move slow tests to integration (#1063)

paul-nechifor · web-flow · commit 35acee905755 · 2026-01-20T11:27:29.000+02:00
* move slow tests to integration

* run integration tests too

* remove fast

* fix dimos/control/test_control.py
diff --git a/.github/workflows/docker.yml b/.github/workflows/docker.yml
@@ -205,6 +205,21 @@ jobs:
       cmd: "pytest -m lcm"
       dev-image: dev:${{ (needs.check-changes.outputs.python == 'true' || needs.check-changes.outputs.dev == 'true') && needs.dev.result == 'success' && needs.check-changes.outputs.branch-tag || 'dev' }}
 
+  run-integration-tests:
+    needs: [check-changes, dev]
+    if: always()
+    uses: ./.github/workflows/tests.yml
+    secrets: inherit
+    with:
+      should-run: ${{
+        needs.check-changes.result == 'success' &&
+        ((needs.dev.result == 'success') ||
+        (needs.dev.result == 'skipped' &&
+        needs.check-changes.outputs.tests == 'true'))
+        }}
+      cmd: "pytest -m integration"
+      dev-image: dev:${{ (needs.check-changes.outputs.python == 'true' || needs.check-changes.outputs.dev == 'true') && needs.dev.result == 'success' && needs.check-changes.outputs.branch-tag || 'dev' }}
+
   run-mypy:
     needs: [check-changes, ros-dev]
     if: always()
diff --git a/bin/pytest-slow b/bin/pytest-slow
@@ -0,0 +1,6 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+
+. .venv/bin/activate
+exec pytest "$@" -m 'not (tool or cuda or gpu or module or temporal)' dimos
diff --git a/dimos/agents/skills/test_navigation.py b/dimos/agents/skills/test_navigation.py
@@ -12,12 +12,12 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import pytest
 
 from dimos.msgs.geometry_msgs import PoseStamped, Vector3
 from dimos.utils.transform_utils import euler_to_quaternion
 
 
-# @pytest.mark.skip
 def test_stop_movement(create_navigation_agent, navigation_skill_container, mocker) -> None:
     cancel_goal_mock = mocker.Mock()
     stop_exploration_mock = mocker.Mock()
@@ -35,6 +35,7 @@ def test_stop_movement(create_navigation_agent, navigation_skill_container, mock
     stop_exploration_mock.assert_called_once_with()
 
 
+@pytest.mark.integration
 def test_take_a_look_around(create_navigation_agent, navigation_skill_container, mocker) -> None:
     explore_mock = mocker.Mock()
     is_exploration_active_mock = mocker.Mock()
diff --git a/dimos/agents/test_agent_fake.py b/dimos/agents/test_agent_fake.py
@@ -12,13 +12,17 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import pytest
 
+
+@pytest.mark.integration
 def test_what_is_your_name(create_potato_agent) -> None:
     agent = create_potato_agent(fixture="test_what_is_your_name.json")
     response = agent.query("hi there, please tell me what's your name?")
     assert "Mr. Potato" in response
 
 
+@pytest.mark.integration
 def test_how_much_is_124181112_plus_124124(create_potato_agent) -> None:
     agent = create_potato_agent(fixture="test_how_much_is_124181112_plus_124124.json")
 
@@ -29,6 +33,7 @@ def test_how_much_is_124181112_plus_124124(create_potato_agent) -> None:
     assert "999000000" in response.replace(",", "")
 
 
+@pytest.mark.integration
 def test_what_do_you_see_in_this_picture(create_potato_agent) -> None:
     agent = create_potato_agent(fixture="test_what_do_you_see_in_this_picture.json")
 
diff --git a/dimos/agents/test_mock_agent.py b/dimos/agents/test_mock_agent.py
@@ -30,6 +30,7 @@
 from dimos.robot.unitree_webrtc.type.lidar import LidarMessage
 
 
+@pytest.mark.integration
 def test_tool_call() -> None:
     """Test agent initialization and tool call execution."""
     # Create a fake model that will respond with tool calls
@@ -74,6 +75,7 @@ def test_tool_call() -> None:
     agent.stop()
 
 
+@pytest.mark.integration
 def test_image_tool_call() -> None:
     """Test agent with image tool call execution."""
     dimos = start(2)
diff --git a/dimos/control/test_control.py b/dimos/control/test_control.py
@@ -36,6 +36,7 @@
     TrajectoryState,
 )
 from dimos.control.tick_loop import TickLoop
+from dimos.hardware.manipulators.spec import ManipulatorBackend
 from dimos.msgs.trajectory_msgs import JointTrajectory, TrajectoryPoint
 
