evotraders/backend/api/runtime.py

# -*- coding: utf-8 -*-
"""Runtime API routes - Control Plane for managing Gateway processes."""

from __future__ import annotations

import asyncio
import json
import logging
import os
import signal
import subprocess
import sys
from datetime import datetime
from pathlib import Path
from typing import Any, Dict, List, Optional

logger = logging.getLogger(__name__)

from fastapi import APIRouter, HTTPException, BackgroundTasks
from pydantic import BaseModel, Field

from backend.runtime.agent_runtime import AgentRuntimeState
from backend.runtime.manager import TradingRuntimeManager, get_global_runtime_manager

router = APIRouter(prefix="/api/runtime", tags=["runtime"])

runtime_manager: Optional[TradingRuntimeManager] = None
PROJECT_ROOT = Path(__file__).resolve().parents[2]

# Gateway process management
_gateway_process: Optional[subprocess.Popen] = None
_gateway_port: int = 8765


class RunContextResponse(BaseModel):
    config_name: str
    run_dir: str
    bootstrap_values: Dict[str, Any]


class RuntimeAgentState(BaseModel):
    agent_id: str
    status: str
    last_session: Optional[str] = None
    last_updated: str


class RuntimeAgentsResponse(BaseModel):
    agents: List[RuntimeAgentState]


class RuntimeEvent(BaseModel):
    timestamp: str
    event: str
    details: Dict[str, Any]
    session: Optional[str]


class RuntimeEventsResponse(BaseModel):
    events: List[RuntimeEvent]


class LaunchConfig(BaseModel):
    """Configuration for launching a new trading task."""
    tickers: List[str] = Field(default_factory=list, description="股票池")
    schedule_mode: str = Field(default="daily", description="调度模式: daily, interval")
    interval_minutes: int = Field(default=60, ge=1, description="间隔分钟数")
    trigger_time: str = Field(default="09:30", description="触发时间 HH:MM")
    max_comm_cycles: int = Field(default=2, ge=1, description="最大会商轮数")
    initial_cash: float = Field(default=100000.0, gt=0, description="初始资金")
    margin_requirement: float = Field(default=0.0, ge=0, description="保证金要求")
    enable_memory: bool = Field(default=False, description="是否启用长期记忆")
    mode: str = Field(default="live", description="运行模式: live, backtest")
    start_date: Optional[str] = Field(default=None, description="回测开始日期 YYYY-MM-DD")
    end_date: Optional[str] = Field(default=None, description="回测结束日期 YYYY-MM-DD")
    poll_interval: int = Field(default=10, ge=1, le=300, description="市场数据轮询间隔(秒)")
    enable_mock: bool = Field(default=False, description="是否启用模拟模式(使用模拟价格数据)")


class LaunchResponse(BaseModel):
    run_id: str
    status: str
    run_dir: str
    gateway_port: int
    message: str


class StopResponse(BaseModel):
    status: str
    message: str


class GatewayStatusResponse(BaseModel):
    is_running: bool
    port: int
    run_id: Optional[str] = None


def _generate_run_id() -> str:
    """Generate timestamp-based run ID: YYYYMMDD_HHMMSS"""
    return datetime.now().strftime("%Y%m%d_%H%M%S")


def _get_run_dir(run_id: str) -> Path:
    """Return the run directory for a given run ID."""
    return PROJECT_ROOT / "runs" / run_id


def _find_available_port(start_port: int = 8765, max_port: int = 9000) -> int:
    """Find an available port for Gateway."""
    import socket
    for port in range(start_port, max_port):
        with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
            if s.connect_ex(('localhost', port)) != 0:
                return port
    raise RuntimeError("No available port found")


def _is_gateway_running() -> bool:
    """Check if Gateway process is running."""
    global _gateway_process
    if _gateway_process is None:
        return False
    return _gateway_process.poll() is None


def _stop_gateway() -> bool:
    """Stop the Gateway process."""
    global _gateway_process
    if _gateway_process is None:
        return False

    try:
        # Try graceful shutdown first
        _gateway_process.terminate()
        try:
            _gateway_process.wait(timeout=5)
        except subprocess.TimeoutExpired:
            # Force kill if graceful shutdown fails
            _gateway_process.kill()
            _gateway_process.wait()
    except Exception as e:
        logger.warning(f"Error during gateway shutdown: {e}")
    finally:
        _gateway_process = None

    return True


def _start_gateway_process(
    run_id: str,
    run_dir: Path,
    bootstrap: Dict[str, Any],
    port: int
) -> subprocess.Popen:
    """Start Gateway as a separate process."""
    # Prepare environment
    env = os.environ.copy()

