File size: 12,357 Bytes
ae73c7f | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 | """
Test scenarios required before this codebase's core contracts could be
considered complete (per project guidelines: nothing is done without
test scenarios for duplicate requests, edge cases, and race conditions).
Run: python -m pytest tests/test_contracts.py -v
"""
from __future__ import annotations
import os
import sys
import shutil
import tempfile
from pathlib import Path
import numpy as np
import pytest
import torch
sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
from src.data_real import LocalWellHDF5, get_dataset, get_synthetic_dataset
from src.provenance import (
DataLoadError,
SchemaValidationError,
TrajectoryTooShortError,
EmptyDatasetError,
DatasetAlreadyUsedError,
DatasetInProgressError,
DatasetRegistry,
CheckpointStore,
hash_config,
hash_dataset,
hash_code,
combined_identity_hash,
validate_trajectory_lengths,
)
@pytest.fixture
def tmpdir():
d = tempfile.mkdtemp()
yield d
shutil.rmtree(d, ignore_errors=True)
# --------------------------------------------------------------------- #
# Scenario 1: missing real-data directory -> hard fail, no substitution
# --------------------------------------------------------------------- #
def test_get_dataset_hard_fails_when_no_real_data_directory_exists(tmpdir):
nonexistent = os.path.join(tmpdir, "does_not_exist")
with pytest.raises(DataLoadError) as exc_info:
get_dataset(search_roots=[nonexistent])
assert exc_info.value.outcome_code == "NO_DATA_DIRECTORY"
def test_get_dataset_hard_fails_on_empty_real_data_directory(tmpdir):
empty_dir = os.path.join(tmpdir, "real")
os.makedirs(empty_dir)
with pytest.raises(DataLoadError) as exc_info:
get_dataset(search_roots=[empty_dir])
assert exc_info.value.outcome_code == "NO_FILES_FOUND"
def test_get_dataset_never_silently_returns_synthetic_data(tmpdir):
nonexistent = os.path.join(tmpdir, "nope")
try:
ds, provenance = get_dataset(search_roots=[nonexistent])
assert False, "should have raised, not returned a dataset"
except DataLoadError:
pass # correct: hard failure, no dataset returned at all
# --------------------------------------------------------------------- #
# Scenario 2: HDF5 with schema that doesn't match declared expectations
# --------------------------------------------------------------------- #
def test_local_hdf5_rejects_wrong_channel_count(tmpdir):
h5py = pytest.importorskip("h5py")
real_dir = os.path.join(tmpdir, "real")
os.makedirs(real_dir)
fp = os.path.join(real_dir, "sample.hdf5")
with h5py.File(fp, "w") as f:
# (T=10, C=5, H=16, W=16) but caller expects 2 channels
f.create_dataset("fields", data=np.random.randn(10, 5, 16, 16).astype("float32"))
with pytest.raises(SchemaValidationError) as exc_info:
LocalWellHDF5(real_dir, expected_channels=2, channel_layout="channels_first", strict=True)
assert exc_info.value.outcome_code in ("CHANNEL_COUNT_MISMATCH", "NO_VALID_TRAJECTORIES")
def test_local_hdf5_rejects_wrong_ndim(tmpdir):
h5py = pytest.importorskip("h5py")
real_dir = os.path.join(tmpdir, "real")
os.makedirs(real_dir)
fp = os.path.join(real_dir, "sample.hdf5")
with h5py.File(fp, "w") as f:
f.create_dataset("fields", data=np.random.randn(10, 16, 16).astype("float32")) # 3D, no channel axis
with pytest.raises(SchemaValidationError):
LocalWellHDF5(real_dir, expected_channels=2, strict=True)
def test_local_hdf5_accepts_correctly_shaped_data(tmpdir):
h5py = pytest.importorskip("h5py")
real_dir = os.path.join(tmpdir, "real")
os.makedirs(real_dir)
fp = os.path.join(real_dir, "sample.hdf5")
with h5py.File(fp, "w") as f:
f.create_dataset("fields", data=np.random.randn(10, 2, 16, 16).astype("float32"))
ds = LocalWellHDF5(real_dir, expected_channels=2, channel_layout="channels_first", strict=True)
assert len(ds) == 1
