【Hackathon 9th No.22】add unit tests for share_external_data (#3744)

Author: ooooo-create

tests/operators/test_share_external_data.py

@@ -0,0 +1,131 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import multiprocessing as mp
+import os
+import queue
+import unittest
+from multiprocessing import Process, Queue
+
+import numpy as np
+import paddle
+
+from fastdeploy.model_executor.ops.gpu import set_data_ipc, share_external_data
+
+
+def _create_test_tensor(shape, dtype):
+    if "float" in str(dtype):
+        return paddle.rand(shape=shape, dtype=dtype)
+    elif "int" in str(dtype):
+        return paddle.randint(-100, 100, shape=shape, dtype=dtype)
+    elif "bool" in str(dtype):
+        return paddle.rand(shape=shape, dtype=dtype) > 0.5
+
+
+def _producer_proc(shm_name, shape, dtype, ready_q, done_q, error_q):
+    # Create shared memory
+    try:
+        paddle.device.set_device("gpu:0")
+        t = _create_test_tensor(shape, dtype)
+        set_data_ipc(t, shm_name)
+        ready_q.put(("ready", t.numpy().tolist()))
+        _ = done_q.get(timeout=20)
+    except Exception as e:
+        error_q.put(("producer_error", str(e)))
+
+
+def _consumer_proc(shm_name, shape, dtype, result_q, error_q):
+    # Shard data
+    try:
+        paddle.device.set_device("gpu:0")
+        dummy = paddle.zeros(shape, dtype=dtype)
+        shared = share_external_data(dummy, shm_name, shape)
+        result_q.put(("ok", shared.numpy().tolist()))
+    except Exception as e:
+        error_q.put(("consumer_error", str(e)))
+
+
+# Use spawn to avoid forking CUDA contexts
+try:
+    mp.set_start_method("spawn", force=True)
+except RuntimeError:
+    pass
+
+
+class TestShareExternalData(unittest.TestCase):
+    def setUp(self):
+        paddle.seed(2024)
+        np.random.seed(42)
+
+        if not paddle.device.is_compiled_with_cuda():
+            self.skipTest("CUDA not available, skipping GPU tests")
+
+        # Set device to GPU
+        paddle.device.set_device("gpu:0")
+
+        self.test_shape = [4, 8]
+        self.dtype = paddle.float32
+        self.shm_prefix = f"test_share_external_{os.getpid()}"
+
+    def _run_minimal_cross_process(self):
+        ready_q = Queue()
+        result_q = Queue()
+        error_q = Queue()
+        done_q = Queue()
+
+        p = Process(
+            target=_producer_proc, args=(self.shm_prefix, self.test_shape, self.dtype, ready_q, done_q, error_q)
+        )
+        p.start()
+
+        # wait producer ready
+        try:
+            status, original_data = ready_q.get(timeout=20)
+            self.assertEqual(status, "ready")
+        except Exception:
+            p.terminate()
+            self.fail("Producer did not become ready in time")
+
+        c = Process(target=_consumer_proc, args=(self.shm_prefix, self.test_shape, self.dtype, result_q, error_q))
+        c.start()
+        c.join(timeout=30)
+
+        # signal producer to exit now
+        done_q.put("done")
+        p.join(timeout=30)
+
+        # check errors first (non-blocking)
+        errors = []
+        try:
+            while True:
+                errors.append(error_q.get_nowait())
+        except queue.Empty:
+            pass
+        self.assertFalse(errors, f"Errors occurred: {errors}")
+
+        # verify data
+        self.assertFalse(result_q.empty(), "No result from consumer")
+        status, shared_data = result_q.get()
+        self.assertEqual(status, "ok")
+        np.testing.assert_allclose(np.array(original_data), np.array(shared_data), rtol=1e-5)
+
+    def test_producer_consumer_processes(self):
+        self._run_minimal_cross_process()
+
+    def tearDown(self):
+        paddle.device.cuda.empty_cache()
+
+
+if __name__ == "__main__":
+    unittest.main()