    # Create command arguments
    cmd = [
        sys.executable,
        "-m", "backend.gateway_server",
        "--run-id", run_id,
        "--run-dir", str(run_dir),
        "--port", str(port),
        "--bootstrap", json.dumps(bootstrap)
    ]

    # Start process
    process = subprocess.Popen(
        cmd,
        env=env,
        stdout=subprocess.PIPE,
        stderr=subprocess.PIPE,
        cwd=PROJECT_ROOT
    )

    return process


@router.get("/context", response_model=RunContextResponse)
async def get_run_context() -> RunContextResponse:
    """Return the most recent run context."""
    snapshot_path = PROJECT_ROOT.glob("runs/*/state/runtime_state.json")
    snapshots = sorted(snapshot_path, key=lambda p: p.stat().st_mtime, reverse=True)

    if not snapshots:
        raise HTTPException(status_code=404, detail="No run context available")

    latest = json.loads(snapshots[0].read_text(encoding="utf-8"))
    context = latest.get("context")
    if context is None:
        raise HTTPException(status_code=404, detail="Run context is not ready")

    return RunContextResponse(
        config_name=context["config_name"],
        run_dir=context["run_dir"],
        bootstrap_values=context["bootstrap_values"],
    )


@router.get("/agents", response_model=RuntimeAgentsResponse)
async def get_runtime_agents() -> RuntimeAgentsResponse:
    """Return agent states from the most recent run."""
    snapshot_path = PROJECT_ROOT.glob("runs/*/state/runtime_state.json")
    snapshots = sorted(snapshot_path, key=lambda p: p.stat().st_mtime, reverse=True)

    if not snapshots:
        raise HTTPException(status_code=404, detail="No runtime state available")

    latest = json.loads(snapshots[0].read_text(encoding="utf-8"))
    agents = latest.get("agents", [])

    return RuntimeAgentsResponse(
        agents=[RuntimeAgentState(**a) for a in agents]
    )


@router.get("/events", response_model=RuntimeEventsResponse)
async def get_runtime_events() -> RuntimeEventsResponse:
    """Return events from the most recent run."""
    snapshot_path = PROJECT_ROOT.glob("runs/*/state/runtime_state.json")
    snapshots = sorted(snapshot_path, key=lambda p: p.stat().st_mtime, reverse=True)

    if not snapshots:
        raise HTTPException(status_code=404, detail="No runtime state available")

    latest = json.loads(snapshots[0].read_text(encoding="utf-8"))
    events = latest.get("events", [])

    return RuntimeEventsResponse(
        events=[RuntimeEvent(**e) for e in events]
    )


@router.get("/gateway/status", response_model=GatewayStatusResponse)
async def get_gateway_status() -> GatewayStatusResponse:
    """Get Gateway process status and port."""
    global _gateway_port

    is_running = _is_gateway_running()
    run_id = None

    if is_running:
        # Try to find run_id from runtime state
        snapshot_path = PROJECT_ROOT.glob("runs/*/state/runtime_state.json")
        snapshots = sorted(snapshot_path, key=lambda p: p.stat().st_mtime, reverse=True)
        if snapshots:
            try:
                latest = json.loads(snapshots[0].read_text(encoding="utf-8"))
                run_id = latest.get("context", {}).get("config_name")
            except Exception as e:
                logger.warning(f"Failed to parse latest snapshot: {e}")

    return GatewayStatusResponse(
        is_running=is_running,
        port=_gateway_port,
        run_id=run_id
    )


@router.get("/gateway/port")
async def get_gateway_port() -> Dict[str, Any]:
    """Get WebSocket Gateway port for frontend connection."""
    global _gateway_port
    return {
        "port": _gateway_port,
        "is_running": _is_gateway_running(),
        "ws_url": f"ws://localhost:{_gateway_port}"
    }


@router.post("/start", response_model=LaunchResponse)
async def start_runtime(
    config: LaunchConfig,
    background_tasks: BackgroundTasks
) -> LaunchResponse:
    """Start a new trading runtime with the given configuration.

    1. Stop existing Gateway if running
    2. Generate run ID and directory
    3. Create runtime manager
    4. Start Gateway as subprocess (Data Plane)
    5. Return Gateway port for WebSocket connection
    """
    global _gateway_process, _gateway_port

    # 1. Stop existing Gateway
    if _is_gateway_running():
        _stop_gateway()
        await asyncio.sleep(1)  # Wait for port release

