langwatch-scenario 0.4.0__py3-none-any.whl → 0.7.1__py3-none-any.whl

scenario/scenario_executor.py

@@ -12,37 +12,55 @@ from typing import (
     Callable,
     Dict,
     List,
-    Any,
     Optional,
     Set,
     Tuple,
     Union,
+    TypedDict,
 )
 import time
+import warnings
 import termcolor
 import asyncio
 import concurrent.futures
 
 from scenario.config import ScenarioConfig
-from scenario.utils import (
-    await_if_awaitable,
-    check_valid_return_type,
+from scenario._utils import (
     convert_agent_return_types_to_openai_messages,
+    check_valid_return_type,
     print_openai_messages,
     show_spinner,
+    await_if_awaitable,
+    get_or_create_batch_run_id,
+    generate_scenario_run_id,
 )
 from openai.types.chat import (
     ChatCompletionMessageParam,
     ChatCompletionUserMessageParam,
+    ChatCompletionAssistantMessageParam,
 )
 
 from .types import AgentInput, AgentRole, ScenarioResult, ScriptStep
-from .error_messages import agent_response_not_awaitable
+from ._error_messages import agent_response_not_awaitable
 from .cache import context_scenario
 from .agent_adapter import AgentAdapter
 from .script import proceed
 from pksuid import PKSUID
 from .scenario_state import ScenarioState
+from ._events import (
+    ScenarioEventBus,
+    ScenarioEvent,
+    ScenarioRunStartedEvent,
+    ScenarioMessageSnapshotEvent,
+    ScenarioRunFinishedEvent,
+    ScenarioRunStartedEventMetadata,
+    ScenarioRunFinishedEventResults,
+    ScenarioRunFinishedEventVerdict,
+    ScenarioRunFinishedEventStatus,
+    convert_messages_to_api_client_messages,
+)
+from rx.subject.subject import Subject
+from rx.core.observable.observable import Observable
 
 
 class ScenarioExecutor:
@@ -66,41 +84,8 @@ class ScenarioExecutor:
         agents: List of agent adapters participating in the scenario
         script: Optional list of script steps to control scenario flow
         config: Configuration settings for execution behavior
-
-    Example:
-        ```python
-        # Direct instantiation (less common)
-        executor = ScenarioExecutor(
-            name="weather query test",
-            description="User asks about weather, agent should provide helpful response",
-            agents=[
-                weather_agent,
-                scenario.UserSimulatorAgent(),
-                scenario.JudgeAgent(criteria=["Agent provides helpful weather info"])
-            ],
-            max_turns=10,
-            verbose=True
-        )
-        result = await executor._run()
-
-        # Preferred high-level API
-        result = await scenario.run(
-            name="weather query test",
-            description="User asks about weather, agent should provide helpful response",
-            agents=[
-                weather_agent,
-                scenario.UserSimulatorAgent(),
-                scenario.JudgeAgent(criteria=["Agent provides helpful weather info"])
-            ]
-        )
-        ```
-
-    Note:
-        - Scenarios run in isolated thread pools to support parallel execution
-        - All agent interactions are cached when cache_key is configured
-        - Debug mode allows step-by-step execution with user intervention
-        - Results include detailed timing information and conversation history
     """
+
     name: str
     description: str
     agents: List[AgentAdapter]
@@ -115,6 +100,11 @@ class ScenarioExecutor:
     _pending_roles_on_turn: List[AgentRole] = []
     _pending_agents_on_turn: Set[AgentAdapter] = set()
     _agent_times: Dict[int, float] = {}
+    _events: Subject
+
+    event_bus: ScenarioEventBus
+
+    batch_run_id: str
 
