breakpilot-pwa/backend/llm_gateway/routes/health.py

"""
Health Check Route.
"""

import logging
from datetime import datetime
from fastapi import APIRouter
from pydantic import BaseModel

from ..config import get_config

logger = logging.getLogger(__name__)

router = APIRouter(tags=["Health"])


class ComponentStatus(BaseModel):
    """Status einer Komponente."""
    name: str
    status: str  # healthy, degraded, unhealthy
    message: str = ""


class HealthResponse(BaseModel):
    """Health Check Response."""
    status: str  # ok, degraded, error
    ts: str
    version: str
    components: list[ComponentStatus]


@router.get("/health", response_model=HealthResponse)
async def health_check():
    """
    Health Check Endpoint.

    Prüft den Status aller Komponenten:
    - Gateway selbst
    - LLM Backend Erreichbarkeit
    - Datenbank (wenn konfiguriert)
    """
    config = get_config()
    components = []
    overall_status = "ok"

    # Gateway selbst
    components.append(ComponentStatus(
        name="gateway",
        status="healthy",
        message="Gateway is running",
    ))

    # Ollama Backend
    if config.ollama and config.ollama.enabled:
        try:
            import httpx
            async with httpx.AsyncClient(timeout=5.0) as client:
                response = await client.get(f"{config.ollama.base_url}/api/tags")
                if response.status_code == 200:
                    components.append(ComponentStatus(
                        name="ollama",
                        status="healthy",
                        message="Ollama is reachable",
                    ))
                else:
                    components.append(ComponentStatus(
                        name="ollama",
                        status="degraded",
                        message=f"Ollama returned status {response.status_code}",
                    ))
                    overall_status = "degraded"
        except Exception as e:
            components.append(ComponentStatus(
                name="ollama",
                status="unhealthy",
                message=f"Cannot reach Ollama: {str(e)}",
            ))
            # Nicht critical wenn andere Backends verfügbar
            if not (config.vllm and config.vllm.enabled) and not (config.anthropic and config.anthropic.enabled):
                overall_status = "error"

    # vLLM Backend
    if config.vllm and config.vllm.enabled:
        try:
            import httpx
            headers = {}
            if config.vllm.api_key:
                headers["Authorization"] = f"Bearer {config.vllm.api_key}"
            async with httpx.AsyncClient(timeout=5.0) as client:
                response = await client.get(
                    f"{config.vllm.base_url}/v1/models",
                    headers=headers,
                )
                if response.status_code == 200:
                    components.append(ComponentStatus(
                        name="vllm",
                        status="healthy",
                        message="vLLM is reachable",
                    ))
                else:
                    components.append(ComponentStatus(
                        name="vllm",
                        status="degraded",
                        message=f"vLLM returned status {response.status_code}",
                    ))
                    overall_status = "degraded"
        except Exception as e:
            components.append(ComponentStatus(
                name="vllm",
                status="unhealthy",
                message=f"Cannot reach vLLM: {str(e)}",
            ))

    # Anthropic Backend
    if config.anthropic and config.anthropic.enabled:
        components.append(ComponentStatus(
            name="anthropic",
            status="healthy",
            message="Anthropic API configured (not checked)",
        ))

    return HealthResponse(
        status=overall_status,
        ts=datetime.utcnow().isoformat() + "Z",
        version="0.1.0",
        components=components,
    )