| 1 | # SPDX-License-Identifier: Apache-2.0 |
| 2 | # SPDX-FileCopyrightText: Copyright contributors to the vLLM project |
| 3 | |
| 4 | import multiprocessing |
| 5 | import random |
| 6 | import time |
| 7 | import traceback |
| 8 | import unittest |
| 9 | from multiprocessing import Lock |
| 10 | |
| 11 | import torch |
| 12 | |
| 13 | # Assuming these are imported from your module |
| 14 | from vllm.distributed.device_communicators.shm_object_storage import ( |
| 15 | MsgpackSerde, |
| 16 | SingleWriterShmObjectStorage, |
| 17 | SingleWriterShmRingBuffer, |
| 18 | ) |
| 19 | from vllm.multimodal.inputs import ( |
| 20 | MultiModalFieldElem, |
| 21 | MultiModalKwargsItem, |
| 22 | MultiModalSharedField, |
| 23 | ) |
| 24 | |
| 25 | |
| 26 | def _dummy_elem(size: int): |
| 27 | return MultiModalFieldElem( |
| 28 | data=torch.empty((size,), dtype=torch.int8), |
| 29 | field=MultiModalSharedField(batch_size=1), |
| 30 | ) |
| 31 | |
| 32 | |
| 33 | def _dummy_item(size_by_key: dict[str, int]): |
| 34 | return MultiModalKwargsItem( |
| 35 | {key: _dummy_elem(size) for key, size in size_by_key.items()} |
| 36 | ) |
| 37 | |
| 38 | |
| 39 | class TestSingleWriterShmObjectStorage(unittest.TestCase): |
| 40 | def setUp(self): |
| 41 | """Set up test fixtures before each test method.""" |
| 42 | ring_buffer = SingleWriterShmRingBuffer( |
| 43 | data_buffer_size=1024 * 100, |
| 44 | create=True, # 10 MB buffer |
| 45 | ) |
| 46 | self.storage = SingleWriterShmObjectStorage( |
| 47 | max_object_size=1024 * 10, # 10KB max object |
| 48 | n_readers=2, |
| 49 | ring_buffer=ring_buffer, |
| 50 | serde_class=MsgpackSerde, |
| 51 | reader_lock=Lock(), |
| 52 | ) |
| 53 | |
| 54 | def tearDown(self): |
| 55 | """Clean up after each test.""" |
| 56 | if self.storage: |
| 57 | self.storage.close() |
| 58 | |
| 59 | def test_minimal_put_get_cycle(self): |
| 60 | """Test basic put and get operations.""" |
| 61 | key = "test_key" |
| 62 | value = _dummy_item({"field1": 10, "field2": 20}) |
| 63 | |
| 64 | # Put operation |
| 65 | address, monotonic_id = self.storage.put(key, value) |
| 66 | |
| 67 | # Verify key is in index |
| 68 | self.assertIn(key, self.storage.key_index) |
| 69 | self.assertEqual(self.storage.key_index[key], (address, monotonic_id)) |
| 70 | self.assertEqual(self.storage.id_index[monotonic_id], key) |
| 71 | |
| 72 | # Get operation |
| 73 | result = self.storage.get(address, monotonic_id) |
| 74 | |
| 75 | # Verify result |
| 76 | self.assertEqual(result, value) |
| 77 | |
| 78 | def test_put_same_key_twice(self): |
| 79 | """Test behavior when putting the same key multiple times.""" |
| 80 | key = "duplicate_key" |
| 81 | value1 = "first value" |
| 82 | value2 = "second value" |
| 83 | |
| 84 | # First put |
| 85 | address1, id1 = self.storage.put(key, value1) |
| 86 | retrieved1 = self.storage.get(address1, id1) |
| 87 | self.assertEqual(retrieved1, value1) |
| 88 | |
| 89 | # should raise an error on second put |
| 90 | with self.assertRaises(ValueError) as context: |
| 91 | self.storage.put(key, value2) |
| 92 | |
| 93 | self.assertIn("already exists in the storage", str(context.exception)) |
| 94 | |
| 95 | def test_large_object_rejection(self): |
| 96 | """Test that objects exceeding max_object_size are rejected.""" |
| 97 | # Create an object larger than max_object_size |
| 98 | large_data = "x" * (self.storage.max_object_size + 100) |
| 99 | |
| 100 | with self.assertRaises(ValueError) as context: |
| 101 | self.storage.put("large_key", large_data) |
| 102 | |
| 103 | self.assertIn("exceeds max object size", str(context.exception)) |
| 104 | |
| 105 | def test_buffer_overflow_and_cleanup(self): |
| 106 | """Test behavior when buffer fills up and needs cleanup.""" |
| 107 | # Fill up the buffer with many small objects |
| 108 | stored_items = [] |
| 109 | |
