tests/e2e/test_cross_service_workflows_e2e.py

"""
Cross-Service E2E Workflow Tests

Tests for end-to-end workflows spanning multiple services.

These tests validate that:
1. Agent execution leads to canvas presentation
2. Governance checks integrate with LLM calls
3. Canvas presentation creates episode records
4. Episode retrieval integrates with feedback
5. Multiple services chain together correctly
6. Workflow rollback works on failure
7. Workflow timeout handling is correct
8. Concurrent users don't interfere

Key Workflows Tested:
- Agent → Canvas presentation
- Governance → LLM call
- Canvas → Episode creation
- Episode → Feedback aggregation
- Multi-service chaining
- Rollback on failure
"""

import pytest
import time
from datetime import datetime, timedelta
from typing import Dict, Any

from sqlalchemy.orm import Session

from core.models import (
    AgentRegistry,
    AgentExecution,
    CanvasAudit,
    AgentEpisode,
    AgentFeedback
)


class TestAgentToCanvasWorkflow:
    """Test agent execution to canvas presentation workflow."""

    def test_agent_to_canvas_workflow(self, db_session: Session):
        """
        E2E: Agent execution → canvas presentation.

        Tests that agent execution can generate insights that are
        presented as a canvas to the user.

        Workflow:
        1. Create agent
        2. Execute agent task
        3. Generate insights
        4. Create canvas with insights
        5. Verify canvas audit record
        """
        # Step 1: Create agent
        agent = AgentRegistry(
            id="workflow-agent-1",
            name="Workflow Agent 1",
            description="Test agent-to-canvas workflow",
            category="Testing",
            module_path="test.workflow",
            class_name="WorkflowAgent1",
            status="AUTONOMOUS",
            confidence_score=0.95
        )
        db_session.add(agent)
        db_session.commit()

        # Step 2: Execute agent
        execution = AgentExecution(
            agent_id=agent.id,
            user_id="workflow-test-user",
            task="Generate insights from data",
            status="completed",
            input_data={"query": "analyze sales data"},
            output_data={
                "insights": [
                    "Sales increased by 20%",
                    "Top product: Widget A",
                    "Region: North America"
                ]
            },
            started_at=datetime.utcnow(),
            completed_at=datetime.utcnow()
        )
        db_session.add(execution)
        db_session.commit()

        # Step 3-4: Create canvas with insights
        canvas_id = "workflow-canvas-1"
        canvas = CanvasAudit(
            id=str(canvas_id),
            canvas_id=canvas_id,
            tenant_id="test-tenant",
            action_type="present",
            canvas_type="docs",
            canvas_data={
                "title": "Sales Insights",
                "content": execution.output_data["insights"],
                "source_execution_id": execution.id
            },
            created_at=datetime.utcnow()
        )
        db_session.add(canvas)
        db_session.commit()

        # Step 5: Verify workflow complete
        retrieved_execution = db_session.query(AgentExecution).filter(
            AgentExecution.id == execution.id
        ).first()
        assert retrieved_execution is not None
        assert retrieved_execution.status == "completed"
        assert "insights" in retrieved_execution.output_data

        retrieved_canvas = db_session.query(CanvasAudit).filter(
            CanvasAudit.canvas_id == canvas_id
        ).first()
        assert retrieved_canvas is not None
        assert retrieved_canvas.canvas_type == "docs"
        assert "insights" in retrieved_canvas.canvas_data

    def test_multi_agent_to_canvas_workflow(self, db_session: Session):
        """
        E2E: Multiple agents executing to single canvas.

        Tests that insights from multiple agents can be combined
        into a single canvas presentation.

        Workflow:
        1. Create multiple agents
        2. Execute all agents
        3. Combine insights
        4. Create unified canvas
        """
        # Step 1: Create multiple agents
        agents = []
        for i in range(3):
            agent = AgentRegistry(
                id=f"multi-agent-{i}",
                name=f"Multi Agent {i}",
                description="Test multi-agent workflow",
                category="Testing",
                module_path="test.multi",
                class_name=f"MultiAgent{i}",
                status="AUTONOMOUS",
                confidence_score=0.95
            )
            db_session.add(agent)
            agents.append(agent)
        db_session.commit()

