synth-ai 0.2.9.dev4__py3-none-any.whl → 0.2.9.dev6__py3-none-any.whl

synth_ai/tracing_v3/turso/manager.py

@@ -1,774 +0,0 @@
-from __future__ import annotations
-"""Async SQLAlchemy-based trace manager for Turso/sqld.
-
-This module provides the database interface for the tracing system using
-async SQLAlchemy with a Turso/sqld backend. It handles all database operations
-including schema creation, session storage, and analytics queries.
-
-Key Features:
-------------
-- Async-first design using aiosqlite for local SQLite
-- Automatic schema creation and migration
-- Batch insert capabilities for high-throughput scenarios
-- Analytics views for efficient querying
-- Connection pooling and retry logic
-
-Performance Considerations:
---------------------------
-- Uses NullPool for SQLite to avoid connection issues
-- Implements busy timeout for concurrent access
-- Batches inserts to reduce transaction overhead
-- Creates indexes for common query patterns
-"""
-
-import asyncio
-import logging
-from contextlib import asynccontextmanager
-from datetime import datetime
-from typing import Any
-
-# Optional pandas import: fall back to records (list[dict]) if unavailable
-try:  # pragma: no cover - exercised in environments without pandas
-    import pandas as pd  # type: ignore
-except Exception:  # pragma: no cover
-    pd = None  # type: ignore[assignment]
-from sqlalchemy import select, text, update
-from sqlalchemy.exc import IntegrityError
-from sqlalchemy.ext.asyncio import AsyncEngine, AsyncSession, create_async_engine
-from sqlalchemy import event
-from sqlalchemy.orm import selectinload, sessionmaker
-from sqlalchemy.pool import NullPool
-
-from ..abstractions import (
-    EnvironmentEvent,
-    LMCAISEvent,
-    RuntimeEvent,
-    SessionTrace,
-)
-from ..config import CONFIG
-from .models import (
-    Base,
-    analytics_views,
-)
-from .models import (
-    Event as DBEvent,
-)
-from .models import (
-    Experiment as DBExperiment,
-)
-from .models import (
-    Message as DBMessage,
-)
-from .models import (
-    SessionTimestep as DBSessionTimestep,
-)
-from .models import (
-    SessionTrace as DBSessionTrace,
-)
-from .models import (
-    OutcomeReward as DBOutcomeReward,
-)
-from .models import (
-    EventReward as DBEventReward,
-)
-
-logger = logging.getLogger(__name__)
-
-
-class AsyncSQLTraceManager:
-    """Async trace storage manager using SQLAlchemy and Turso/sqld.
-
-    Handles all database operations for the tracing system. Designed to work
-    with both local SQLite (via aiosqlite) and remote Turso databases.
-
-    The manager handles:
-    - Connection lifecycle management
-    - Schema creation and verification
-    - Transaction management
-    - Batch operations for efficiency
-    - Analytics view creation
-    """
-
-    def __init__(self, db_url: str | None = None):
-        self.db_url = db_url or CONFIG.db_url
-        self.engine: AsyncEngine | None = None
-        self.SessionLocal: sessionmaker | None = None
-        self._schema_lock = asyncio.Lock()
-        self._schema_ready = False
-
-    async def initialize(self):
-        """Initialize the database connection and schema.
-
-        This method is idempotent and thread-safe. It:
-        1. Creates the async engine with appropriate settings
-        2. Verifies database file exists (for SQLite)
-        3. Creates schema if needed
-        4. Sets up analytics views
-
-        The schema lock ensures only one worker creates the schema in
-        concurrent scenarios.