| 110 | try: |
| 111 | for i in range(1000): # Try to store many items |
| 112 | key = f"item_{i}" |
| 113 | value = f"data_{i}" * 100 # Make it reasonably sized |
| 114 | address, monotonic_id = self.storage.put(key, value) |
| 115 | stored_items.append((key, value, address, monotonic_id)) |
| 116 | except MemoryError: |
| 117 | print(f"Buffer filled after {len(stored_items)} items") |
| 118 | |
| 119 | # Verify that some items are still accessible |
| 120 | accessible_count = 0 |
| 121 | for key, original_value, address, monotonic_id in stored_items: |
| 122 | for i in range(self.storage.n_readers): |
| 123 | retrieved = self.storage.get(address, monotonic_id) |
| 124 | if retrieved == original_value: |
| 125 | accessible_count += 1 |
| 126 | |
| 127 | self.assertEqual(accessible_count, len(stored_items)) |
| 128 | |
| 129 | try: |
| 130 | for i in range(len(stored_items), 1000): # Try to store many items |
| 131 | key = f"item_{i}" |
| 132 | value = f"data_{i}" * 100 # Make it reasonably sized |
| 133 | address, monotonic_id = self.storage.put(key, value) |
| 134 | stored_items.append((key, value, address, monotonic_id)) |
| 135 | except MemoryError: |
| 136 | print(f"Buffer filled after {len(stored_items)} items") |
| 137 | |
| 138 | # Verify that some items are still accessibles |
| 139 | for key, original_value, address, monotonic_id in stored_items: |
| 140 | try: |
| 141 | for i in range(self.storage.n_readers): |
| 142 | retrieved = self.storage.get(address, monotonic_id) |
| 143 | if retrieved == original_value: |
| 144 | accessible_count += 1 |
| 145 | except ValueError as e: |
| 146 | print(f"Error retrieving {key}: {e}") |
| 147 | |
| 148 | # some items from the first batch may still be accessible |
| 149 | self.assertGreaterEqual(accessible_count, len(stored_items)) |
| 150 | |
| 151 | def test_blocking_unread_object(self): |
| 152 | """Test behavior when buffer fills up and needs cleanup.""" |
| 153 | # Fill up the buffer with many small objects |
| 154 | stored_items = [] |
| 155 | |
| 156 | try: |
| 157 | for i in range(1000): # Try to store many items |
| 158 | key = f"item_{i}" |
| 159 | value = f"data_{i}" * 100 # Make it reasonably sized |
| 160 | address, monotonic_id = self.storage.put(key, value) |
| 161 | stored_items.append((key, value, address, monotonic_id)) |
| 162 | except MemoryError: |
| 163 | print(f"Buffer filled after {len(stored_items)} items") |
| 164 | |
| 165 | # read all items except the first one |
| 166 | # to simulate a blocking situation |
| 167 | accessible_count = 0 |
| 168 | for key, original_value, address, monotonic_id in stored_items[1:]: |
| 169 | for i in range(self.storage.n_readers): |
| 170 | retrieved = self.storage.get(address, monotonic_id) |
| 171 | if retrieved == original_value: |
| 172 | accessible_count += 1 |
| 173 | |
| 174 | self.assertEqual(accessible_count, len(stored_items) - 1) |
| 175 | |
| 176 | try: |
| 177 | key = f"item_{len(stored_items)}" |
| 178 | value = f"data_{len(stored_items)}" * 100 |
| 179 | address, monotonic_id = self.storage.put(key, value) |
| 180 | except MemoryError: |
| 181 | print(f"Buffer filled after {len(stored_items)} items") |
| 182 | |
| 183 | # read the first item |
| 184 | for i in range(self.storage.n_readers): |
| 185 | key, original_value, address, monotonic_id = stored_items[0] |
| 186 | retrieved = self.storage.get(address, monotonic_id) |
| 187 | self.assertEqual(retrieved, original_value) |
| 188 | |
| 189 | try: |
| 190 | for i in range(len(stored_items), 1000): # Try to store many items |
| 191 | key = f"item_{i}" |
| 192 | value = f"data_{i}" * 100 # Make it reasonably sized |
| 193 | address, monotonic_id = self.storage.put(key, value) |
| 194 | stored_items.append((key, value, address, monotonic_id)) |
| 195 | except MemoryError: |
