(self, tmp_path: pathlib.Path) -> None: |
| 145 | """ |
| 146 | Writer thread: overwrites the same file 500 times. |
| 147 | Reader thread: reads the file concurrently, checks for 0-byte or corrupt reads. |
| 148 | """ |
| 149 | root = _make_repo(tmp_path) |
| 150 | rec = _record("reservation", "res-concurrent", payload_size=4096) |
| 151 | # Prime the file |
| 152 | _write_remote(root, [rec]) |
| 153 | |
| 154 | target = _remote_path(root, "reservation", "res-concurrent") |
| 155 | zero_byte_reads: list[int] = [] |
| 156 | corrupt_reads: list[str] = [] |
| 157 | stop = threading.Event() |
| 158 | |
| 159 | def writer() -> None: |
| 160 | for _ in range(500): |
| 161 | _write_remote(root, [rec]) |
| 162 | |
| 163 | def reader() -> None: |
| 164 | iteration = 0 |
| 165 | while not stop.is_set(): |
| 166 | try: |
| 167 | content = target.read_text() |
| 168 | if not content.strip(): |
| 169 | zero_byte_reads.append(iteration) |
| 170 | else: |
| 171 | json.loads(content) |
| 172 | except json.JSONDecodeError as exc: |
| 173 | corrupt_reads.append(str(exc)) |
| 174 | except FileNotFoundError: |
| 175 | pass # transient — rename may briefly make old name disappear |
| 176 | iteration += 1 |
| 177 | |
| 178 | t_writer = threading.Thread(target=writer) |
| 179 | t_reader = threading.Thread(target=reader) |
| 180 | t_reader.start() |
| 181 | t_writer.start() |
| 182 | t_writer.join() |
| 183 | stop.set() |
| 184 | t_reader.join() |
| 185 | |
| 186 | assert zero_byte_reads == [], ( |
| 187 | f"Saw {len(zero_byte_reads)} zero-byte reads — write_text truncation window exposed.\n" |
| 188 | f"First occurrence at reader iteration {zero_byte_reads[0]}.\n" |
| 189 | f"Fix: use write_text_atomic() instead of write_text()." |
| 190 | ) |
| 191 | assert corrupt_reads == [], ( |
| 192 | f"Saw {len(corrupt_reads)} corrupt-JSON reads — torn write occurred.\n" |
| 193 | f"First error: {corrupt_reads[0]}" |
| 194 | ) |
| 195 | |
| 196 | def test_concurrent_writes_different_uuids_all_valid(self, tmp_path: pathlib.Path) -> None: |
| 197 | """ |
| 198 | 8 threads each writing their own UUID 100 times. |
| 199 | All files must be valid JSON after all threads complete. |
| 200 | """ |
| 201 | root = _make_repo(tmp_path) |
| 202 | n_threads = 8 |
| 203 | n_writes = 100 |
| 204 | |
| 205 | def worker(idx: int) -> None: |
| 206 | rec = _record("intent", f"intent-{idx:04d}", payload_size=2048) |
| 207 | for _ in range(n_writes): |
| 208 | _write_remote(root, [rec]) |
| 209 | |
| 210 | threads = [threading.Thread(target=worker, args=(i,)) for i in range(n_threads)] |
| 211 | for t in threads: |
| 212 | t.start() |
| 213 | for t in threads: |
| 214 | t.join() |
| 215 | |
| 216 | # All files must be valid JSON |
| 217 | for i in range(n_threads): |
| 218 | path = _remote_path(root, "intent", f"intent-{i:04d}") |
| 219 | assert path.exists(), f"intent-{i:04d}.json missing" |
| 220 | content = path.read_text() |
| 221 | assert content, f"intent-{i:04d}.json is empty" |
| 222 | data = json.loads(content) |
| 223 | assert data["record_uuid"] == f"intent-{i:04d}" |
| 224 | |
| 225 | def test_concurrent_pull_and_read_no_corruption(self, tmp_path: pathlib.Path) -> None: |
| 226 | """ |
| 227 | Simulates two concurrent coord sync pull operations writing to the same remote dir. |
| 228 | Both pull the same record. Reader checks consistency throughout. |
| 229 | """ |
| 230 | root = _make_repo(tmp_path) |
| 231 | rec = _record("release", "rel-kernel-6-14", payload_size=8192) |
| 232 | _write_remote(root, [rec]) # prime |
| 233 | |
| 234 | target = _remote_path(root, "release", "rel-kernel-6-14") |
| 235 | errors: list[str] = [] |
| 236 | stop = threading.Event() |
| 237 | |
| 238 | def puller() -> None: |
| 239 | for _ in range(300): |
| 240 | _write_remote(root, [rec]) |
| 241 | |
| 242 | def reader() -> None: |
| 243 | while not stop.is_set(): |
| 244 | try: |
| 245 | content = target.read_text() |
| 246 | if not content.strip(): |
| 247 | errors.append("zero-byte file read") |
| 248 | else: |
| 249 | json.loads(content) |
| 250 | except json.JSONDecodeError as exc: |