-        """
-        if self.engine is None:
-            logger.debug(f"🔗 Initializing database connection to: {self.db_url}")
-
-            # For SQLite, use NullPool to avoid connection pool issues
-            # SQLite doesn't handle concurrent connections well, so we create
-            # a new connection for each operation
-            if self.db_url.startswith("sqlite"):
-                # Extract the file path from the URL
-                db_path = self.db_url.replace("sqlite+aiosqlite:///", "")
-                import os
-
-                # Check if database file exists
-                if not os.path.exists(db_path):
-                    logger.debug(f"⚠️  Database file not found: {db_path}")
-                    logger.debug(
-                        "🔧 Make sure './serve.sh' is running to start the turso/sqld service"
-                    )
-                else:
-                    logger.debug(f"✅ Found database file: {db_path}")
-
-                # Set a high busy timeout to handle concurrent access
-                # This allows SQLite to wait instead of immediately failing
-                connect_args = {"timeout": 30.0}  # 30 second busy timeout
-                self.engine = create_async_engine(
-                    self.db_url,  # Use instance db_url, not CONFIG
-                    poolclass=NullPool,  # No connection pooling for SQLite
-                    connect_args=connect_args,
-                    echo=CONFIG.echo_sql,
-                )
-                # Ensure PRAGMA foreign_keys=ON for every connection
-                try:
-                    @event.listens_for(self.engine.sync_engine, "connect")
-                    def _set_sqlite_pragma(dbapi_connection, connection_record):  # type: ignore[no-redef]
-                        try:
-                            cursor = dbapi_connection.cursor()
-                            cursor.execute("PRAGMA foreign_keys=ON")
-                            cursor.close()
-                        except Exception:
-                            pass
-                except Exception:
-                    pass
-            else:
-                connect_args = CONFIG.get_connect_args()
-                engine_kwargs = CONFIG.get_engine_kwargs()
-                self.engine = create_async_engine(
-                    self.db_url,  # Use instance db_url, not CONFIG
-                    connect_args=connect_args,
-                    **engine_kwargs,
-                )
-
-            self.SessionLocal = sessionmaker(
-                self.engine, class_=AsyncSession, expire_on_commit=False
-            )
-
-        await self._ensure_schema()
-
-    async def _ensure_schema(self):
-        """Ensure database schema is created.
-
-        Uses a lock to prevent race conditions when multiple workers start
-        simultaneously. The checkfirst=True parameter handles cases where
-        another worker already created the schema.
-        """
-        async with self._schema_lock:
-            if self._schema_ready:
-                return
-
-            logger.debug("📊 Initializing database schema...")
-
-            async with self.engine.begin() as conn:
-                # Use a transaction to ensure atomic schema creation
-                # checkfirst=True prevents errors if tables already exist
-                try:
-                    await conn.run_sync(
-                        lambda sync_conn: Base.metadata.create_all(sync_conn, checkfirst=True)
-                    )
-                    # logger.info("✅ Database schema created/verified successfully")
-                except Exception as e:
-                    # If tables already exist, that's fine - another worker created them
-                    if "already exists" not in str(e):
-                        logger.error(f"❌ Failed to create database schema: {e}")
-                        raise
-                    else:
-                        logger.debug("✅ Database schema already exists")
-
-                # Enable foreign keys for SQLite - critical for data integrity
-                # This must be done for each connection in SQLite
-                if CONFIG.foreign_keys:
-                    await conn.execute(text("PRAGMA foreign_keys = ON"))
-
-                # Set journal mode
-                if CONFIG.journal_mode:
-                    await conn.execute(text(f"PRAGMA journal_mode = {CONFIG.journal_mode}"))
-
-                # Create analytics views for efficient querying
-                # These are materialized as views to avoid recalculation
-                for view_name, view_sql in analytics_views.items():
-                    try:
-                        await conn.execute(text(view_sql))
-                    except Exception as e:
-                        # Views might already exist from another worker
-                        if "already exists" not in str(e):
-                            logger.warning(f"Could not create view {view_name}: {e}")
-
-            self._schema_ready = True
-            # logger.debug("🎯 Database ready for use!")
-
-    @asynccontextmanager
-    async def session(self):
-        """Get an async database session."""
-        if not self.SessionLocal:
-            await self.initialize()
-        async with self.SessionLocal() as session:
-            yield session
-
-    async def insert_session_trace(self, trace: SessionTrace) -> str:
-        """Insert a complete session trace.
-
-        This method handles the complex task of inserting a complete session
-        with all its timesteps, events, and messages. It uses a single
-        transaction for atomicity and flushes after timesteps to get their
-        auto-generated IDs for foreign keys.
