fallback.py

import asyncio
import base64
import httpx
import structlog

logger = structlog.get_logger()

class FallbackRouter:
    """
    Asynchronous fallback routing engine for Sentinel.
    Provides backup API inference paths (OpenRouter, OpenBMB, OpenAI) 
    when local edge/GPU Modal execution fails or times out.
    """

    def __init__(self, openrouter_key: str, openbmb_key: str, openai_key: str):
        """
        Initializes the FallbackRouter with required keys.
        """
        self.openrouter_key = openrouter_key
        self.openbmb_key = openbmb_key
        self.openai_key = openai_key
        
        # Initialize an AsyncClient with a 30s timeout
        self.client = httpx.AsyncClient(timeout=30.0)

    async def close(self):
        """
        Closes the HTTP client session.
        """
        await self.client.aclose()

    async def _handle_http_error(self, provider_name: str, error: Exception) -> bool:
        """
        Helper method to handle exceptions. 
        If a rate limit (HTTP 429) is encountered, waits 2 seconds.
        Returns True if a retry/fallback should continue, False otherwise.
        """
        if isinstance(error, httpx.HTTPStatusError):
            status_code = error.response.status_code
            logger.warn(
                "HTTP Error during fallback call", 
                provider=provider_name, 
                status_code=status_code, 
                error=str(error)
            )
            if status_code == 429:
                logger.info("Rate limit (429) hit. Backing off for 2.0s...", provider=provider_name)
                await asyncio.sleep(2.0)
        elif isinstance(error, httpx.TimeoutException):
            logger.warn("Request timed out during fallback call", provider=provider_name, error=str(error))
        else:
            logger.error("Unexpected error during fallback call", provider=provider_name, error=str(error))
        
        return True

    async def fallback_reason(self, prompt: str, system_prompt: str = "") -> dict:
        """
        Executes text reasoning using fallback models.
        Primary: OpenRouter (nvidia/nemotron-3-nano-30b-a3b:free) — 30B MoE, free tier
        Secondary: OpenAI (gpt-4o-mini)
        """
        fallback_chain = []
        
        # 1. Try OpenRouter (Nemotron-3-Nano-30B-A3B MoE, free tier)
        if self.openrouter_key:
            fallback_chain.append("openrouter")
            try:
                logger.info("Attempting reasoning fallback via OpenRouter (Nemotron-30B-A3B)...")
                headers = {
                    "Authorization": f"Bearer {self.openrouter_key}",
                    "Content-Type": "application/json",
                    "HTTP-Referer": "https://huggingface.co/spaces",
                    "X-Title": "Sentinel Safety Guardian"
                }
                payload = {
                    "model": "nvidia/nemotron-3-nano-30b-a3b:free",
                    "messages": [
                        {"role": "system", "content": system_prompt},
                        {"role": "user", "content": prompt}
                    ],
                    "max_tokens": 512,
                    "temperature": 0.4
                }
                
                response = await self.client.post(
                    "https://openrouter.ai/api/v1/chat/completions",
                    headers=headers,
                    json=payload
                )
                response.raise_for_status()
                result = response.json()
                
                text = result["choices"][0]["message"]["content"]
                tokens = result.get("usage", {}).get("total_tokens", 0)
                
                logger.info("OpenRouter reasoning fallback successful", model="nemotron-3-nano-30b-a3b:free")
                return {
                    "text": text,
                    "tokens": tokens,
                    "model": "openrouter:nemotron-3-nano-30b-a3b:free",
                    "fallback_chain": fallback_chain
                }
            except Exception as e:
                await self._handle_http_error("openrouter", e)
        else:
            logger.warn("OpenRouter API key missing. Skipping to next fallback.")

        # 2. Try OpenAI (gpt-4o-mini)
        if self.openai_key:
            fallback_chain.append("openai")
            try:
                logger.info("Attempting reasoning fallback via OpenAI (gpt-4o-mini)...")
                headers = {
                    "Authorization": f"Bearer {self.openai_key}",
                    "Content-Type": "application/json"
                }
                payload = {
                    "model": "gpt-4o-mini",
                    "messages": [
                        {"role": "system", "content": system_prompt},
                        {"role": "user", "content": prompt}
                    ]
                }
                
                response = await self.client.post(
                    "https://api.openai.com/v1/chat/completions",
                    headers=headers,
                    json=payload
                )
                response.raise_for_status()
                result = response.json()
                
                text = result["choices"][0]["message"]["content"]
                tokens = result.get("usage", {}).get("total_tokens", 0)
                
                logger.info("OpenAI reasoning fallback successful")
                return {
                    "text": text,
                    "tokens": tokens,
                    "model": "openai:gpt-4o-mini",
                    "fallback_chain": fallback_chain
                }
            except Exception as e:
                await self._handle_http_error("openai", e)
        else:
            logger.warn("OpenAI API key missing. Cannot execute backup reasoning fallback.")