        # Step 2: Execute all agents
        executions = []
        for agent in agents:
            execution = AgentExecution(
                agent_id=agent.id,
                user_id="multi-agent-user",
                task=f"Task {agent.id}",
                status="completed",
                output_data={"insight": f"Insight from {agent.id}"},
                started_at=datetime.utcnow(),
                completed_at=datetime.utcnow()
            )
            db_session.add(execution)
            executions.append(execution)
        db_session.commit()

        # Step 3-4: Combine insights into unified canvas
        combined_insights = [
            exec.output_data["insight"] for exec in executions
        ]

        canvas_id = "multi-agent-canvas"
        canvas = CanvasAudit(
            id=str(canvas_id),
            canvas_id=canvas_id,
            tenant_id="test-tenant",
            action_type="present",
            canvas_type="docs",
            canvas_data={
                "title": "Combined Insights",
                "insights": combined_insights,
                "agent_count": len(agents)
            },
            created_at=datetime.utcnow()
        )
        db_session.add(canvas)
        db_session.commit()

        # Verify unified canvas created
        retrieved = db_session.query(CanvasAudit).filter(
            CanvasAudit.canvas_id == canvas_id
        ).first()
        assert retrieved is not None
        assert retrieved.canvas_data["agent_count"] == 3
        assert len(retrieved.canvas_data["insights"]) == 3


class TestGovernanceToLLMWorkflow:
    """Test governance check to LLM call workflow."""

    def test_governance_to_llm_workflow(self, db_session: Session):
        """
        E2E: Governance check → LLM call.

        Tests that governance checks are performed before LLM calls,
        and that agents with appropriate maturity can invoke LLM.

        Workflow:
        1. Create AUTONOMOUS agent
        2. Perform governance check
        3. Check passes (AUTONOMOUS can use LLM)
        4. Simulate LLM call result
        5. Record execution with LLM output
        """
        # Step 1: Create AUTONOMOUS agent
        agent = AgentRegistry(
            id="llm-workflow-agent",
            name="LLM Workflow Agent",
            description="Test governance to LLM workflow",
            category="Testing",
            module_path="test.llm",
            class_name="LLMWorkflowAgent",
            status="AUTONOMOUS",
            confidence_score=0.95
        )
        db_session.add(agent)
        db_session.commit()

        # Step 2-3: Governance check (simulated - would use AgentGovernanceService)
        can_use_llm = agent.status == "AUTONOMOUS"  # Simplified check
        assert can_use_llm, "AUTONOMOUS agent should be able to use LLM"

        # Step 4-5: Simulate LLM call and record execution
        execution = AgentExecution(
            agent_id=agent.id,
            user_id="llm-workflow-user",
            task="Generate summary with LLM",
            status="completed",
            input_data={"prompt": "Summarize the data"},
            output_data={
                "llm_response": "This is a summary of the data...",
                "tokens_used": 150,
                "model": "gpt-4"
            },
            started_at=datetime.utcnow(),
            completed_at=datetime.utcnow()
        )
        db_session.add(execution)
        db_session.commit()

        # Verify LLM execution recorded
        retrieved = db_session.query(AgentExecution).filter(
            AgentExecution.agent_id == agent.id
        ).first()
        assert retrieved is not None
        assert "llm_response" in retrieved.output_data
        assert retrieved.output_data["tokens_used"] == 150

    def test_governance_blocks_student_from_llm(self, db_session: Session):
        """
        E2E: STUDENT agent blocked from LLM call.

        Tests that governance properly blocks STUDENT agents from
        making LLM calls (read-only maturity).

        Workflow:
        1. Create STUDENT agent
        2. Perform governance check
        3. Check fails (STUDENT cannot use LLM)
        4. Verify no LLM execution created
        """
        # Step 1: Create STUDENT agent
        agent = AgentRegistry(
            id="student-llm-agent",
            name="Student LLM Agent",
            description="Test STUDENT blocked from LLM",
            category="Testing",
            module_path="test.student",
            class_name="StudentLLMAgent",
            status="STUDENT",
            confidence_score=0.4
        )
        db_session.add(agent)
        db_session.commit()

        # Step 2-3: Governance check
        can_use_llm = agent.status == "AUTONOMOUS"  # Simplified check
        assert not can_use_llm, "STUDENT agent should be blocked from LLM"

        # Step 4: Verify no LLM execution created
        executions = db_session.query(AgentExecution).filter(
            AgentExecution.agent_id == agent.id,
            AgentExecution.output_data["llm_response"].isnot(None)  # type: ignore
        ).all()
        assert len(executions) == 0, "STUDENT agent should not have LLM executions"


class TestCanvasToEpisodeWorkflow:
    """Test canvas presentation to episode creation workflow."""

    def test_canvas_to_episode_workflow(self, db_session: Session):
        """
        E2E: Canvas presentation → episode creation.