-
-        Args:
-            trace: The complete session trace to store
-
-        Returns:
-            The session ID
-
-        Raises:
-            IntegrityError: If session ID already exists (handled gracefully)
-        """
-        async with self.session() as sess:
-            try:
-                # Convert to cents for cost storage - avoids floating point
-                # precision issues and allows for integer arithmetic
-                def to_cents(cost: float | None) -> int | None:
-                    return int(cost * 100) if cost is not None else None
-
-                # Insert session
-                db_session = DBSessionTrace(
-                    session_id=trace.session_id,
-                    created_at=trace.created_at,
-                    num_timesteps=len(trace.session_time_steps),
-                    num_events=len(trace.event_history),
-                    num_messages=len(trace.markov_blanket_message_history),
-                    session_metadata=trace.metadata or {},
-                )
-                sess.add(db_session)
-
-                # Track timestep IDs for foreign keys - we need these to link
-                # events and messages to their respective timesteps
-                step_id_map: dict[str, int] = {}
-
-                # Insert timesteps
-                for step in trace.session_time_steps:
-                    db_step = DBSessionTimestep(
-                        session_id=trace.session_id,
-                        step_id=step.step_id,
-                        step_index=step.step_index,
-                        turn_number=step.turn_number,
-                        started_at=step.timestamp,
-                        completed_at=step.completed_at,
-                        num_events=len(step.events),
-                        num_messages=len(step.markov_blanket_messages),
-                        step_metadata=step.step_metadata or {},
-                    )
-                    sess.add(db_step)
-                    # Flush to get the auto-generated ID without committing
-                    # This allows us to use the ID for foreign keys while
-                    # maintaining transaction atomicity
-                    await sess.flush()  # Get the auto-generated ID
-                    step_id_map[step.step_id] = db_step.id
-
-                # Insert events - handle different event types with their
-                # specific fields while maintaining a unified storage model
-                for event in trace.event_history:
-                    event_data = {
-                        "session_id": trace.session_id,
-                        "timestep_id": step_id_map.get(event.metadata.get("step_id")),
-                        "system_instance_id": event.system_instance_id,
-                        "event_time": event.time_record.event_time,
-                        "message_time": event.time_record.message_time,
-                        "event_metadata_json": event.metadata or {},
-                        "event_extra_metadata": event.event_metadata,
-                    }
-
-                    if isinstance(event, LMCAISEvent):
-                        # Serialize call_records if present
-                        call_records_data = None
-                        if event.call_records:
-                            from dataclasses import asdict
-
-                            call_records_data = [asdict(record) for record in event.call_records]
-
-                        event_data.update(
-                            {
-                                "event_type": "cais",
-                                "model_name": event.model_name,
-                                "provider": event.provider,
-                                "input_tokens": event.input_tokens,
-                                "output_tokens": event.output_tokens,
-                                "total_tokens": event.total_tokens,
-                                "cost_usd": to_cents(event.cost_usd),
-                                "latency_ms": event.latency_ms,
-                                "span_id": event.span_id,
-                                "trace_id": event.trace_id,
-                                "system_state_before": event.system_state_before,
-                                "system_state_after": event.system_state_after,
-                                "call_records": call_records_data,  # Store in the proper column
-                            }
-                        )
-                    elif isinstance(event, EnvironmentEvent):
-                        event_data.update(
-                            {
-                                "event_type": "environment",
-                                "reward": event.reward,
-                                "terminated": event.terminated,
-                                "truncated": event.truncated,
-                                "system_state_before": event.system_state_before,
-                                "system_state_after": event.system_state_after,
-                            }
-                        )
-                    elif isinstance(event, RuntimeEvent):
-                        event_data.update(
-                            {
-                                "event_type": "runtime",
-                                "event_metadata_json": {**event.metadata, "actions": event.actions},
-                            }
-                        )
-                    else:
-                        event_data["event_type"] = event.__class__.__name__.lower()
-
-                    db_event = DBEvent(**event_data)
-                    sess.add(db_event)
-
-                # Insert messages
-                for msg in trace.markov_blanket_message_history:
-                    db_msg = DBMessage(
-                        session_id=trace.session_id,
-                        timestep_id=step_id_map.get(msg.metadata.get("step_id"))
-                        if hasattr(msg, "metadata")
-                        else None,
-                        message_type=msg.message_type,
-                        content=msg.content,
-                        event_time=msg.time_record.event_time,
-                        message_time=msg.time_record.message_time,
-                        message_metadata=msg.metadata if hasattr(msg, "metadata") else {},
-                    )
-                    sess.add(db_msg)
-
-                # Commit the entire transaction atomically
-                await sess.commit()
-                return trace.session_id
-            except IntegrityError as e:
-                # Handle duplicate session IDs gracefully - this can happen
-                # in distributed systems or retries. We return the existing
-                # ID to maintain idempotency
-                if "UNIQUE constraint failed: session_traces.session_id" in str(e):
-                    await sess.rollback()
-                    return trace.session_id  # Return existing ID
-                raise
-
-    async def get_session_trace(self, session_id: str) -> dict[str, Any] | None:
-        """Retrieve a session trace by ID."""
