已合并
【API】Add custom validation cases for HealthCheckServer APIs #34493
Jinfan Liu创建于 4月27日
【API】Add custom validation cases for HealthCheckServer APIs #34493
已合并
共 1 个文件变更+95-0
| @@ -0,0 +1,95 @@ | |||
| 1 | +""" | ||
| 2 | +Add validation cases for torch.distributed.elastic.agent.server.health_check_server APIs. | ||
| 3 | + | ||
| 4 | +1. PyTorch community tests do not cover HealthCheckServer APIs, so this file is added. | ||
| 5 | +2. This file validates : | ||
| 6 | +torch.distributed.elastic.agent.server.health_check_server.HealthCheckServer | ||
| 7 | +torch.distributed.elastic.agent.server.health_check_server.HealthCheckServer.start | ||
| 8 | +torch.distributed.elastic.agent.server.health_check_server.HealthCheckServer.stop | ||
| 9 | +torch.distributed.elastic.agent.server.health_check_server.create_healthcheck_server | ||
| 10 | +(extendable) | ||
| 11 | +""" | ||
| 12 | + | ||
| 13 | +import logging | ||
| 14 | +from unittest.mock import MagicMock | ||
| 15 | + | ||
| 16 | +from torch.distributed.elastic.agent.server.health_check_server import ( | ||
| 17 | + HealthCheckServer, | ||
| 18 | + create_healthcheck_server, | ||
| 19 | +) | ||
| 20 | +from torch_npu.testing.testcase import TestCase, run_tests | ||
| 21 | + | ||
| 22 | +LOGGER_NAME = "torch.distributed.elastic.agent.server.health_check_server" | ||
| 23 | + | ||
| 24 | + | ||
| 25 | +class TestHealthCheckServer(TestCase): | ||
| 26 | + """Unit tests for HealthCheckServer and factory method create_healthcheck_server.""" | ||
| 27 | + | ||
| 28 | + def test_init(self): | ||
| 29 | + """Test HealthCheckServer initialization, verify attributes are assigned correctly | ||
| 30 | + and callback is not called prematurely. | ||
| 31 | + """ | ||
| 32 | + alive_callback = MagicMock(return_value=123) | ||
| 33 | + server = HealthCheckServer(alive_callback, 0, 30) | ||
| 34 | + | ||
| 35 | + self.assertIs(server._alive_callback, alive_callback) | ||
| 36 | + self.assertEqual(server._port, 0) | ||
| 37 | + self.assertEqual(server._timeout, 30) | ||
| 38 | + alive_callback.assert_not_called() | ||
| 39 | + | ||
| 40 | + def test_create_healthcheck_server(self): | ||
| 41 | + """Test the factory function create_healthcheck_server, returns a valid HealthCheckServer instance.""" | ||
| 42 | + alive_callback = MagicMock(return_value=123) | ||
| 43 | + server = create_healthcheck_server(alive_callback, 0, 45) | ||
| 44 | + | ||
| 45 | + self.assertIsInstance(server, HealthCheckServer) | ||
| 46 | + self.assertIs(server._alive_callback, alive_callback) | ||
| 47 | + self.assertEqual(server._port, 0) | ||
| 48 | + self.assertEqual(server._timeout, 45) | ||
| 49 | + alive_callback.assert_not_called() | ||
| 50 | + | ||
| 51 | + def test_start_logs_warning(self): | ||
| 52 | + """Test start() outputs expected WARNING log (current noop stub behavior).""" | ||
| 53 | + server = HealthCheckServer(lambda: 0, 0, 30) | ||
| 54 | + | ||
| 55 | + with self.assertLogs(LOGGER_NAME, level=logging.WARNING) as captured: | ||
| 56 | + server.start() | ||
| 57 | + | ||
| 58 | + self.assertEqual(len(captured.records), 1) | ||
| 59 | + self.assertEqual( | ||
| 60 | + captured.records[0].getMessage(), | ||
| 61 | + "No health check server started", | ||
| 62 | + ) | ||
| 63 | + | ||
| 64 | + def test_stop_logs_info(self): | ||
| 65 | + """Test stop() outputs expected INFO log (current noop stub behavior).""" | ||
| 66 | + server = HealthCheckServer(lambda: 0, 0, 30) | ||
| 67 | + | ||
| 68 | + with self.assertLogs(LOGGER_NAME, level=logging.INFO) as captured: | ||
| 69 | + server.stop() | ||
| 70 | + | ||
| 71 | + self.assertEqual(len(captured.records), 1) | ||
| 72 | + self.assertEqual( | ||
| 73 | + captured.records[0].getMessage(), | ||
| 74 | + "Stopping noop health check server.", | ||
| 75 | + ) | ||
| 76 | + | ||
| 77 | + def test_start_stop_lifecycle_safe(self): | ||
| 78 | + """Verify start -> stop full lifecycle runs safely without exceptions.""" | ||
| 79 | + server = HealthCheckServer(lambda: True, 0, 5) | ||
| 80 | + server.start() | ||
| 81 | + server.stop() | ||
| 82 | + # Passes if no exception is thrown, no additional assertions needed | ||
| 83 | + | ||
| 84 | + def test_stop_idempotent(self): | ||
| 85 | + """Verify stop() is idempotent: repeated calls do not throw exceptions.""" | ||
| 86 | + server = HealthCheckServer(lambda: True, 0, 5) | ||
| 87 | + server.start() | ||
| 88 | + server.stop() | ||
| 89 | + server.stop() | ||
| 90 | + server.stop() | ||
| 91 | + # Passes if no exception is thrown, no additional assertions needed | ||
| 92 | + | ||
| 93 | + | ||
| 94 | +if __name__ == "__main__": | ||
| 95 | + run_tests() | ||