| 251 | errors.append(f"corrupt JSON: {exc}") |
| 252 | except FileNotFoundError: |
| 253 | pass |
| 254 | |
| 255 | t1 = threading.Thread(target=puller) |
| 256 | t2 = threading.Thread(target=puller) |
| 257 | t_reader = threading.Thread(target=reader) |
| 258 | |
| 259 | t_reader.start() |
| 260 | t1.start() |
| 261 | t2.start() |
| 262 | t1.join() |
| 263 | t2.join() |
| 264 | stop.set() |
| 265 | t_reader.join() |
| 266 | |
| 267 | assert errors == [], ( |
| 268 | f"{len(errors)} corruption events during concurrent pull simulation.\n" |
| 269 | f"First: {errors[0]}" |
| 270 | ) |
| 271 | |
| 272 | |
| 273 | # ============================================================================= |
| 274 | # 3. DATA INTEGRITY — written content survives overwrite correctly |
| 275 | # ============================================================================= |
| 276 | |
| 277 | class TestWriteRemoteDataIntegrity: |
| 278 | |
| 279 | def test_large_payload_written_completely(self, tmp_path: pathlib.Path) -> None: |
| 280 | """50KB payload must be written and read back completely.""" |
| 281 | root = _make_repo(tmp_path) |
| 282 | rec = _record("dependency", "dep-kernel-mm", payload_size=50 * 1024) |
| 283 | _write_remote(root, [rec]) |
| 284 | |
| 285 | path = _remote_path(root, "dependency", "dep-kernel-mm") |
| 286 | data = json.loads(path.read_text()) |
| 287 | assert len(data["payload"]["data"]) == 50 * 1024 |
| 288 | |
| 289 | def test_sequential_overwrites_produce_correct_final_state(self, tmp_path: pathlib.Path) -> None: |
| 290 | """100 sequential overwrites of the same UUID — final content is correct.""" |
| 291 | root = _make_repo(tmp_path) |
| 292 | for version in range(100): |
| 293 | rec = _record("reservation", "res-overwrite") |
| 294 | rec["payload"]["version"] = version |
| 295 | _write_remote(root, [rec]) |
| 296 | |
| 297 | path = _remote_path(root, "reservation", "res-overwrite") |
| 298 | data = json.loads(path.read_text()) |
| 299 | assert data["payload"]["version"] == 99 |
| 300 | |
| 301 | def test_batch_write_all_files_present(self, tmp_path: pathlib.Path) -> None: |
| 302 | """Writing 1000 records in one call — all files must be present and valid.""" |
| 303 | root = _make_repo(tmp_path) |
| 304 | records = [ |
| 305 | _record(_ALL_KINDS[i % len(_ALL_KINDS)], f"rec-{i:06d}") |
| 306 | for i in range(1000) |
| 307 | ] |
| 308 | _write_remote(root, records) |
| 309 | |
| 310 | for i, rec in enumerate(records): |
| 311 | path = _remote_path(root, rec["kind"], f"rec-{i:06d}") |
| 312 | assert path.exists(), f"rec-{i:06d} missing" |
| 313 | data = json.loads(path.read_text()) |
| 314 | assert data["record_uuid"] == f"rec-{i:06d}" |
| 315 | |
| 316 | def test_file_never_partially_written_single_thread(self, tmp_path: pathlib.Path) -> None: |
| 317 | """Single-threaded: read immediately after each write must be complete.""" |
| 318 | root = _make_repo(tmp_path) |
| 319 | for i in range(50): |
| 320 | rec = _record("task", "task-sequential", payload_size=1024 * (i % 10 + 1)) |
| 321 | _write_remote(root, [rec]) |
| 322 | path = _remote_path(root, "task", "task-sequential") |
| 323 | content = path.read_text() |
| 324 | assert content, f"empty file after write {i}" |
| 325 | data = json.loads(content) |
| 326 | assert data == rec, f"content mismatch at write {i}" |
| 327 | |
| 328 | |
| 329 | # ============================================================================= |
| 330 | # 4. SECURITY — path validation still enforced (regression) |
| 331 | # ============================================================================= |
| 332 | |
| 333 | class TestWriteRemoteSecurity: |
| 334 | """Verify the fix doesn't break existing security validation.""" |
| 335 | |
| 336 | def test_unknown_kind_rejected(self, tmp_path: pathlib.Path) -> None: |
| 337 | root = _make_repo(tmp_path) |
| 338 | rec = {"kind": "../evil", "record_uuid": "safe-uuid", "payload": {}} |
| 339 | _write_remote(root, [rec]) |
| 340 | evil_path = root / ".muse" / "coordination" / "remote" / ".." / "evil" / "safe-uuid.json" |
| 341 | assert not evil_path.resolve().exists(), "path traversal via kind succeeded" |