        Tests that canvas presentations create episode records
        for episodic memory tracking.

        Workflow:
        1. Create agent and execute
        2. Present canvas
        3. Create episode linked to canvas
        4. Verify episode references canvas
        """
        # Step 1: Create agent and execute
        agent = AgentRegistry(
            id="canvas-episode-agent",
            name="Canvas Episode Agent",
            description="Test canvas to episode workflow",
            category="Testing",
            module_path="test.canvas_ep",
            class_name="CanvasEpisodeAgent",
            status="AUTONOMOUS",
            confidence_score=0.95
        )
        db_session.add(agent)
        db_session.commit()

        execution = AgentExecution(
            agent_id=agent.id,
            user_id="canvas-ep-user",
            task="Present data visualization",
            status="completed",
            started_at=datetime.utcnow(),
            completed_at=datetime.utcnow()
        )
        db_session.add(execution)
        db_session.commit()

        # Step 2: Present canvas
        canvas_id = "canvas-episode-1"
        canvas = CanvasAudit(
            id=str(canvas_id),
            canvas_id=canvas_id,
            tenant_id="test-tenant",
            action_type="present",
            canvas_type="chart",
            canvas_data={"type": "line", "data": [1, 2, 3]},
            created_at=datetime.utcnow()
        )
        db_session.add(canvas)
        db_session.commit()

        # Step 3: Create episode linked to canvas
        episode = AgentEpisode(
            id=f"episode-{canvas_id}",
            agent_id=agent.id,
            tenant_id="test-tenant",
            execution_id=execution.id,
            task_description="Present data visualization",
            maturity_at_time="autonomous",
            constitutional_score=1.0,
            outcome="success"
        )
        db_session.add(episode)
        db_session.commit()

        # Step 4: Verify episode references canvas
        retrieved = db_session.query(AgentEpisode).filter(
            AgentEpisode.id == episode.id
        ).first()
        assert retrieved is not None
        assert retrieved.execution_id == execution.id


class TestEpisodeToFeedbackWorkflow:
    """Test episode retrieval to feedback aggregation workflow."""

    def test_episode_to_feedback_workflow(self, db_session: Session):
        """
        E2E: Episode retrieval → feedback aggregation.

        Tests that user feedback on episodes is properly aggregated
        and linked for retrieval weighting.

        Workflow:
        1. Create episode
        2. User provides feedback
        3. Feedback linked to episode
        4. Verify feedback aggregation
        """
        # Step 1: Create episode
        agent = AgentRegistry(
            id="feedback-episode-agent",
            name="Feedback Episode Agent",
            description="Test episode to feedback workflow",
            category="Testing",
            module_path="test.fb_ep",
            class_name="FeedbackEpisodeAgent",
            status="AUTONOMOUS",
            confidence_score=0.95
        )
        db_session.add(agent)
        db_session.commit()

        episode = AgentEpisode(
            id="feedback-episode-1",
            agent_id=agent.id,
            tenant_id="test-tenant",
            task_description="Task for feedback testing",
            maturity_at_time="autonomous",
            constitutional_score=1.0,
            outcome="success"
        )
        db_session.add(episode)
        db_session.commit()

        # Step 2: User provides feedback
        feedback1 = AgentFeedback(
            agent_id=agent.id,
            user_id="feedback-user-1",
            feedback_type="thumbs_up",
            rating=1.0,
            comments="Great job!",
            episode_id=episode.id,
            timestamp=datetime.utcnow()
        )
        db_session.add(feedback1)

        feedback2 = AgentFeedback(
            agent_id=agent.id,
            user_id="feedback-user-2",
            feedback_type="thumbs_up",
            rating=1.0,
            comments="Very helpful",
            episode_id=episode.id,
            timestamp=datetime.utcnow()
        )
        db_session.add(feedback2)
        db_session.commit()

        # Step 3-4: Verify feedback aggregation
        feedbacks = db_session.query(AgentFeedback).filter(
            AgentFeedback.episode_id == episode.id
        ).all()
        assert len(feedbacks) == 2

        # Calculate average rating
        avg_rating = sum(f.rating for f in feedbacks) / len(feedbacks)
        assert avg_rating == 1.0


class TestMultiServiceChaining:
    """Test multiple services chained together."""

    def test_multi_service_chaining(self, db_session: Session):
        """
        E2E: Multiple services chained in sequence.

