import logging
from typing import List
from time import time

from sentence_transformers import SentenceTransformer
import torch

from config import CONFIG
from metrics import prom_metrics


logger = logging.getLogger(__name__)


class EmbeddingService:
    def __init__(self) -> None:
        self.model = None

    def initialized(self):
        return self.model is not None

    def initialize(self) -> None:
        device = CONFIG.embedding_device

        if device == "cuda" and not torch.cuda.is_available():
            device = "cpu"
            logger.warning("CUDA requested but not availaable, falling back to CPU")

        if device == "cuda":
            logger.info("Using CUDA")
        else:
            logger.info("Using CPU")

        self.model = SentenceTransformer(CONFIG.embedding_model, device=device)

    def encode(self, text: str) -> List[float]:
        if not text or not text.strip():
            return [0.0] * CONFIG.embedding_size

        status = "error"
        start_time = time()
        try:
            vector = self.model.encode(text, convert_to_numpy=True)
            status = "ok"
            return vector.tolist()
        except Exception as e:
            logger.error(f"Error getting embedding: {e}")
            raise e
        finally:
            prom_metrics.embedding_performed.labels(status=status).inc()
            prom_metrics.embedding_duration.labels(status=status).observe(
                time() - start_time
            )


EMBEDDING_SERVICE = EmbeddingService()