        # If all providers fail
        logger.error("All reasoning fallbacks failed.")
        return {
            "error": "all_providers_failed",
            "text": "I'm having trouble thinking right now, please try again in a moment.",
            "fallback_chain": fallback_chain
        }

    async def fallback_see(self, image_base64: str, question: str) -> dict:
        """
        Executes vision-language inference using fallback models.
        Primary: OpenBMB API (MiniCPM-V-4.6-Instruct)
        Secondary: OpenAI (gpt-4o-mini)
        """
        fallback_chain = []
        
        # Ensure correct prefix for base64 image data URI
        if not image_base64.startswith("data:"):
            image_data_uri = f"data:image/jpeg;base64,{image_base64}"
        else:
            image_data_uri = image_base64

        # 1. Try OpenBMB (ModelBest API)
        if self.openbmb_key:
            fallback_chain.append("openbmb")
            try:
                logger.info("Attempting vision fallback via OpenBMB (ModelBest)...")
                headers = {
                    "Authorization": f"Bearer {self.openbmb_key}",
                    "Content-Type": "application/json"
                }
                payload = {
                    "model": "MiniCPM-V-4.6-Instruct",
                    "messages": [
                        {
                            "role": "user",
                            "content": [
                                {"type": "image_url", "image_url": {"url": image_data_uri}},
                                {"type": "text", "text": question}
                            ]
                        }
                    ]
                }
                
                response = await self.client.post(
                    "https://api.modelbest.cn/v1/chat/completions",
                    headers=headers,
                    json=payload
                )
                response.raise_for_status()
                result = response.json()
                
                text = result["choices"][0]["message"]["content"]
                tokens = result.get("usage", {}).get("total_tokens", 0)
                
                logger.info("OpenBMB vision fallback successful")
                return {
                    "text": text,
                    "tokens": tokens,
                    "model": "openbmb:minicpm-v-4.6",
                    "fallback_chain": fallback_chain
                }
            except Exception as e:
                await self._handle_http_error("openbmb", e)
        else:
            logger.warn("OpenBMB API key missing. Skipping to next fallback.")

        # 2. Try OpenAI (gpt-4o-mini vision)
        if self.openai_key:
            fallback_chain.append("openai")
            try:
                logger.info("Attempting vision fallback via OpenAI (gpt-4o-mini)...")
                headers = {
                    "Authorization": f"Bearer {self.openai_key}",
                    "Content-Type": "application/json"
                }
                payload = {
                    "model": "gpt-4o-mini",
                    "messages": [
                        {
                            "role": "user",
                            "content": [
                                {"type": "text", "text": question},
                                {"type": "image_url", "image_url": {"url": image_data_uri}}
                            ]
                        }
                    ]
                }
                
                response = await self.client.post(
                    "https://api.openai.com/v1/chat/completions",
                    headers=headers,
                    json=payload
                )
                response.raise_for_status()
                result = response.json()
                
                text = result["choices"][0]["message"]["content"]
                tokens = result.get("usage", {}).get("total_tokens", 0)
                
                logger.info("OpenAI vision fallback successful")
                return {
                    "text": text,
                    "tokens": tokens,
                    "model": "openai:gpt-4o-mini-vision",
                    "fallback_chain": fallback_chain
                }
            except Exception as e:
                await self._handle_http_error("openai", e)
        else:
            logger.warn("OpenAI API key missing. Cannot execute backup vision fallback.")

        # If all providers fail
        logger.error("All vision fallbacks failed.")
        return {
            "error": "all_providers_failed",
            "text": "I'm having trouble seeing right now, please check my connection and try again.",
            "fallback_chain": fallback_chain
        }

    async def fallback_tts(self, text: str) -> dict:
        """
        Synthesizes warning announcement audio using OpenAI TTS API.
        Model: tts-1, Voice: alloy
        """
        if not self.openai_key:
            logger.error("Cannot perform TTS fallback: OpenAI API key is missing.")
            return {"error": "openai_key_missing", "text": "OpenAI API key missing."}
            
        try:
            logger.info("Attempting TTS fallback via OpenAI...")
            headers = {
                "Authorization": f"Bearer {self.openai_key}",
                "Content-Type": "application/json"
            }
            payload = {
                "model": "tts-1",
                "input": text,
                "voice": "alloy"
            }
            
            response = await self.client.post(
                "https://api.openai.com/v1/audio/speech",
                headers=headers,
                json=payload
            )
            response.raise_for_status()
            
            audio_bytes = response.content
            audio_base64 = base64.b64encode(audio_bytes).decode('utf-8')
            
            logger.info("OpenAI TTS fallback successful")
            return {
                "audio_base64": audio_base64,
                "model": "openai-tts-1"
            }
        except Exception as e:
            logger.error("OpenAI TTS fallback failed", error=str(e))
            return {"error": f"TTS fallback failed: {str(e)}"}