-        async with self.session() as sess:
-            result = await sess.execute(
-                select(DBSessionTrace)
-                .options(
-                    selectinload(DBSessionTrace.timesteps),
-                    selectinload(DBSessionTrace.events),
-                    selectinload(DBSessionTrace.messages),
-                )
-                .where(DBSessionTrace.session_id == session_id)
-            )
-            session = result.scalar_one_or_none()
-
-            if not session:
-                return None
-
-            return {
-                "session_id": session.session_id,
-                "created_at": session.created_at,
-                "num_timesteps": session.num_timesteps,
-                "num_events": session.num_events,
-                "num_messages": session.num_messages,
-                "metadata": session.session_metadata,
-                "timesteps": [
-                    {
-                        "step_id": step.step_id,
-                        "step_index": step.step_index,
-                        "turn_number": step.turn_number,
-                        "started_at": step.started_at,
-                        "completed_at": step.completed_at,
-                        "metadata": step.step_metadata,
-                    }
-                    for step in sorted(session.timesteps, key=lambda s: s.step_index)
-                ],
-            }
-
-    async def query_traces(
-        self, query: str, params: dict[str, Any] | None = None
-    ) -> Any:
-        """Execute a query and return results.
-
-        Returns a pandas DataFrame when pandas is available; otherwise a
-        list of dict records. Callers should handle both.
-        """
-        async with self.session() as sess:
-            result = await sess.execute(text(query), params or {})
-            rows = result.mappings().all()
-            if pd is not None:
-                return pd.DataFrame(rows)
-            return [dict(r) for r in rows]
-
-    async def get_model_usage(
-        self,
-        start_date: datetime | None = None,
-        end_date: datetime | None = None,
-        model_name: str | None = None,
-    ) -> Any:
-        """Get model usage statistics.
-
-        Returns a pandas DataFrame when pandas is available; otherwise a list
-        of dict records.
-        """
-        query = """
-            SELECT * FROM model_usage_stats
-            WHERE 1=1
-        """
-        params = {}
-
-        if start_date:
-            query += " AND last_used >= :start_date"
-            params["start_date"] = start_date
-
-        if end_date:
-            query += " AND first_used <= :end_date"
-            params["end_date"] = end_date
-
-        if model_name:
-            query += " AND model_name = :model_name"
-            params["model_name"] = model_name
-
-        query += " ORDER BY usage_count DESC"
-
-        return await self.query_traces(query, params)
-
-    async def create_experiment(
-        self,
-        experiment_id: str,
-        name: str,
-        description: str | None = None,
-        configuration: dict[str, Any] | None = None,
-    ) -> str:
-        """Create a new experiment."""
-        async with self.session() as sess:
-            experiment = DBExperiment(
-                experiment_id=experiment_id,
-                name=name,
-                description=description,
-                configuration=configuration or {},
-            )
-            sess.add(experiment)
-            await sess.commit()
-            return experiment_id
-
-    async def link_session_to_experiment(self, session_id: str, experiment_id: str):
-        """Link a session to an experiment."""
-        async with self.session() as sess:
-            await sess.execute(
-                update(DBSessionTrace)
-                .where(DBSessionTrace.session_id == session_id)
-                .values(experiment_id=experiment_id)
-            )
-            await sess.commit()
-
-    async def batch_insert_sessions(
-        self, traces: list[SessionTrace], batch_size: int | None = None
-    ) -> list[str]:
-        """Batch insert multiple session traces.
-
-        Processes traces in batches to balance memory usage and performance.
-        Each batch is inserted in a separate transaction to avoid holding
-        locks for too long.
-
-        Args:
-            traces: List of session traces to insert
-            batch_size: Number of traces per batch (defaults to config)
-
-        Returns:
-            List of inserted session IDs
-        """
-        batch_size = batch_size or CONFIG.batch_size
-        inserted_ids = []
-
-        # Process in chunks to avoid memory issues with large datasets
-        for i in range(0, len(traces), batch_size):
-            batch = traces[i : i + batch_size]
-            # Insert each trace in the batch - could be optimized further
-            # with bulk inserts if needed
-            for trace in batch:
-                session_id = await self.insert_session_trace(trace)
-                inserted_ids.append(session_id)
-
-        return inserted_ids
-
-    async def get_sessions_by_experiment(
-        self, experiment_id: str, limit: int | None = None
-    ) -> list[dict[str, Any]]:
-        """Get all sessions for an experiment."""