     def __init__(
         self,
@@ -127,6 +117,7 @@ class ScenarioExecutor:
         verbose: Optional[Union[bool, int]] = None,
         cache_key: Optional[str] = None,
         debug: Optional[bool] = None,
+        event_bus: Optional[ScenarioEventBus] = None,
     ):
         """
         Initialize a scenario executor.
@@ -147,26 +138,7 @@ class ScenarioExecutor:
                       Overrides global configuration for this scenario.
             debug: Whether to enable debug mode with step-by-step execution.
                   Overrides global configuration for this scenario.
-
-        Example:
-            ```python
-            executor = ScenarioExecutor(
-                name="customer service test",
-                description="Customer has a billing question and needs help",
-                agents=[
-                    customer_service_agent,
-                    scenario.UserSimulatorAgent(),
-                    scenario.JudgeAgent(criteria=[
-                        "Agent is polite and professional",
-                        "Agent addresses the billing question",
-                        "Agent provides clear next steps"
-                    ])
-                ],
-                max_turns=15,
-                verbose=True,
-                debug=False
-            )
-            ```
+            event_bus: Optional event bus that will subscribe to this executor's events
         """
         self.name = name
         self.description = description
@@ -183,115 +155,33 @@ class ScenarioExecutor:
 
         self.reset()
 
-    @classmethod
-    async def run(
-        cls,
-        name: str,
-        description: str,
-        agents: List[AgentAdapter] = [],
-        max_turns: Optional[int] = None,
-        verbose: Optional[Union[bool, int]] = None,
-        cache_key: Optional[str] = None,
-        debug: Optional[bool] = None,
-        script: Optional[List[ScriptStep]] = None,
-    ) -> ScenarioResult:
-        """
-        High-level interface for running a scenario test.
-
-        This is the main entry point for executing scenario tests. It creates a
-        ScenarioExecutor instance and runs it in an isolated thread pool to support
-        parallel execution and prevent blocking.
+        # Create executor's own event stream
+        self._events = Subject()
 
-        Args:
-            name: Human-readable name for the scenario
-            description: Detailed description of what the scenario tests
-            agents: List of agent adapters (agent under test, user simulator, judge)
-            max_turns: Maximum conversation turns before timeout (default: 10)
-            verbose: Show detailed output during execution
-            cache_key: Cache key for deterministic behavior
-            debug: Enable debug mode for step-by-step execution
-            script: Optional script steps to control scenario flow
+        # Create and configure event bus to subscribe to our events
+        self.event_bus = event_bus or ScenarioEventBus()
+        self.event_bus.subscribe_to_events(self._events)
 
-        Returns:
-            ScenarioResult containing the test outcome, conversation history,
-            success/failure status, and detailed reasoning
-
-        Example:
-            ```python
-            import scenario
+        self.batch_run_id = get_or_create_batch_run_id()
 
-            # Simple scenario with automatic flow
-            result = await scenario.run(
-                name="help request",
-                description="User asks for help with a technical problem",
-                agents=[
-                    my_agent,
-                    scenario.UserSimulatorAgent(),
-                    scenario.JudgeAgent(criteria=["Agent provides helpful response"])
-                ]
-            )
+    @property
+    def events(self) -> Observable:
+        """Expose event stream for subscribers like the event bus."""
+        return self._events
 
-            # Scripted scenario with custom evaluations
-            result = await scenario.run(
-                name="custom interaction",
-                description="Test specific conversation flow",
-                agents=[
-                    my_agent,
-                    scenario.UserSimulatorAgent(),
-                    scenario.JudgeAgent(criteria=["Agent provides helpful response"])
-                ],
-                script=[
-                    scenario.user("Hello"),
-                    scenario.agent(),
-                    custom_eval,
-                    scenario.succeed()
-                ]
-            )
+    def _emit_event(self, event: ScenarioEvent) -> None:
+        """
+        Emit a domain event to all subscribers.
 
-            # Results analysis
-            print(f"Test {'PASSED' if result.success else 'FAILED'}")
-            print(f"Reasoning: {result.reasoning}")
-            print(f"Conversation had {len(result.messages)} messages")
-            ```
+        This method publishes scenario events to the internal event stream,
+        which subscribers (like the event bus) can observe and react to.
+        The timestamp is automatically set to the current time.
 
