Merge branch 'main' into features/ui/add-tooltip-to-metric-summary

Author: DaltheCow

pyproject.toml

@@ -160,7 +160,7 @@ profile = "black"
 
 
 [tool.mypy]
-files = ["src/guidellm", "tests"]
+files = ["src/guidellm"]
 python_version = '3.10'
 warn_redundant_casts = true
 warn_unused_ignores = false

src/guidellm/scheduler/__init__.py

@@ -55,25 +55,20 @@
     "BackendInterface",
     "BackendT",
     "ConcurrentStrategy",
-    "ConstantRateRequestTimings",
     "Constraint",
     "ConstraintInitializer",
     "ConstraintsInitializerFactory",
     "Environment",
-    "LastCompletionRequestTimings",
     "MaxDurationConstraint",
     "MaxErrorRateConstraint",
     "MaxErrorsConstraint",
     "MaxGlobalErrorRateConstraint",
     "MaxNumberConstraint",
     "MultiTurnRequestT",
-    "NoDelayRequestTimings",
     "NonDistributedEnvironment",
-    "PoissonRateRequestTimings",
     "PydanticConstraintInitializer",
     "RequestT",
     "ResponseT",
-    "ScheduledRequestTimings",
     "Scheduler",
     "SchedulerMessagingPydanticRegistry",
     "SchedulerState",

tests/integration/scheduler/test_scheduler.py

@@ -16,12 +16,12 @@
     Environment,
     MaxNumberConstraint,
     NonDistributedEnvironment,
-    ScheduledRequestInfo,
     Scheduler,
     SchedulerState,
     SchedulingStrategy,
     SynchronousStrategy,
 )
+from guidellm.schemas import RequestInfo
 
 
 def async_timeout(delay: float):
@@ -91,6 +91,7 @@ async def resolve(self, request: MockRequest, request_info, request_history):
         yield f"response_for_{request.payload}", request_info
 
 
+@pytest.mark.xfail(reason="old and broken", run=False)
 @pytest.mark.smoke
 @pytest.mark.asyncio
 @async_timeout(10.0)
@@ -127,12 +128,13 @@ async def test_scheduler_run_integration(
         requests=[MockRequest(payload=f"req_{ind}") for ind in range(num_requests)],
         backend=MockBackend(),
         strategy=strategy,
+        startup_duration=0.1,
         env=env,
         **constraints,
     ):
         assert req is not None
         assert isinstance(req, MockRequest)
-        assert isinstance(info, ScheduledRequestInfo)
+        assert isinstance(info, RequestInfo)
         assert info.status != "cancelled"
         assert isinstance(state, SchedulerState)
         if info.status == "completed":

tests/integration/scheduler/test_worker_group.py

@@ -27,13 +27,13 @@
     MaxErrorsConstraint,
     MaxGlobalErrorRateConstraint,
     MaxNumberConstraint,
-    MeasuredRequestTimings,
     SynchronousStrategy,
     ThroughputStrategy,
     WorkerProcessGroup,
 )
 from guidellm.scheduler.constraints import ConstraintInitializer
 from guidellm.scheduler.strategies import SchedulingStrategy
+from guidellm.schemas import RequestTimings
 
 
 def async_timeout(delay):
@@ -47,7 +47,7 @@ async def new_func(*args, **kwargs):
     return decorator
 
 
-class MockRequestTimings(MeasuredRequestTimings):
+class MockRequestTimings(RequestTimings):
     """Mock timing implementation for integration testing."""
 
 
@@ -102,6 +102,7 @@ async def resolve(self, request, request_info, request_history):
 
 
 class TestWorkerGroup:
+    @pytest.mark.xfail(reason="old and broken", run=False)
     @pytest.mark.smoke
     @pytest.mark.asyncio
     @async_timeout(5)
@@ -138,10 +139,10 @@ async def test_lifecycle(
             backend=backend,
             requests=requests,
             strategy=strategy,
+            startup_duration=0.1,
             constraints={
                 key: init.create_constraint() for key, init in constraints_inits.items()
             },
-            infinite_requests=False,
         )
 
         try:

tests/unit/backends/test_backend.py

@@ -11,11 +11,11 @@
 import pytest
 
 from guidellm.backends.backend import Backend, BackendType
-from guidellm.scheduler import BackendInterface, ScheduledRequestInfo
-from guidellm.schemas.response import (
+from guidellm.schemas import (
     GenerationRequest,
-    GenerationRequestTimings,
+    RequestInfo,
 )
+from guidellm.schemas.request import GenerationRequestArguments
 from guidellm.utils import RegistryMixin
 from tests.unit.testing_utils import async_timeout
 
@@ -41,6 +41,7 @@ def valid_instances(self, request):
         constructor_args = request.param
 
         class TestBackend(Backend):
+            @property
             def info(self) -> dict[str, Any]:
                 return {"type": self.type_}
 
@@ -68,7 +69,11 @@ async def default_model(self) -> str | None:
     def test_class_signatures(self):
         """Test Backend inheritance and type relationships."""
         assert issubclass(Backend, RegistryMixin)
-        assert isinstance(Backend, BackendInterface)
+        # Check that Backend implements BackendInterface methods
+        assert hasattr(Backend, "resolve")
+        assert hasattr(Backend, "process_startup")
+        assert hasattr(Backend, "process_shutdown")
+        assert hasattr(Backend, "validate")
         assert hasattr(Backend, "create")
         assert hasattr(Backend, "register")
         assert hasattr(Backend, "get_registered_object")
@@ -100,6 +105,7 @@ def test_invalid_initialization_values(self, field, value):
         """Test Backend with invalid field values."""
 
         class TestBackend(Backend):
+            @property
             def info(self) -> dict[str, Any]:
                 return {}
 
@@ -147,15 +153,10 @@ async def test_interface_compatibility(self, valid_instances):
         instance, _ = valid_instances
 
         # Test that Backend uses the correct generic types
-        request = GenerationRequest(content="test")
-        request_info = ScheduledRequestInfo(
-            request_id="test-id",
-            status="pending",
-            scheduler_node_id=1,
-            scheduler_process_id=1,
-            scheduler_start_time=123.0,
-            request_timings=GenerationRequestTimings(),
+        request = GenerationRequest(
+            request_type="text_completions", arguments=GenerationRequestArguments()
         )
+        request_info = RequestInfo(request_id="test-id")
 
         # Test resolve method
         async for response, info in instance.resolve(request, request_info):