-        async with self.session() as sess:
-            query = (
-                select(DBSessionTrace)
-                .where(DBSessionTrace.experiment_id == experiment_id)
-                .order_by(DBSessionTrace.created_at.desc())
-            )
-
-            if limit:
-                query = query.limit(limit)
-
-            result = await sess.execute(query)
-            sessions = result.scalars().all()
-
-            return [
-                {
-                    "session_id": s.session_id,
-                    "created_at": s.created_at,
-                    "num_timesteps": s.num_timesteps,
-                    "num_events": s.num_events,
-                    "num_messages": s.num_messages,
-                    "metadata": s.metadata,
-                }
-                for s in sessions
-            ]
-
-    async def delete_session(self, session_id: str) -> bool:
-        """Delete a session and all related data."""
-        async with self.session() as sess:
-            # Get the session object to trigger cascade deletes
-            result = await sess.execute(
-                select(DBSessionTrace).where(DBSessionTrace.session_id == session_id)
-            )
-            session = result.scalar_one_or_none()
-
-            if session:
-                await sess.delete(session)
-                await sess.commit()
-                return True
-            return False
-
-    async def close(self):
-        """Close the database connection.
-
-        Properly disposes of the engine and all connections. This is important
-        for cleanup, especially with SQLite which can leave lock files.
-        """
-        if self.engine:
-            # Dispose of all connections in the pool
-            await self.engine.dispose()
-            # Clear all state to allow re-initialization if needed
-            self.engine = None
-            self.SessionLocal = None
-            self._schema_ready = False
-
-    # -------------------------------
-    # Incremental insert helpers
-    # -------------------------------
-
-    async def ensure_session(self, session_id: str, *, created_at: datetime | None = None, metadata: dict[str, Any] | None = None):
-        """Ensure a DB session row exists for session_id."""
-        async with self.session() as sess:
-            result = await sess.execute(select(DBSessionTrace).where(DBSessionTrace.session_id == session_id))
-            existing = result.scalar_one_or_none()
-            if existing:
-                return
-            row = DBSessionTrace(
-                session_id=session_id,
-                created_at=created_at or datetime.utcnow(),
-                num_timesteps=0,
-                num_events=0,
-                num_messages=0,
-                session_metadata=metadata or {},
-            )
-            sess.add(row)
-            await sess.commit()
-
-    async def ensure_timestep(self, session_id: str, *, step_id: str, step_index: int, turn_number: int | None = None, started_at: datetime | None = None, completed_at: datetime | None = None, metadata: dict[str, Any] | None = None) -> int:
-        """Ensure a timestep row exists; return its DB id."""
-        async with self.session() as sess:
-            result = await sess.execute(
-                select(DBSessionTimestep).where(DBSessionTimestep.session_id == session_id, DBSessionTimestep.step_id == step_id)
-            )
-            row = result.scalar_one_or_none()
-            if row:
-                return row.id
-            row = DBSessionTimestep(
-                session_id=session_id,
-                step_id=step_id,
-                step_index=step_index,
-                turn_number=turn_number,
-                started_at=started_at or datetime.utcnow(),
-                completed_at=completed_at,
-                num_events=0,
-                num_messages=0,
-                step_metadata=metadata or {},
-            )
-            sess.add(row)
-            await sess.flush()
-            # increment session num_timesteps
-            await sess.execute(
-                update(DBSessionTrace)
-                .where(DBSessionTrace.session_id == session_id)
-                .values(num_timesteps=DBSessionTrace.num_timesteps + 1)
-            )
-            await sess.commit()
-            return row.id
-
-    async def insert_message_row(self, session_id: str, *, timestep_db_id: int | None, message_type: str, content: str, event_time: float | None = None, message_time: int | None = None, metadata: dict[str, Any] | None = None) -> int:
-        """Insert a message and return its id."""
-        async with self.session() as sess:
-            db_msg = DBMessage(
-                session_id=session_id,
-                timestep_id=timestep_db_id,
-                message_type=message_type,
-                content=content,
-                event_time=event_time,
-                message_time=message_time,
-                message_metadata=metadata or {},
-            )
-            sess.add(db_msg)
-            await sess.flush()
-            # increment session num_messages
-            await sess.execute(
-                update(DBSessionTrace)
-                .where(DBSessionTrace.session_id == session_id)
-                .values(num_messages=DBSessionTrace.num_messages + 1)
-            )
-            await sess.commit()
-            return db_msg.id
-
-    async def insert_event_row(self, session_id: str, *, timestep_db_id: int | None, event: EnvironmentEvent | LMCAISEvent | RuntimeEvent, metadata_override: dict[str, Any] | None = None) -> int:
-        """Insert an event and return its id."""