        Tests complete workflow: Agent → LLM → Canvas → Episode → Feedback

        Workflow:
        1. Execute agent with LLM
        2. Present canvas with LLM output
        3. Create episode
        4. Collect feedback
        5. Verify complete chain
        """
        # Step 1: Execute agent with LLM
        agent = AgentRegistry(
            id="chain-agent",
            name="Chain Agent",
            description="Test multi-service chaining",
            category="Testing",
            module_path="test.chain",
            class_name="ChainAgent",
            status="AUTONOMOUS",
            confidence_score=0.95
        )
        db_session.add(agent)
        db_session.commit()

        execution = AgentExecution(
            agent_id=agent.id,
            user_id="chain-user",
            task="Analyze and present",
            status="completed",
            output_data={
                "llm_response": "Analysis complete",
                "insights": ["Insight 1", "Insight 2"]
            },
            started_at=datetime.utcnow(),
            completed_at=datetime.utcnow()
        )
        db_session.add(execution)
        db_session.commit()

        # Step 2: Present canvas with LLM output
        canvas_id = "chain-canvas"
        canvas = CanvasAudit(
            id=str(canvas_id),
            canvas_id=canvas_id,
            tenant_id="test-tenant",
            action_type="present",
            canvas_type="docs",
            canvas_data={
                "title": "Analysis Results",
                "content": execution.output_data["llm_response"]
            },
            created_at=datetime.utcnow()
        )
        db_session.add(canvas)
        db_session.commit()

        # Step 3: Create episode
        episode = AgentEpisode(
            id="chain-episode",
            agent_id=agent.id,
            tenant_id="test-tenant",
            execution_id=execution.id,
            task_description="Analyze and present",
            maturity_at_time="autonomous",
            constitutional_score=1.0,
            outcome="success"
        )
        db_session.add(episode)
        db_session.commit()

        # Step 4: Collect feedback
        feedback = AgentFeedback(
            agent_id=agent.id,
            user_id="chain-user",
            feedback_type="thumbs_up",
            rating=1.0,
            comments="Excellent analysis",
            canvas_id=canvas_id,
            episode_id=episode.id,
            timestamp=datetime.utcnow()
        )
        db_session.add(feedback)
        db_session.commit()

        # Step 5: Verify complete chain
        assert execution.id is not None
        assert canvas.canvas_id == canvas_id
        assert episode.execution_id == execution.id
        assert feedback.canvas_id == canvas_id
        assert feedback.episode_id == episode.id


class TestWorkflowRollbackOnFailure:
    """Test workflow rollback when intermediate step fails."""

    def test_workflow_rollback_on_failure(self, db_session: Session):
        """
        E2E: Workflow rollback on intermediate step failure.

        Tests that when an intermediate step fails, the workflow
        doesn't leave inconsistent state.

        Workflow:
        1. Execute agent (succeeds)
        2. Try to create canvas (fails - invalid data)
        3. Verify agent execution still recorded
        4. Verify canvas not created
        """
        # Step 1: Execute agent
        agent = AgentRegistry(
            id="rollback-agent",
            name="Rollback Agent",
            description="Test workflow rollback",
            category="Testing",
            module_path="test.rollback",
            class_name="RollbackAgent",
            status="AUTONOMOUS",
            confidence_score=0.95
        )
        db_session.add(agent)
        db_session.commit()

        execution = AgentExecution(
            agent_id=agent.id,
            user_id="rollback-user",
            task="Task that will fail at canvas stage",
            status="completed",
            started_at=datetime.utcnow(),
            completed_at=datetime.utcnow()
        )
        db_session.add(execution)
        db_session.commit()

        # Step 2: Try to create canvas with invalid data (would fail in real scenario)
        # In this test, we simulate by not creating the canvas

        # Step 3-4: Verify state
        retrieved_exec = db_session.query(AgentExecution).filter(
            AgentExecution.id == execution.id
        ).first()
        assert retrieved_exec is not None, "Agent execution should be recorded"

        canvas_count = db_session.query(CanvasAudit).filter(
            CanvasAudit.canvas_id == "rollback-canvas"
        ).count()
        assert canvas_count == 0, "Canvas should not be created"


class TestWorkflowTimeoutHandling:
    """Test workflow timeout handling."""

    def test_workflow_timeout_handling(self, db_session: Session):
        """
        E2E: Workflow timeout doesn't leave inconsistent state.