| 196 | print(f"Buffer filled after {len(stored_items)} items") |
| 197 | |
| 198 | # some items from the first batch may still be accessible |
| 199 | self.assertGreaterEqual(len(stored_items), accessible_count + 10) |
| 200 | |
| 201 | def test_invalid_get_operations(self): |
| 202 | """Test various invalid get operations.""" |
| 203 | # Test with non-existent address |
| 204 | with self.assertRaises(ValueError): # Could be various exceptions |
| 205 | self.storage.get(99999, 1) |
| 206 | |
| 207 | # Store something first |
| 208 | address, monotonic_id = self.storage.put("test", "value") |
| 209 | |
| 210 | # Test with wrong monotonic_id |
| 211 | with self.assertRaises(ValueError) as context: |
| 212 | self.storage.get(address, monotonic_id + 100) |
| 213 | |
| 214 | self.assertIn("has been modified or is invalid", str(context.exception)) |
| 215 | |
| 216 | def test_clear_storage(self): |
| 217 | """Test clearing the storage.""" |
| 218 | # Store some items |
| 219 | for i in range(5): |
| 220 | self.storage.put(f"item_{i}", f"value_{i}") |
| 221 | |
| 222 | # Clear the storage |
| 223 | self.storage.clear() |
| 224 | |
| 225 | # Verify that all indices are empty |
| 226 | self.assertEqual(len(self.storage.key_index), 0) |
| 227 | self.assertEqual(len(self.storage.id_index), 0) |
| 228 | self.assertEqual(len(self.storage.ring_buffer.metadata), 0) |
| 229 | |
| 230 | # Verify that new items can be added after clearing. IDs keep |
| 231 | # increasing so handles issued before the clear cannot alias new data. |
| 232 | address, monotonic_id = self.storage.put("new_item", "new_value") |
| 233 | self.assertIn("new_item", self.storage.key_index) |
| 234 | self.assertEqual((address, monotonic_id), (0, 5)) |
| 235 | |
| 236 | def _open_reader(self) -> SingleWriterShmObjectStorage: |
| 237 | reader = SingleWriterShmObjectStorage.create_from_handle(self.storage.handle()) |
| 238 | self.addCleanup(reader.close) |
| 239 | return reader |
| 240 | |
| 241 | def test_reader_get_and_touch_each_require_valid_signature(self): |
| 242 | address, monotonic_id = self.storage.put("key", "value") |
| 243 | signature = self.storage.get_signature("key") |
| 244 | reader = self._open_reader() |
| 245 | |
| 246 | reader.touch("key", address, monotonic_id, signature) |
| 247 | self.assertEqual(reader.get(address, monotonic_id, signature, "key"), "value") |
| 248 | |
| 249 | flipped = [signature[0] ^ 1, *signature[1:]] |
| 250 | for bad in (None, flipped, [signature], signature[:-1], [256] * 32): |
| 251 | with self.assertRaisesRegex(ValueError, "SHM handle signature"): |
| 252 | reader.get(address, monotonic_id, bad, "key") |
| 253 | with self.assertRaisesRegex(ValueError, "SHM handle signature"): |
| 254 | reader.touch("key", address, monotonic_id, bad) |
| 255 | |
| 256 | def test_signature_is_bound_to_key_and_address(self): |
| 257 | address, monotonic_id = self.storage.put("key", "value") |
| 258 | signature = self.storage.get_signature("key") |
| 259 | reader = self._open_reader() |
| 260 | |
| 261 | with self.assertRaisesRegex(ValueError, "SHM handle signature"): |
| 262 | reader.get(address, monotonic_id, signature, "other_key") |
| 263 | with self.assertRaisesRegex(ValueError, "SHM handle signature"): |
| 264 | reader.get(address + 1, monotonic_id, signature, "key") |
| 265 | |
| 266 | def test_writer_rejects_handle_no_longer_in_key_index(self): |
| 267 | address, monotonic_id = self.storage.put("key", "value") |
| 268 | signature = self.storage.get_signature("key") |
| 269 | self.storage.verify_signature("key", address, monotonic_id, signature) |
| 270 | |
| 271 | self.storage.clear() |
| 272 | |
| 273 | with self.assertRaisesRegex(ValueError, "SHM handle signature"): |
| 274 | self.storage.verify_signature("key", address, monotonic_id, signature) |
| 275 | |
| 276 | def test_handle_issued_before_clear_reads_own_data_until_reused(self): |