 # =============================================================================
@@ -46,9 +47,8 @@
 @pytest.fixture
 def mock_backend():
     """Create a mock manipulator backend."""
-    backend = MagicMock()
+    backend = MagicMock(spec=ManipulatorBackend)
     backend.get_dof.return_value = 6
-    backend.get_joint_names.return_value = [f"joint{i + 1}" for i in range(6)]
     backend.read_joint_positions.return_value = [0.0] * 6
     backend.read_joint_velocities.return_value = [0.0] * 6
     backend.read_joint_efforts.return_value = [0.0] * 6
diff --git a/dimos/core/test_blueprints.py b/dimos/core/test_blueprints.py
@@ -162,6 +162,7 @@ def test_global_config() -> None:
     assert blueprint_set.global_config_overrides["option2"] == 42
 
 
+@pytest.mark.integration
 def test_build_happy_path() -> None:
     pubsub.lcm.autoconf()
 
@@ -272,6 +273,7 @@ class Module3(Module):
     blueprint_set_remapped._verify_no_name_conflicts()
 
 
+@pytest.mark.integration
 def test_remapping() -> None:
     """Test that remapping connections works correctly."""
     pubsub.lcm.autoconf()
@@ -351,6 +353,7 @@ def test_future_annotations_support() -> None:
     )
 
 
+@pytest.mark.integration
 def test_future_annotations_autoconnect() -> None:
     """Test that autoconnect works with modules using `from __future__ import annotations`."""
 
diff --git a/dimos/e2e_tests/test_control_orchestrator.py b/dimos/e2e_tests/test_control_orchestrator.py
@@ -33,6 +33,7 @@
 
 
 @pytest.mark.skipif(bool(os.getenv("CI")), reason="LCM doesn't work in CI.")
+@pytest.mark.e2e
 class TestControlOrchestratorE2E:
     """End-to-end tests for ControlOrchestrator."""
 
diff --git a/dimos/e2e_tests/test_dimos_cli_e2e.py b/dimos/e2e_tests/test_dimos_cli_e2e.py
@@ -19,6 +19,7 @@
 
 @pytest.mark.skipif(bool(os.getenv("CI")), reason="LCM spy doesn't work in CI.")
 @pytest.mark.skipif(not os.getenv("OPENAI_API_KEY"), reason="OPENAI_API_KEY not set.")
+@pytest.mark.e2e
 def test_dimos_skills(lcm_spy, start_blueprint, human_input) -> None:
     lcm_spy.save_topic("/rpc/DemoCalculatorSkill/set_AgentSpec_register_skills/res")
     lcm_spy.save_topic("/rpc/HumanInput/start/res")
diff --git a/dimos/mapping/occupancy/test_extrude_occupancy.py b/dimos/mapping/occupancy/test_extrude_occupancy.py
@@ -12,10 +12,13 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import pytest
+
 from dimos.mapping.occupancy.extrude_occupancy import generate_mujoco_scene
 from dimos.utils.data import get_data
 
 
+@pytest.mark.integration
 def test_generate_mujoco_scene(occupancy) -> None:
     with open(get_data("expected_occupancy_scene.xml")) as f:
         expected = f.read()
diff --git a/dimos/msgs/sensor_msgs/image_impls/test_image_backends.py b/dimos/msgs/sensor_msgs/image_impls/test_image_backends.py
@@ -356,6 +356,7 @@ def test_perf_resize(alloc_timer) -> None:
         print(f"resize (avg per call) cpu={cpu_t:.6f}s")
 
 
+@pytest.mark.integration
 def test_perf_sharpness(alloc_timer) -> None:
     """Test sharpness performance with NumpyImage always, add CudaImage when available."""
     arr = _prepare_image(ImageFormat.BGR, (480, 640, 3))
diff --git a/dimos/perception/experimental/temporal_memory/test_temporal_memory_module.py b/dimos/perception/experimental/temporal_memory/test_temporal_memory_module.py
@@ -81,6 +81,7 @@ def stop(self) -> None:
 
 @pytest.mark.skipif(bool(os.getenv("CI")), reason="LCM replay + dataset not CI-safe.")
 @pytest.mark.skipif(not os.getenv("OPENAI_API_KEY"), reason="OPENAI_API_KEY not set.")
+@pytest.mark.temporal  # TODO: This never finishes for me.
 class TestTemporalMemoryModule:
     @pytest.fixture(scope="function")
     def temp_dir(self):
diff --git a/dimos/protocol/mcp/test_mcp_module.py b/dimos/protocol/mcp/test_mcp_module.py
@@ -133,6 +133,7 @@ async def wait_for_updates(self) -> bool:
     assert "Error:" in response["result"]["content"][0]["text"]
 
 
+@pytest.mark.integration
 def test_mcp_end_to_end_lcm_bridge() -> None:
     try:
         import lcm  # type: ignore[import-untyped]
diff --git a/dimos/protocol/rpc/test_spec.py b/dimos/protocol/rpc/test_spec.py
@@ -293,6 +293,7 @@ def callback(val) -> None:
             unsub_server()
 