item = ds[0]
assert item["fields"].shape == (10, 2, 16, 16)
# --------------------------------------------------------------------- #
# Scenario 3: trajectory-length validation fails loudly, before training
# --------------------------------------------------------------------- #
def test_validate_trajectory_lengths_raises_on_short_trajectories():
ds, _ = get_synthetic_dataset(max_samples=8, n_steps=5) # too short
with pytest.raises(TrajectoryTooShortError) as exc_info:
validate_trajectory_lengths(ds, required_length=10)
assert exc_info.value.outcome_code == "TRAJECTORY_TOO_SHORT"
def test_validate_trajectory_lengths_passes_on_sufficient_trajectories():
ds, _ = get_synthetic_dataset(max_samples=8, n_steps=14)
result = validate_trajectory_lengths(ds, required_length=8)
assert result["success"] is True
def test_validate_trajectory_lengths_raises_on_empty_dataset():
class Empty:
def __len__(self):
return 0
with pytest.raises(EmptyDatasetError):
validate_trajectory_lengths(Empty(), required_length=8)
# --------------------------------------------------------------------- #
# Scenario 4: checkpoint atomicity — no partial/corrupt file ever visible
# --------------------------------------------------------------------- #
def test_checkpoint_save_is_atomic_no_temp_file_left_behind(tmpdir):
store = CheckpointStore(checkpoints_dir=os.path.join(tmpdir, "ckpts"))
result = store.save(
model_state={"w": torch.randn(4, 4)},
config={"lr": 0.001, "hidden": 32},
dataset_hash="a" * 64,
code_hash="b" * 64,
data_provenance="SYNTHETIC",
)
assert result["success"] is True
assert result["outcome_code"] == "SAVED"
ckpt_dir = Path(tmpdir) / "ckpts"
files = list(ckpt_dir.iterdir())
# No .tmp_* files should remain after a successful save.
assert not any(f.name.startswith(".tmp_") for f in files)
# Exactly the final .pt and .meta.json should exist.
assert any(f.suffix == ".pt" for f in files)
assert any(f.name.endswith(".meta.json") for f in files)
def test_checkpoint_save_is_idempotent_for_identical_inputs(tmpdir):
store = CheckpointStore(checkpoints_dir=os.path.join(tmpdir, "ckpts"))
kwargs = dict(
model_state={"w": torch.randn(4, 4)},
config={"lr": 0.001},
dataset_hash="c" * 64,
code_hash="d" * 64,
data_provenance="SYNTHETIC",
)
r1 = store.save(**kwargs)
r2 = store.save(**kwargs)
assert r1["identity_hash"] == r2["identity_hash"]
assert r1["outcome_code"] == "SAVED"
assert r2["outcome_code"] == "DUPLICATE_EXISTS"
ckpt_dir = Path(tmpdir) / "ckpts"
pt_files = [f for f in ckpt_dir.iterdir() if f.suffix == ".pt"]
assert len(pt_files) == 1 # not duplicated
def test_checkpoint_save_rejects_missing_dataset_hash(tmpdir):
store = CheckpointStore(checkpoints_dir=os.path.join(tmpdir, "ckpts"))
with pytest.raises(Exception) as exc_info:
store.save(
model_state={"w": torch.randn(2, 2)},
config={"lr": 0.001},
dataset_hash="",
code_hash="e" * 64,
data_provenance="SYNTHETIC",
)
assert "MISSING_DATASET_HASH" in str(exc_info.value)
def test_checkpoint_load_detects_missing_meta_as_integrity_failure(tmpdir):
from src.provenance import CheckpointIntegrityError
store = CheckpointStore(checkpoints_dir=os.path.join(tmpdir, "ckpts"))
with pytest.raises(CheckpointIntegrityError):
store.load("nonexistent" * 8)
def test_checkpoint_extra_with_tensors_does_not_crash_meta_write(tmpdir):
store = CheckpointStore(checkpoints_dir=os.path.join(tmpdir, "ckpts"))
result = store.save(
model_state={"w": torch.randn(3, 3)},
config={"lr": 0.001},
dataset_hash="9" * 64,
code_hash="8" * 64,
data_provenance="SYNTHETIC",
extra={
"normalizer": {"mean": torch.tensor([0.1, 0.2]), "std": torch.tensor([1.0, 1.0])},
"ppo_returns": [1.0, 2.5, -3.2],
},
)
assert result["success"] is True
loaded = store.load(result["identity_hash"])
assert loaded["meta"]["extra"]["normalizer"]["mean"] == [pytest.approx(0.1), pytest.approx(0.2)]