| 342 | |
| 343 | def test_unsafe_uuid_rejected(self, tmp_path: pathlib.Path) -> None: |
| 344 | root = _make_repo(tmp_path) |
| 345 | rec = {"kind": "reservation", "record_uuid": "../../../etc/passwd", "payload": {}} |
| 346 | _write_remote(root, [rec]) |
| 347 | evil = root / ".muse" / "coordination" / "remote" / "reservation" / "../../../etc/passwd.json" |
| 348 | assert not pathlib.Path("/etc/passwd.json").exists() or True # sanity |
| 349 | # The file must not be written outside the remote/ dir |
| 350 | remote_dir = root / ".muse" / "coordination" / "remote" |
| 351 | written = list(remote_dir.rglob("*.json")) if remote_dir.exists() else [] |
| 352 | assert written == [], f"unsafe UUID produced files: {written}" |
| 353 | |
| 354 | def test_null_byte_uuid_rejected(self, tmp_path: pathlib.Path) -> None: |
| 355 | root = _make_repo(tmp_path) |
| 356 | rec = {"kind": "reservation", "record_uuid": "valid\x00evil", "payload": {}} |
| 357 | _write_remote(root, [rec]) |
| 358 | remote_dir = root / ".muse" / "coordination" / "remote" |
| 359 | written = list(remote_dir.rglob("*.json")) if remote_dir.exists() else [] |
| 360 | assert written == [], f"null-byte UUID produced files: {written}" |
| 361 | |
| 362 | def test_valid_records_still_written_after_invalid_ones_skipped(self, tmp_path: pathlib.Path) -> None: |
| 363 | root = _make_repo(tmp_path) |
| 364 | records = [ |
| 365 | {"kind": "../evil", "record_uuid": "uuid-1", "payload": {}}, # bad kind |
| 366 | _record("reservation", "res-valid-001"), # good |
| 367 | {"kind": "reservation", "record_uuid": "../bad", "payload": {}}, # bad uuid |
| 368 | _record("task", "task-valid-001"), # good |
| 369 | ] |
| 370 | _write_remote(root, records) |
| 371 | |
| 372 | assert _remote_path(root, "reservation", "res-valid-001").exists() |
| 373 | assert _remote_path(root, "task", "task-valid-001").exists() |
| 374 | # Only 2 valid files |
| 375 | remote_dir = root / ".muse" / "coordination" / "remote" |
| 376 | written = list(remote_dir.rglob("*.json")) |
| 377 | assert len(written) == 2 |
| 378 | |
| 379 | |
| 380 | # ============================================================================= |
| 381 | # 5. PERFORMANCE — atomic write is not dramatically slower than write_text |
| 382 | # ============================================================================= |
| 383 | |
| 384 | class TestWriteRemotePerformance: |
| 385 | |
| 386 | def test_500_records_completes_under_3s(self, tmp_path: pathlib.Path) -> None: |
| 387 | root = _make_repo(tmp_path) |
| 388 | records = [ |
| 389 | _record(_ALL_KINDS[i % len(_ALL_KINDS)], f"perf-{i:06d}", payload_size=512) |
| 390 | for i in range(500) |
| 391 | ] |
| 392 | t0 = time.monotonic() |
| 393 | _write_remote(root, records) |
| 394 | elapsed = time.monotonic() - t0 |
| 395 | assert elapsed < 3.0, f"500 atomic writes took {elapsed:.3f}s (> 3s)" |
| 396 | |
| 397 | def test_atomic_write_throughput_records_per_second(self, tmp_path: pathlib.Path) -> None: |
| 398 | root = _make_repo(tmp_path) |
| 399 | records = [_record("reservation", f"tput-{i:06d}") for i in range(200)] |
| 400 | |
| 401 | t0 = time.monotonic() |
| 402 | _write_remote(root, records) |
| 403 | elapsed = time.monotonic() - t0 |
| 404 | |
| 405 | rps = 200 / elapsed if elapsed > 0 else float("inf") |
| 406 | assert rps >= 50, f"write_remote throughput {rps:.0f} rec/s is below minimum 50/s" |
File History
5 commits
sha256:660fcac1df3ab28f61862e961890bd2ca8b754fa0242079d93ca1e25037ec8a6
chore(tests): add docstring to tests/__init__.py so rc14 tr…
Human
26 days ago
sha256:d8316ffae901be06347e16ab55be11868eb519dd16ade3e8aa16a99e662f7e62
baseline: rc14 re-baseline after rc3 store corruption recovery
Human
patch
26 days ago
sha256:50d413a635f57761c3fce1f70251b957b6423821525bf330747ef4007339222e
Merge branch 'dev' into main
Human
29 days ago
sha256:fb67fed5a4d3e40de84bdd163de94ef1386570bef1dd1a020a732c8a038962ce
Merge branch 'dev' into main
Human
48 days ago
sha256:1c4b3e3a9a1f300774c3ee662b572a698d5fd405bf765a71e6011a2e9c3eaaaa
feat: Muse — version control for the agent era
Human
100 days ago