-        Note:
-            - Runs in isolated thread pool to support parallel execution
-            - Blocks until scenario completes or times out
-            - All agent calls are automatically cached when cache_key is set
-            - Exception handling ensures clean resource cleanup
+        Args:
+            event: The scenario event to emit
         """
-        scenario = cls(
-            name=name,
-            description=description,
-            agents=agents,
-            max_turns=max_turns,
-            verbose=verbose,
-            cache_key=cache_key,
-            debug=debug,
-            script=script,
-        )
-
-        # We'll use a thread pool to run the execution logic, we
-        # require a separate thread because even though asyncio is
-        # being used throughout, any user code on the callback can
-        # be blocking, preventing them from running scenarios in parallel
-        with concurrent.futures.ThreadPoolExecutor() as executor:
-
-            def run_in_thread():
-                loop = asyncio.new_event_loop()
-                asyncio.set_event_loop(loop)
-
-                try:
-                    return loop.run_until_complete(scenario._run())
-                finally:
-                    loop.close()
-
-            # Run the function in the thread pool and await its result
-            # This converts the thread's execution into a Future that the current
-            # event loop can await without blocking
-            loop = asyncio.get_event_loop()
-            result = await loop.run_in_executor(executor, run_in_thread)
-            return result
+        event.timestamp = int(time.time() * 1000)
+        self._events.on_next(event)
 
     def reset(self):
         """
@@ -300,18 +190,6 @@ class ScenarioExecutor:
         This method reinitializes all internal state for a fresh scenario run,
         including conversation history, turn counters, and agent timing information.
         Called automatically during initialization and can be used to rerun scenarios.
-
-        Example:
-            ```python
-            executor = ScenarioExecutor(...)
-
-            # Run first test
-            result1 = await executor._run()
-
-            # Reset and run again
-            executor.reset()
-            result2 = await executor._run()
-            ```
         """
         self._state = ScenarioState(
             description=self.description,
@@ -349,24 +227,24 @@ class ScenarioExecutor:
                            Used to avoid broadcasting the message back to its creator.
 
         Example:
-            ```python
+            ```
             def inject_system_message(state: ScenarioState) -> None:
-                state._executor.add_message({
+                state.add_message({
                     "role": "system",
                     "content": "The user is now in a hurry"
                 })
 
             # Use in script
             result = await scenario.run(
-                name="system message test",
-                agents=[agent, user_sim, judge],
-                script=[
-                    scenario.user("Hello"),
-                    scenario.agent(),
-                    inject_system_message,
-                    scenario.user(),  # Will see the system message
-                    scenario.succeed()
-                ]
+               name="system message test",
+               agents=[agent, user_sim, judge],
+               script=[
+                   scenario.user("Hello"),
+                   scenario.agent(),
+                   inject_system_message,
+                   scenario.user(),  # Will see the system message
+                   scenario.succeed()
+               ]
             )
             ```
         """
@@ -396,7 +274,7 @@ class ScenarioExecutor:
             from_agent_idx: Index of the agent that generated these messages
 
         Example:
-            ```python
+            ```
             # Agent returns multiple messages for a complex interaction
             messages = [
                 {"role": "assistant", "content": "Let me search for that..."},
@@ -476,7 +354,11 @@ class ScenarioExecutor:
         self, role: AgentRole
     ) -> Tuple[int, Optional[AgentAdapter]]:
         for idx, agent in enumerate(self.agents):
-            if role == agent.role and agent in self._pending_agents_on_turn:
+            if (
+                role == agent.role
+                and agent in self._pending_agents_on_turn
+                and agent.role in self._pending_roles_on_turn
+            ):
                 return idx, agent
         return -1, None
 
@@ -503,7 +385,7 @@ class ScenarioExecutor:
             agent_time=agent_time,
         )
 
-    async def _run(self) -> ScenarioResult:
+    async def run(self) -> ScenarioResult:
         """
         Run a scenario against the agent under test.
 