| 277 | address, monotonic_id = self.storage.put("key", "value") |
| 278 | signature = self.storage.get_signature("key") |
| 279 | reader = self._open_reader() |
| 280 | |
| 281 | self.storage.clear() |
| 282 | |
| 283 | # Requests still draining after a clear keep reading their own data. |
| 284 | self.assertEqual(reader.get(address, monotonic_id, signature, "key"), "value") |
| 285 | |
| 286 | self.storage.put("key", "new_value") |
| 287 | |
| 288 | with self.assertRaisesRegex(ValueError, "modified or is invalid"): |
| 289 | reader.get(address, monotonic_id, signature, "key") |
| 290 | with self.assertRaisesRegex(ValueError, "modified or is invalid"): |
| 291 | reader.touch("key", address, monotonic_id, signature) |
| 292 | |
| 293 | |
| 294 | class TestShmObjectStorageReferenceCounting(unittest.TestCase): |
| 295 | """One writer and four readers, as with tensor-parallel workers.""" |
| 296 | |
| 297 | def setUp(self): |
| 298 | ring_buffer = SingleWriterShmRingBuffer( |
| 299 | data_buffer_size=1024 * 100, create=True |
| 300 | ) |
| 301 | self.writer = SingleWriterShmObjectStorage( |
| 302 | max_object_size=1024 * 10, |
| 303 | n_readers=4, |
| 304 | ring_buffer=ring_buffer, |
| 305 | serde_class=MsgpackSerde, |
| 306 | ) |
| 307 | self.addCleanup(self.writer.close) |
| 308 | handle = self.writer.handle() |
| 309 | handle.reader_lock = Lock() |
| 310 | self.readers = [ |
| 311 | SingleWriterShmObjectStorage.create_from_handle(handle) for _ in range(4) |
| 312 | ] |
| 313 | for reader in self.readers: |
| 314 | self.addCleanup(reader.close) |
| 315 | |
| 316 | def _reference(self, key: str) -> tuple[int, int, list[int], str]: |
| 317 | """Take one writer reference to the key, as one request item does.""" |
| 318 | if self.writer.is_cached(key): |
| 319 | address, monotonic_id = self.writer.get_cached(key) |
| 320 | else: |
| 321 | address, monotonic_id = self.writer.put(key, "value") |
| 322 | return address, monotonic_id, self.writer.get_signature(key), key |
| 323 | |
| 324 | @staticmethod |
| 325 | def _touch(reader: SingleWriterShmObjectStorage, reference) -> None: |
| 326 | address, monotonic_id, signature, key = reference |
| 327 | reader.touch(key, address, monotonic_id, signature) |
| 328 | |
| 329 | def _is_evictable(self, key: str) -> bool: |
| 330 | self.writer.free_unused() |
| 331 | return not self.writer.is_cached(key) |
| 332 | |
| 333 | def test_item_referenced_again_before_every_reader_fetched_it(self): |
| 334 | """A second request must not leave the item unevictable forever.""" |
| 335 | first = self._reference("image") |
| 336 | self.writer.touch("image") |
| 337 | second = self._reference("image") |
| 338 | self.writer.release_touches() |
| 339 | |
| 340 | # Every reader starts on the first request; the ones that finish it |
| 341 | # go on to the second while the others are still fetching. |
| 342 | for reader in self.readers: |
| 343 | self._touch(reader, first) |
| 344 | for reader in self.readers: |
| 345 | reader.get(*first) |
| 346 | self._touch(reader, second) |
| 347 | reader.get(*second) |
| 348 | |
| 349 | self.assertTrue(self._is_evictable("image")) |
| 350 | |
| 351 | def test_new_item_twice_in_one_request_is_not_evicted_early(self): |
| 352 | """The item must stay until the last reader has fetched it.""" |
| 353 | self.writer.touch("image") |
| 354 | references = [self._reference("image"), self._reference("image")] |
| 355 | self.writer.release_touches() |
| 356 | |
| 357 | for reader in self.readers[:-1]: |
| 358 | for reference in references: |
| 359 | self._touch(reader, reference) |
| 360 | for reference in references: |
| 361 | reader.get(*reference) |
| 362 | for reference in references: |
| 363 | self._touch(self.readers[-1], reference) |
| 364 | self.assertFalse(self._is_evictable("image")) |
| 365 | |