        Tests that when a workflow times out, partial state is
        properly handled.

        Workflow:
        1. Start agent execution
        2. Simulate timeout
        3. Mark execution as failed
        4. Verify no partial canvas/episode created
        """
        # Step 1: Start agent execution
        agent = AgentRegistry(
            id="timeout-agent",
            name="Timeout Agent",
            description="Test workflow timeout",
            category="Testing",
            module_path="test.timeout",
            class_name="TimeoutAgent",
            status="AUTONOMOUS",
            confidence_score=0.95
        )
        db_session.add(agent)
        db_session.commit()

        execution = AgentExecution(
            agent_id=agent.id,
            user_id="timeout-user",
            task="Long running task",
            status="in_progress",
            started_at=datetime.utcnow()
        )
        db_session.add(execution)
        db_session.commit()

        # Step 2-3: Simulate timeout
        execution.status = "failed"
        execution.error_message = "Workflow timeout after 30 seconds"
        execution.completed_at = datetime.utcnow()
        db_session.commit()

        # Step 4: Verify no partial artifacts
        # (In real scenario, canvas/episode shouldn't be created for timed-out execution)
        retrieved = db_session.query(AgentExecution).filter(
            AgentExecution.id == execution.id
        ).first()
        assert retrieved.status == "failed"
        assert "timeout" in retrieved.error_message.lower()


class TestWorkflowWithConcurrentUsers:
    """Test workflow with multiple concurrent users."""

    def test_workflow_with_concurrent_users(self, db_session: Session):
        """
        E2E: Multiple users executing workflows simultaneously.

        Tests that concurrent users don't interfere with each other's
        workflow state.

        Workflow:
        1. User A executes agent and creates canvas
        2. User B executes agent and creates canvas concurrently
        3. Verify both workflows complete independently
        4. Verify no data mixing between users
        """
        # Create shared agent
        agent = AgentRegistry(
            id="concurrent-users-agent",
            name="Concurrent Users Agent",
            description="Test concurrent user workflows",
            category="Testing",
            module_path="test.concurrent_users",
            class_name="ConcurrentUsersAgent",
            status="AUTONOMOUS",
            confidence_score=0.95
        )
        db_session.add(agent)
        db_session.commit()

        # User A workflow
        execution_a = AgentExecution(
            agent_id=agent.id,
            user_id="user-a",
            task="User A task",
            status="completed",
            started_at=datetime.utcnow(),
            completed_at=datetime.utcnow()
        )
        db_session.add(execution_a)

        canvas_a = CanvasAudit(
            id="canvas-a",
            canvas_id="canvas-a",
            tenant_id="test-tenant",
            action_type="present",
            canvas_type="docs",
            canvas_data={"user": "A"},
            created_at=datetime.utcnow()
        )
        db_session.add(canvas_a)

        # User B workflow
        execution_b = AgentExecution(
            agent_id=agent.id,
            user_id="user-b",
            task="User B task",
            status="completed",
            started_at=datetime.utcnow(),
            completed_at=datetime.utcnow()
        )
        db_session.add(execution_b)

        canvas_b = CanvasAudit(
            id="canvas-b",
            canvas_id="canvas-b",
            tenant_id="test-tenant",
            action_type="present",
            canvas_type="docs",
            canvas_data={"user": "B"},
            created_at=datetime.utcnow()
        )
        db_session.add(canvas_b)
        db_session.commit()

        # Verify both workflows complete
        exec_a = db_session.query(AgentExecution).filter(
            AgentExecution.user_id == "user-a"
        ).first()
        exec_b = db_session.query(AgentExecution).filter(
            AgentExecution.user_id == "user-b"
        ).first()

        assert exec_a is not None
        assert exec_b is not None
        assert exec_a.user_id != exec_b.user_id

        # Verify no data mixing
        canvas_a_retrieved = db_session.query(CanvasAudit).filter(
            CanvasAudit.canvas_id == "canvas-a"
        ).first()
        canvas_b_retrieved = db_session.query(CanvasAudit).filter(
            CanvasAudit.canvas_id == "canvas-b"
        ).first()

        assert canvas_a_retrieved.canvas_data["user"] == "A"
        assert canvas_b_retrieved.canvas_data["user"] == "B"