@@ -513,30 +395,63 @@ class ScenarioExecutor:
         Returns:
             ScenarioResult containing the test outcome
         """
+        scenario_run_id = generate_scenario_run_id()
 
-        if self.config.verbose:
-            print("")  # new line
-
-        self.reset()
-
-        for script_step in self.script:
-            callable = script_step(self._state)
-            if isinstance(callable, Awaitable):
-                result = await callable
-            else:
-                result = callable
+        try:
+            self._emit_run_started_event(scenario_run_id)
 
-            if isinstance(result, ScenarioResult):
-                return result
+            if self.config.verbose:
+                print("")  # new line
+
+            self.reset()
+
+            for script_step in self.script:
+                callable = script_step(self._state)
+                if isinstance(callable, Awaitable):
+                    result = await callable
+                else:
+                    result = callable
+                self._emit_message_snapshot_event(scenario_run_id)
+
+                if isinstance(result, ScenarioResult):
+                    status = (
+                        ScenarioRunFinishedEventStatus.SUCCESS
+                        if result.success
+                        else ScenarioRunFinishedEventStatus.FAILED
+                    )
+                    self._emit_run_finished_event(scenario_run_id, result, status)
+                    return result
 
-        return self._reached_max_turns(
-            """Reached end of script without conclusion, add one of the following to the end of the script:
+            result = self._reached_max_turns(
+                """Reached end of script without conclusion, add one of the following to the end of the script:
 
 - `scenario.proceed()` to let the simulation continue to play out
 - `scenario.judge()` to force criteria judgement
 - `scenario.succeed()` or `scenario.fail()` to end the test with an explicit result
-            """
-        )
+                """
+            )
+
+            status = (
+                ScenarioRunFinishedEventStatus.SUCCESS
+                if result.success
+                else ScenarioRunFinishedEventStatus.FAILED
+            )
+            self._emit_run_finished_event(scenario_run_id, result, status)
+            return result
+
+        except Exception as e:
+            # Publish failure event before propagating the error
+            error_result = ScenarioResult(
+                success=False,
+                messages=self._state.messages,
+                reasoning=f"Scenario failed with error: {str(e)}",
+                total_time=time.time() - self._total_start_time,
+                agent_time=0,
+            )
+            self._emit_run_finished_event(
+                scenario_run_id, error_result, ScenarioRunFinishedEventStatus.ERROR
+            )
+            raise  # Re-raise the exception after cleanup
 
     async def _call_agent(
         self, idx: int, role: AgentRole, request_judgment: bool = False
@@ -577,16 +492,19 @@ class ScenarioExecutor:
         ):
             start_time = time.time()
 
-            agent_response = agent.call(
-                AgentInput(
-                    # TODO: test thread_id
-                    thread_id=self._state.thread_id,
-                    messages=self._state.messages,
-                    new_messages=self._pending_messages.get(idx, []),
-                    judgment_request=request_judgment,
-                    scenario_state=self._state,
+            # Prevent pydantic validation warnings which should already be disabled
+            with warnings.catch_warnings():
+                warnings.simplefilter("ignore")
+                agent_response = agent.call(
+                    AgentInput(
+                        # TODO: test thread_id
+                        thread_id=self._state.thread_id,
+                        messages=self._state.messages,
+                        new_messages=self._pending_messages.get(idx, []),
+                        judgment_request=request_judgment,
+                        scenario_state=self._state,
+                    )
                 )
-            )
             if not isinstance(agent_response, Awaitable):
                 raise Exception(
                     agent_response_not_awaitable(agent.__class__.__name__),
@@ -708,15 +626,24 @@ class ScenarioExecutor:
             reasoning=reasoning or "Scenario marked as failed with scenario.fail()",
         )
 