    # 2. Generate run ID and directory
    run_id = _generate_run_id()
    run_dir = _get_run_dir(run_id)

    # 3. Prepare bootstrap config
    bootstrap = {
        "tickers": config.tickers,
        "schedule_mode": config.schedule_mode,
        "interval_minutes": config.interval_minutes,
        "trigger_time": config.trigger_time,
        "max_comm_cycles": config.max_comm_cycles,
        "initial_cash": config.initial_cash,
        "margin_requirement": config.margin_requirement,
        "enable_memory": config.enable_memory,
        "mode": config.mode,
        "start_date": config.start_date,
        "end_date": config.end_date,
        "poll_interval": config.poll_interval,
        "enable_mock": config.enable_mock,
    }

    # 4. Create runtime manager
    manager = TradingRuntimeManager(
        config_name=run_id,
        run_dir=run_dir,
        bootstrap=bootstrap,
    )
    manager.prepare_run()
    register_runtime_manager(manager)

    # 5. Write BOOTSTRAP.md
    _write_bootstrap_md(run_dir, bootstrap)

    # 6. Find available port and start Gateway process
    _gateway_port = _find_available_port(start_port=8765)

    try:
        _gateway_process = _start_gateway_process(
            run_id=run_id,
            run_dir=run_dir,
            bootstrap=bootstrap,
            port=_gateway_port
        )

        # Wait briefly to check if process started successfully
        await asyncio.sleep(2)

        if not _is_gateway_running():
            stdout, stderr = _gateway_process.communicate(timeout=1)
            _gateway_process = None
            raise HTTPException(
                status_code=500,
                detail=f"Gateway failed to start: {stderr.decode() if stderr else 'Unknown error'}"
            )

    except Exception as e:
        _stop_gateway()
        raise HTTPException(status_code=500, detail=f"Failed to start Gateway: {str(e)}")

    return LaunchResponse(
        run_id=run_id,
        status="started",
        run_dir=str(run_dir),
        gateway_port=_gateway_port,
        message=f"Runtime started with run_id: {run_id}, Gateway on port: {_gateway_port}",
    )


@router.post("/stop", response_model=StopResponse)
async def stop_runtime(force: bool = True) -> StopResponse:
    """Stop the current running runtime."""
    global _gateway_process

    was_running = _is_gateway_running()

    if not was_running:
        raise HTTPException(status_code=404, detail="No runtime is currently running")

    # Stop Gateway process
    _stop_gateway()

    # Unregister runtime manager
    unregister_runtime_manager()

    return StopResponse(
        status="stopped",
        message="Runtime stopped successfully",
    )


@router.post("/restart")
async def restart_runtime(
    config: LaunchConfig,
    background_tasks: BackgroundTasks
):
    """Restart the runtime with a new configuration."""
    # Stop current runtime
    await stop_runtime(force=True)

    # Start new runtime
    response = await start_runtime(config, background_tasks)

    return {
        "run_id": response.run_id,
        "status": "restarted",
        "gateway_port": response.gateway_port,
        "message": f"Runtime restarted with run_id: {response.run_id}",
    }


@router.get("/current")
async def get_current_runtime():
    """Get information about the currently running runtime."""
    if not _is_gateway_running():
        raise HTTPException(status_code=404, detail="No runtime is currently running")

    # Find latest runtime state
    snapshot_path = PROJECT_ROOT.glob("runs/*/state/runtime_state.json")
    snapshots = sorted(snapshot_path, key=lambda p: p.stat().st_mtime, reverse=True)

    if not snapshots:
        raise HTTPException(status_code=404, detail="No runtime information available")

    latest = json.loads(snapshots[0].read_text(encoding="utf-8"))
    context = latest.get("context", {})

    return {
        "run_id": context.get("config_name"),
        "run_dir": context.get("run_dir"),
        "is_running": True,
        "gateway_port": _gateway_port,
        "bootstrap": context.get("bootstrap_values", {}),
    }


def register_runtime_manager(manager: TradingRuntimeManager) -> None:
    """Allow other modules to expose the runtime manager to the API."""
    global runtime_manager
    runtime_manager = manager


def unregister_runtime_manager() -> None:
    """Drop the runtime manager reference."""
    global runtime_manager
    runtime_manager = None


def _write_bootstrap_md(run_dir: Path, bootstrap: Dict[str, Any]) -> None:
    """Write bootstrap configuration to BOOTSTRAP.md."""
    try:
        import yaml
    except ImportError:
        yaml = None

    bootstrap_path = run_dir / "BOOTSTRAP.md"
    bootstrap_path.parent.mkdir(parents=True, exist_ok=True)

    # Filter out None values
    values = {k: v for k, v in bootstrap.items() if v is not None}

    if yaml:
        front_matter = yaml.safe_dump(values, allow_unicode=True, sort_keys=False)
    else:
        front_matter = json.dumps(values, ensure_ascii=False, indent=2)

    content = f"---\n{front_matter}---\n"
    bootstrap_path.write_text(content, encoding="utf-8")