| 366 | for reference in references: |
| 367 | self.readers[-1].get(*reference) |
| 368 | self.assertTrue(self._is_evictable("image")) |
| 369 | |
| 370 | def test_touch_protects_item_until_released(self): |
| 371 | reference = self._reference("image") |
| 372 | for reader in self.readers: |
| 373 | reader.get(*reference) |
| 374 | |
| 375 | self.writer.touch("image") |
| 376 | self.assertFalse(self._is_evictable("image")) |
| 377 | self.writer.release_touches() |
| 378 | self.assertTrue(self._is_evictable("image")) |
| 379 | |
| 380 | |
| 381 | # Reader process function |
| 382 | def reader_process(process_id, storage_handle, items_to_read): |
| 383 | """Reader process that connects to existing shared memory and reads data.""" |
| 384 | reader_storage = SingleWriterShmObjectStorage.create_from_handle(storage_handle) |
| 385 | |
| 386 | print(f"Reader {process_id} started") |
| 387 | |
| 388 | errors = [] |
| 389 | |
| 390 | for key, original_value, address, monotonic_id, signature in items_to_read: |
| 391 | time.sleep(random.random() / 100) |
| 392 | try: |
| 393 | # Read data from shared memory |
| 394 | retrieved_value = reader_storage.get(address, monotonic_id, signature, key) |
| 395 | |
| 396 | # Verify data integrity |
| 397 | assert retrieved_value == original_value |
| 398 | print(f"Reader {process_id} retrieved {key}: {retrieved_value}") |
| 399 | except Exception as e: |
| 400 | errors.append((key, str(e), type(e).__name__)) |
| 401 | |
| 402 | |
| 403 | def run_multiprocess_example(): |
| 404 | """Run a minimal working example with real shared memory.""" |
| 405 | print("=== Minimal Object Storage Example ===") |
| 406 | |
| 407 | try: |
| 408 | # Create storage instance |
| 409 | ring_buffer = SingleWriterShmRingBuffer( |
| 410 | data_buffer_size=1024 * 100, |
| 411 | create=True, # 10 MB buffer |
| 412 | ) |
| 413 | storage = SingleWriterShmObjectStorage( |
| 414 | max_object_size=1024, |
| 415 | n_readers=3, |
| 416 | ring_buffer=ring_buffer, |
| 417 | serde_class=MsgpackSerde, |
| 418 | reader_lock=Lock(), |
| 419 | ) |
| 420 | |
| 421 | print(f"Created storage (writer: {storage.is_writer})") |
| 422 | |
| 423 | # Test basic data types |
| 424 | test_data = [ |
| 425 | ("user_data", {"name": "Alice", "age": 30, "scores": [95, 87, 92]}), |
| 426 | ("simple_string", "Hello, World!"), |
| 427 | ("number", 42), |
| 428 | ("list_data", [1, 2, 3, "four", 5.0]), |
| 429 | ] |
| 430 | |
| 431 | stored_items = [] |
| 432 | |
| 433 | # Store all data |
| 434 | for key, value in test_data: |
| 435 | print(f"Storing {key}: {value}") |
| 436 | address, monotonic_id = storage.put(key, value) |
| 437 | signature = storage.get_signature(key) |
| 438 | stored_items.append((key, value, address, monotonic_id, signature)) |
| 439 | print(f" -> Stored at address {address}, ID {monotonic_id}") |
| 440 | |
| 441 | print("\n--- Retrieving Data ---") |
| 442 | processes = [] |
| 443 | handle = storage.handle() |
| 444 | # initialize lock for reader processes |
| 445 | handle.reader_lock = Lock() |
| 446 | for i in range(storage.n_readers): |
| 447 | p = multiprocessing.Process( |
| 448 | target=reader_process, args=(i, handle, stored_items) |
| 449 | ) |
| 450 | processes.append(p) |
| 451 | p.start() |
| 452 | |
| 453 | for p in processes: |
| 454 | p.join(timeout=10) |
| 455 | if p.is_alive(): |
| 456 | p.terminate() |
| 457 | p.join() |
| 458 | |
| 459 | except Exception as e: |
| 460 | print(f"Error in minimal example: {e}") |
| 461 | traceback.print_exc() |
| 462 | |
| 463 | |
| 464 | if __name__ == "__main__": |
| 465 | # Run the minimal example first |
| 466 | run_multiprocess_example() |
| 467 | print("\n" + "=" * 50 + "\n") |
| 468 | |
| 469 | # Run the test suite |
| 470 | print("Running comprehensive test suite...") |
| 471 | unittest.main(verbosity=2, exit=False) |