-        def to_cents(cost: float | None) -> int | None:
-            return int(cost * 100) if cost is not None else None
-
-        event_data: dict[str, Any] = {
-            "session_id": session_id,
-            "timestep_id": timestep_db_id,
-            "system_instance_id": event.system_instance_id,
-            "event_time": event.time_record.event_time,
-            "message_time": event.time_record.message_time,
-            "event_metadata_json": metadata_override or event.metadata or {},
-            "event_extra_metadata": getattr(event, "event_metadata", None),
-        }
-        if isinstance(event, LMCAISEvent):
-            call_records_data = None
-            if getattr(event, "call_records", None):
-                from dataclasses import asdict
-
-                call_records_data = [asdict(record) for record in event.call_records]
-            event_data.update({
-                "event_type": "cais",
-                "model_name": event.model_name,
-                "provider": event.provider,
-                "input_tokens": event.input_tokens,
-                "output_tokens": event.output_tokens,
-                "total_tokens": event.total_tokens,
-                "cost_usd": to_cents(event.cost_usd),
-                "latency_ms": event.latency_ms,
-                "span_id": event.span_id,
-                "trace_id": event.trace_id,
-                "system_state_before": event.system_state_before,
-                "system_state_after": event.system_state_after,
-                "call_records": call_records_data,
-            })
-        elif isinstance(event, EnvironmentEvent):
-            event_data.update({
-                "event_type": "environment",
-                "reward": event.reward,
-                "terminated": event.terminated,
-                "truncated": event.truncated,
-                "system_state_before": event.system_state_before,
-                "system_state_after": event.system_state_after,
-            })
-        elif isinstance(event, RuntimeEvent):
-            event_data.update({
-                "event_type": "runtime",
-                "event_metadata_json": {**(event.metadata or {}), "actions": event.actions},
-            })
-        else:
-            event_data["event_type"] = event.__class__.__name__.lower()
-
-        async with self.session() as sess:
-            db_event = DBEvent(**event_data)
-            sess.add(db_event)
-            await sess.flush()
-            # increment session num_events
-            await sess.execute(
-                update(DBSessionTrace)
-                .where(DBSessionTrace.session_id == session_id)
-                .values(num_events=DBSessionTrace.num_events + 1)
-            )
-            await sess.commit()
-            return db_event.id
-
-    # -------------------------------
-    # Reward helpers
-    # -------------------------------
-
-    async def insert_outcome_reward(self, session_id: str, *, total_reward: int, achievements_count: int, total_steps: int, reward_metadata: dict | None = None) -> int:
-        async with self.session() as sess:
-            row = DBOutcomeReward(
-                session_id=session_id,
-                total_reward=total_reward,
-                achievements_count=achievements_count,
-                total_steps=total_steps,
-                reward_metadata=reward_metadata or {},
-            )
-            sess.add(row)
-            await sess.flush()
-            await sess.commit()
-            return row.id
-
-    async def insert_event_reward(self, session_id: str, *, event_id: int, message_id: int | None = None, turn_number: int | None = None, reward_value: float = 0.0, reward_type: str | None = None, key: str | None = None, annotation: dict[str, Any] | None = None, source: str | None = None) -> int:
-        async with self.session() as sess:
-            row = DBEventReward(
-                event_id=event_id,
-                session_id=session_id,
-                message_id=message_id,
-                turn_number=turn_number,
-                reward_value=reward_value,
-                reward_type=reward_type,
-                key=key,
-                annotation=annotation or {},
-                source=source,
-            )
-            sess.add(row)
-            await sess.flush()
-            await sess.commit()
-            return row.id
-
-    async def get_outcome_rewards(self) -> list[dict[str, Any]]:
-        async with self.session() as sess:
-            result = await sess.execute(select(DBOutcomeReward))
-            rows = result.scalars().all()
-            return [
-                {
-                    "id": r.id,
-                    "session_id": r.session_id,
-                    "total_reward": r.total_reward,
-                    "achievements_count": r.achievements_count,
-                    "total_steps": r.total_steps,
-                    "created_at": r.created_at,
-                }
-                for r in rows
-            ]
-
-    async def get_outcome_rewards_by_min_reward(self, min_reward: int) -> list[str]:
-        async with self.session() as sess:
-            result = await sess.execute(
-                select(DBOutcomeReward.session_id).where(DBOutcomeReward.total_reward >= min_reward)
-            )
-            return [row[0] for row in result.all()]