+    def _consume_until_role(self, role: AgentRole) -> None:
+        while len(self._pending_roles_on_turn) > 0:
+            next_role = self._pending_roles_on_turn[0]
+            if next_role == role:
+                break
+            self._pending_roles_on_turn.pop(0)
+
     async def _script_call_agent(
         self,
         role: AgentRole,
         content: Optional[Union[str, ChatCompletionMessageParam]] = None,
         request_judgment: bool = False,
     ) -> Optional[ScenarioResult]:
+        self._consume_until_role(role)
         idx, next_agent = self._next_agent_for_role(role)
         if not next_agent:
             self._new_turn()
+            self._consume_until_role(role)
             idx, next_agent = self._next_agent_for_role(role)
 
             if not next_agent:
@@ -738,11 +665,16 @@ class ScenarioExecutor:
                 )
 
         self._pending_agents_on_turn.remove(next_agent)
-        self._pending_roles_on_turn.remove(role)
 
         if content:
             if isinstance(content, str):
-                message = ChatCompletionUserMessageParam(role="user", content=content)
+                message = (
+                    ChatCompletionUserMessageParam(role="user", content=content)
+                    if role == AgentRole.USER
+                    else ChatCompletionAssistantMessageParam(
+                        role="assistant", content=content
+                    )
+                )
             else:
                 message = content
 
