index.js

import express from 'express';
import cors from 'cors';
import { pipeline, env } from '@xenova/transformers';
import path from "path";
import { fileURLToPath } from 'url';
import textToSpeechAPI from './textToSpeech/textToSpeechAPI.js';

const __filename = fileURLToPath(import.meta.url);
const __dirname = path.dirname(__filename);

const app = express();
app.use(cors());
app.use(express.json());

let extractor = null;
let isInitializing = false;

const init = async () => {
  if (extractor || isInitializing) return extractor;
  isInitializing = true;
  console.log('🔄 Inicializando E5...');
  
  try {
    // Configurações para modelo local
   // env.allowRemoteModels = false;
    //env.allowLocalModels = true;
    env.useBrowserCache = false;
    
    // Define o caminho para a pasta de modelos
    //env.localModelPath = path.join(__dirname, 'models');
    
    const modelPath = 'Xenova/multilingual-e5-small';
    
    extractor = await pipeline('feature-extraction', modelPath, {
      quantized: true,
      device: 'cpu',
      local_files_only: false,
    });
    
    console.log('✅ E5 carregado!');
    return extractor;
  } catch (error) {
    console.error('❌ Erro:', error.message);
    throw error;
  } finally {
    isInitializing = false;
  }
};

console.log('🔄 Inicializando TTS...');
textToSpeechAPI(app);
console.log('✅ TTS carregado!');

// Health check mínimo
app.get('/', (req, res) => {
    const uptimeSeconds = process.uptime(); // tempo desde que o servidor iniciou
    const uptime = `${Math.floor(uptimeSeconds / 60)} min ${Math.floor(uptimeSeconds % 60)} sec`;
  
    const now = new Date();
  
    res.json({
      status: extractor ? 'ready' : 'loading',
      model: 'e5-small',
      api_version: 'v1',
      timestamp: now.toISOString(),
      uptime: uptime,
      server_timezone: Intl.DateTimeFormat().resolvedOptions().timeZone,
      requester_ip: req.headers['x-forwarded-for'] || req.socket.remoteAddress
    });
  });

app.get('/ping', (req, res) => res.json({ pong: true }));

// Endpoint principal otimizado
app.post('/api/embed', async (req, res) => {
  try {
    if (!extractor) await init();
    
    const { text, type = 'query' } = req.body;
    if (!text) return res.status(400).json({ error: 'Texto necessário' });
    
    // Processa com configurações de baixa memória
    const result = await extractor([`${type}: ${text}`], {
      pooling: 'mean',
      normalize: true,
      // Reduz batch size para economizar RAM
      batch_size: 1,
      // Usa menos threads
      num_threads: 1
    });
    
    const embedding = Array.from(result[0].data);
    res.json({ embedding });
    
    // Force garbage collection se disponível
    if (global.gc) global.gc();
  } catch (err) {
    console.error('Erro embed:', err.message);
    res.status(500).json({ error: 'Erro interno' });
  }
});

// Batch otimizado para múltiplos textos
app.post('/api/embed/batch', async (req, res) => {
  try {
    if (!extractor) await init();
    
    const { texts, type = 'query' } = req.body;
    if (!Array.isArray(texts)) return res.status(400).json({ error: 'Array necessário' });
    
    // Processa em chunks pequenos para não estourar memória
    const chunkSize = 3;
    const embeddings = [];
    
    for (let i = 0; i < texts.length; i += chunkSize) {
      const chunk = texts.slice(i, i + chunkSize);
      const prefixed = chunk.map(text => `${type}: ${text}`);
      
      const results = await extractor(prefixed, {
        pooling: 'mean',
        normalize: true,
        batch_size: 1,
        num_threads: 1
      });
      
      embeddings.push(...results.map(r => Array.from(r.data)));
      
      // Pequena pausa entre chunks
      await new Promise(resolve => setTimeout(resolve, 10));
    }
    
    res.json({
      embeddings,
      count: embeddings.length
    });
    
    if (global.gc) global.gc();
  } catch (err) {
    console.error('Erro batch:', err.message);
    res.status(500).json({ error: 'Erro interno' });
  }
});

const PORT = process.env.PORT || 7860;

// Timeout agressivo
const timeout = setTimeout(() => {
  if (isInitializing) {
    console.log('⚠️ Timeout - reiniciando');
    process.exit(1);
  }
}, 90000); // 90s

// Inicializa o modelo automaticamente
init().catch(console.error);

app.listen(PORT, '0.0.0.0', () => {
  console.log(`🚀 E5 na porta ${PORT}`);
});

// Cleanup
process.on('SIGTERM', () => {
  console.log('📴 Finalizando...');
  process.exit(0);
});

process.on('uncaughtException', (err) => {
  console.error('💥 Erro crítico:', err.message);
  process.exit(1);
});