Method fetch_and_verify

backend/python/mlx/test_mlx_cache.py:394–412

(thread_id)

            cache.insert_cache("model", [i], [f"original_{i}"])

        def fetch_and_verify(thread_id):
            try:
                for _ in range(50):
                    token_id = thread_id % 20
                    result, remaining = cache.fetch_nearest_cache("model", [token_id])

                    if result is not None:
                        # Verify data integrity
                        expected_prefix = f"original_{token_id}"
                        if not str(result[0]).startswith("original_"):
                            with lock:
                                errors.append(f"Corrupted data: {result}")

                        # Re-insert to keep cache populated
                        cache.insert_cache("model", [token_id], result)

            except Exception as e:
                with lock:
                    errors.append(str(e))

        with concurrent.futures.ThreadPoolExecutor(max_workers=10) as executor:
            futures = [executor.submit(fetch_and_verify, tid) for tid in range(10)]