 
+@pytest.mark.integration
 @pytest.mark.parametrize("rpc_context, impl_name", testdata)
 def test_timeout(rpc_context, impl_name: str) -> None:
     """Test that RPC calls properly timeout."""
diff --git a/dimos/protocol/skill/test_coordinator.py b/dimos/protocol/skill/test_coordinator.py
@@ -96,6 +96,7 @@ def take_photo(self) -> Image:
         return img
 
 
+@pytest.mark.integration
 @pytest.mark.asyncio  # type: ignore[untyped-decorator]
 async def test_coordinator_parallel_calls() -> None:
     container = SkillContainerTest()
@@ -136,6 +137,7 @@ async def test_coordinator_parallel_calls() -> None:
     skillCoordinator.stop()
 
 
+@pytest.mark.integration
 @pytest.mark.asyncio  # type: ignore[untyped-decorator]
 async def test_coordinator_generator() -> None:
     container = SkillContainerTest()
diff --git a/dimos/robot/drone/test_drone.py b/dimos/robot/drone/test_drone.py
@@ -24,6 +24,7 @@
 from unittest.mock import MagicMock, patch
 
 import numpy as np
+import pytest
 
 from dimos.msgs.geometry_msgs import PoseStamped, Quaternion, Vector3
 from dimos.msgs.sensor_msgs import Image, ImageFormat
diff --git a/dimos/robot/test_all_blueprints.py b/dimos/robot/test_all_blueprints.py
@@ -24,6 +24,7 @@
 }
 
 
+@pytest.mark.integration
 @pytest.mark.parametrize("blueprint_name", all_blueprints.keys())
 def test_all_blueprints_are_valid(blueprint_name: str) -> None:
     """Test that all blueprints in all_blueprints are valid ModuleBlueprintSet instances."""
diff --git a/dimos/types/test_weaklist.py b/dimos/types/test_weaklist.py
@@ -54,6 +54,7 @@ def test_weaklist_basic_operations() -> None:
     assert SampleObject(4) not in wl
 
 
+@pytest.mark.integration
 def test_weaklist_auto_removal() -> None:
     """Test that objects are automatically removed when garbage collected."""
     wl = WeakList()
@@ -136,6 +137,7 @@ def test_weaklist_clear() -> None:
     assert obj1 not in wl
 
 
+@pytest.mark.integration
 def test_weaklist_iteration_during_modification() -> None:
     """Test that iteration works even if objects are deleted during iteration."""
     wl = WeakList()
diff --git a/dimos/utils/test_reactive.py b/dimos/utils/test_reactive.py
@@ -82,6 +82,7 @@ def _dispose() -> None:
     return proxy
 
 
+@pytest.mark.integration
 def test_backpressure_handling() -> None:
     # Create a dedicated scheduler for this test to avoid thread leaks
     test_scheduler = ThreadPoolScheduler(max_workers=8)
@@ -141,6 +142,7 @@ def test_backpressure_handling() -> None:
         test_scheduler.executor.shutdown(wait=True)
 
 
+@pytest.mark.integration
 def test_getter_streaming_blocking() -> None:
     source = dispose_spy(
         rx.interval(0.2).pipe(ops.map(lambda i: np.array([i, i + 1, i + 2])), ops.take(50))
@@ -175,6 +177,7 @@ def test_getter_streaming_blocking_timeout() -> None:
     assert source.is_disposed()
 
 
+@pytest.mark.integration
 def test_getter_streaming_nonblocking() -> None:
     source = dispose_spy(rx.interval(0.2).pipe(ops.take(50)))
 
diff --git a/pyproject.toml b/pyproject.toml
@@ -349,12 +349,15 @@ markers = [
     "lcm: tests that run actual LCM bus (can't execute in CI)",
     "module: tests that need to run directly as modules",
     "gpu: tests that require GPU",
-    "tofix: temporarily disabled test"
+    "tofix: temporarily disabled test",
+    "e2e: end to end tests",
+    "temporal: not working for me",
+    "integration: slower integration tests",
 ]
 env = [
   "GOOGLE_MAPS_API_KEY=AIzafake_google_key"
 ]
-addopts = "-v -p no:warnings -ra --color=yes -m 'not vis and not benchmark and not exclude and not tool and not needsdata and not lcm and not ros and not heavy and not gpu and not module and not tofix'"
+addopts = "-v -p no:warnings -ra --color=yes -m 'not (vis or benchmark or exclude or tool or needsdata or lcm or ros or heavy or gpu or module or tofix or e2e or temporal or integration)'"
 asyncio_mode = "auto"
 asyncio_default_fixture_loop_scope = "function"
 