@@ -756,3 +688,228 @@ class ScenarioExecutor:
         )
         if isinstance(result, ScenarioResult):
             return result
+
+    # Event handling methods
+
+    class _CommonEventFields(TypedDict):
+        """
+        Common fields shared across all scenario events.
+
+        These fields provide consistent identification and timing information
+        for all events emitted during scenario execution.
+
+        Attributes:
+            batch_run_id: Unique identifier for the batch of scenario runs
+            scenario_run_id: Unique identifier for this specific scenario run
+            scenario_id: Human-readable name/identifier for the scenario
+            timestamp: Unix timestamp in milliseconds when the event occurred
+        """
+
+        batch_run_id: str
+        scenario_run_id: str
+        scenario_id: str
+        timestamp: int
+
+    def _create_common_event_fields(self, scenario_run_id: str) -> _CommonEventFields:
+        """
+        Create common fields used across all scenario events.
+
+        This method generates the standard fields that every scenario event
+        must include for proper identification and timing.
+
+        Args:
+            scenario_run_id: Unique identifier for the current scenario run
+
+        Returns:
+            Dictionary containing common event fields with current timestamp
+        """
+        return {
+            "batch_run_id": self.batch_run_id,
+            "scenario_run_id": scenario_run_id,
+            "scenario_id": self.name,
+            "timestamp": int(time.time() * 1000),
+        }
+
+    def _emit_run_started_event(self, scenario_run_id: str) -> None:
+        """
+        Emit a scenario run started event.
+
+        This event is published when a scenario begins execution. It includes
+        metadata about the scenario such as name and description, and is used
+        to track the start of scenario runs in monitoring systems.
+
+        Args:
+            scenario_run_id: Unique identifier for the current scenario run
+        """
+        common_fields = self._create_common_event_fields(scenario_run_id)
+        metadata = ScenarioRunStartedEventMetadata(
+            name=self.name,
+            description=self.description,
+        )
+
+        event = ScenarioRunStartedEvent(
+            **common_fields,
+            metadata=metadata,
+        )
+        self._emit_event(event)
+
+    def _emit_message_snapshot_event(self, scenario_run_id: str) -> None:
+        """
+        Emit a message snapshot event.
+
+        This event captures the current state of the conversation during
+        scenario execution. It's published whenever messages are added to
+        the conversation, allowing real-time tracking of scenario progress.
+        """
+        common_fields = self._create_common_event_fields(scenario_run_id)
+
+        event = ScenarioMessageSnapshotEvent(
+            **common_fields,
+            messages=convert_messages_to_api_client_messages(self._state.messages),
+        )
+        self._emit_event(event)
+
+    def _emit_run_finished_event(
+        self,
+        scenario_run_id: str,
+        result: ScenarioResult,
+        status: ScenarioRunFinishedEventStatus,
+    ) -> None:
+        """
+        Emit a scenario run finished event.
+
+        This event is published when a scenario completes execution, whether
+        successfully or with an error. It includes the final results, verdict,
+        and reasoning for the scenario outcome.
+
+        Args:
+            scenario_run_id: Unique identifier for the current scenario run
+            result: The final scenario result containing success/failure status
+            status: The execution status (SUCCESS, FAILED, or ERROR)
+        """
+        common_fields = self._create_common_event_fields(scenario_run_id)
+
+        results = ScenarioRunFinishedEventResults(
+            verdict=(
+                ScenarioRunFinishedEventVerdict.SUCCESS
+                if result.success
+                else ScenarioRunFinishedEventVerdict.FAILURE
+            ),
+            reasoning=result.reasoning or "",
+            met_criteria=result.passed_criteria,
+            unmet_criteria=result.failed_criteria,
+        )
+
+        event = ScenarioRunFinishedEvent(
+            **common_fields,
+            status=status,
+            results=results,
+        )
+        self._emit_event(event)
+
+        # Signal end of event stream
+        self._events.on_completed()
+
+
+async def run(
+    name: str,
+    description: str,
+    agents: List[AgentAdapter] = [],
+    max_turns: Optional[int] = None,
+    verbose: Optional[Union[bool, int]] = None,
+    cache_key: Optional[str] = None,
+    debug: Optional[bool] = None,
+    script: Optional[List[ScriptStep]] = None,
+) -> ScenarioResult:
+    """
+    High-level interface for running a scenario test.
+
+    This is the main entry point for executing scenario tests. It creates a
+    ScenarioExecutor instance and runs it in an isolated thread pool to support
+    parallel execution and prevent blocking.
+
+    Args:
+        name: Human-readable name for the scenario
+        description: Detailed description of what the scenario tests
+        agents: List of agent adapters (agent under test, user simulator, judge)
+        max_turns: Maximum conversation turns before timeout (default: 10)
+        verbose: Show detailed output during execution
+        cache_key: Cache key for deterministic behavior
+        debug: Enable debug mode for step-by-step execution
+        script: Optional script steps to control scenario flow
+
+    Returns:
+        ScenarioResult containing the test outcome, conversation history,
+        success/failure status, and detailed reasoning
+
+    Example:
+        ```
+        import scenario
+
+        # Simple scenario with automatic flow
+        result = await scenario.run(
+           name="help request",
+           description="User asks for help with a technical problem",
+           agents=[
+               my_agent,
+               scenario.UserSimulatorAgent(),
+               scenario.JudgeAgent(criteria=["Agent provides helpful response"])
+           ]
+        )
+
+        # Scripted scenario with custom evaluations
+        result = await scenario.run(
+           name="custom interaction",
+           description="Test specific conversation flow",
+           agents=[
+               my_agent,
+               scenario.UserSimulatorAgent(),
+               scenario.JudgeAgent(criteria=["Agent provides helpful response"])
+           ],
+           script=[
+               scenario.user("Hello"),
+               scenario.agent(),
+               custom_eval,
+               scenario.succeed()
+           ]
+        )
+
+        # Results analysis
+        print(f"Test {'PASSED' if result.success else 'FAILED'}")
+        print(f"Reasoning: {result.reasoning}")
+        print(f"Conversation had {len(result.messages)} messages")
+        ```
+    """
+    scenario = ScenarioExecutor(
+        name=name,
+        description=description,
+        agents=agents,
+        max_turns=max_turns,
+        verbose=verbose,
+        cache_key=cache_key,
+        debug=debug,
+        script=script,
+    )
+
+    # We'll use a thread pool to run the execution logic, we
+    # require a separate thread because even though asyncio is
+    # being used throughout, any user code on the callback can
+    # be blocking, preventing them from running scenarios in parallel
+    with concurrent.futures.ThreadPoolExecutor() as executor:
+
+        def run_in_thread():
+            loop = asyncio.new_event_loop()
+            asyncio.set_event_loop(loop)
+
+            try:
+                return loop.run_until_complete(scenario.run())
+            finally:
+                scenario.event_bus.drain()
+                loop.close()
+
+        # Run the function in the thread pool and await its result
+        # This converts the thread's execution into a Future that the current
+        # event loop can await without blocking
+        loop = asyncio.get_event_loop()
+        result = await loop.run_in_executor(executor, run_in_thread)
+        return result