def test_checkpoint_pt_without_meta_is_integrity_failure(tmpdir):
from src.provenance import CheckpointIntegrityError
ckpt_dir = Path(tmpdir) / "ckpts"
ckpt_dir.mkdir(parents=True)
fake_identity = "z" * 64
torch.save({"w": torch.randn(2, 2)}, ckpt_dir / f"{fake_identity}.pt")
# deliberately do NOT write the .meta.json sidecar
store = CheckpointStore(checkpoints_dir=str(ckpt_dir))
with pytest.raises(CheckpointIntegrityError):
store.load(fake_identity)
# --------------------------------------------------------------------- #
# Scenario 5: dataset-reuse registry blocks retraining on consumed data
# --------------------------------------------------------------------- #
def test_registry_blocks_retraining_on_already_consumed_dataset(tmpdir):
registry = DatasetRegistry(registry_dir=os.path.join(tmpdir, "registry"))
dataset_hash = "f" * 64
registry.claim(dataset_hash, experiment_id="exp1")
registry.mark_consumed(dataset_hash)
with pytest.raises(DatasetAlreadyUsedError):
registry.claim(dataset_hash, experiment_id="exp2")
def test_registry_blocks_concurrent_in_progress_claim(tmpdir):
registry = DatasetRegistry(registry_dir=os.path.join(tmpdir, "registry"))
dataset_hash = "0" * 64
registry.claim(dataset_hash, experiment_id="exp1") # first claim succeeds
with pytest.raises(DatasetInProgressError):
registry.claim(dataset_hash, experiment_id="exp2") # second is blocked
def test_registry_allows_retry_after_explicit_human_action(tmpdir):
registry = DatasetRegistry(registry_dir=os.path.join(tmpdir, "registry"))
dataset_hash = "1" * 64
registry.claim(dataset_hash, experiment_id="exp1")
registry.mark_failed(dataset_hash, error_detail="crashed")
# Second claim after a FAILED run is blocked until an explicit human
# action (allow_retry) is taken — not automatic.
with pytest.raises(DatasetInProgressError):
registry.claim(dataset_hash, experiment_id="exp2")
registry.allow_retry(dataset_hash)
result = registry.claim(dataset_hash, experiment_id="exp2") # now succeeds
assert result.success is True
def test_registry_status_none_for_unknown_dataset(tmpdir):
registry = DatasetRegistry(registry_dir=os.path.join(tmpdir, "registry"))
assert registry.status("nonexistent" * 8) is None
# --------------------------------------------------------------------- #
# Scenario 6: hashing determinism (identity must be stable and content-based)
# --------------------------------------------------------------------- #
def test_hash_config_is_order_independent():
h1 = hash_config({"lr": 0.001, "hidden": 64})
h2 = hash_config({"hidden": 64, "lr": 0.001})
assert h1 == h2
def test_hash_config_differs_for_different_values():
h1 = hash_config({"lr": 0.001})
h2 = hash_config({"lr": 0.002})
assert h1 != h2
def test_hash_dataset_is_content_based_not_path_based():
ds_a, _ = get_synthetic_dataset(max_samples=4, n_steps=8)
ds_b, _ = get_synthetic_dataset(max_samples=4, n_steps=8)
# Different random seeds inside SyntheticWellLike -> different content
# (this asserts the hash actually reflects content, not just shape).
assert hash_dataset(ds_a) != hash_dataset(ds_b) or torch.allclose(
ds_a[0]["fields"], ds_b[0]["fields"]
)
def test_hash_dataset_rejects_non_tensor_items():
class BadDataset:
def __len__(self):
return 1
def __getitem__(self, idx):
return {"fields": "not a tensor"}
with pytest.raises(SchemaValidationError):
hash_dataset(BadDataset())
def test_combined_identity_hash_changes_if_any_component_changes():
base = combined_identity_hash("cfg", "code", "data")
assert combined_identity_hash("cfg2", "code", "data") != base
assert combined_identity_hash("cfg", "code2", "data") != base
assert combined_identity_hash("cfg", "code", "data2") != base
if __name__ == "__main__":
sys.exit(pytest.main([__file__, "-v"]))
|