📊 Logs, Observabilidade e Monitoramento

// Tabela de precos por modelo (por token)
const PRICING = {
  'gpt-4o':          { input: 2.50 / 1e6, output: 10.00 / 1e6 },
  'claude-sonnet-4': { input: 3.00 / 1e6, output: 15.00 / 1e6 },
  'gpt-4o-mini':     { input: 0.15 / 1e6, output: 0.60 / 1e6 },
};

function calculateCost(model, tokens) {
  const prices = PRICING[model];
  if (!prices) return 0;
  return (tokens.input * prices.input) + (tokens.output * prices.output);
}

// Gravar custo a cada request
async function trackUsage(userId, agentId, model, tokens, latencyMs) {
  const cost = calculateCost(model, tokens);
  await db.insert('usage_log', {
    user_id: userId,
    agent_id: agentId,
    model,
    input_tokens: tokens.input,
    output_tokens: tokens.output,
    cost_usd: cost,
    latency_ms: latencyMs,
    created_at: new Date()
  });
}

// Query: custo por usuario nos ultimos 30 dias
const costByUser = await db.query(`
  SELECT user_id, SUM(cost_usd) as total_cost,
         SUM(input_tokens + output_tokens) as total_tokens,
         COUNT(*) as total_requests
  FROM usage_log
  WHERE created_at > NOW() - INTERVAL '30 days'
  GROUP BY user_id
  ORDER BY total_cost DESC
  LIMIT 20
`);

// Funcao de alerta via Telegram
async function sendAlert(severity, message) {
  const emoji = { warning: '⚠️', high: '🔔', critical: '🚨' };
  const text = `${emoji[severity]} [${severity.toUpperCase()}]\n\n${message}`;

  await fetch(`https://api.telegram.org/bot${process.env.ALERT_BOT_TOKEN}/sendMessage`, {
    method: 'POST',
    headers: { 'Content-Type': 'application/json' },
    body: JSON.stringify({
      chat_id: process.env.ALERT_CHAT_ID,
      text,
      parse_mode: 'HTML'
    })
  });
}

// Checagem periodica (roda a cada 5 min via cron ou setInterval)
async function checkAlerts() {
  const last5min = await db.query(`
    SELECT COUNT(*) as total,
           SUM(CASE WHEN status = 'error' THEN 1 ELSE 0 END) as errors,
           AVG(latency_ms) as avg_latency,
           SUM(cost_usd) as cost
    FROM usage_log
    WHERE created_at > NOW() - INTERVAL '5 minutes'
  `);

  const errorRate = last5min.errors / last5min.total;
  if (errorRate > 0.05) {
    await sendAlert('high',
      `Error rate em ${(errorRate * 100).toFixed(1)}%\n` +
      `${last5min.errors}/${last5min.total} requests falharam nos ultimos 5min`
    );
  }

  if (last5min.avg_latency > 15000) {
    await sendAlert('warning',
      `Latencia media em ${(last5min.avg_latency/1000).toFixed(1)}s\n` +
      `Threshold: 15s. Investigar provider de IA.`
    );
  }
}

// Prompts versionados como objetos
const PROMPTS = {
  'agent-v1': {
    version: '1.0.0',
    system: 'You are a helpful assistant. Be concise.',
    temperature: 0.3,
  },
  'agent-v2': {
    version: '2.0.0',
    system: 'You are an expert assistant. Think step by step. Be precise.',
    temperature: 0.2,
  },
};

// A/B test: 50/50 split
function getActivePrompt(userId) {
  const hash = hashString(userId); // Deterministic per user
  const variant = hash % 2 === 0 ? 'agent-v1' : 'agent-v2';
  return { ...PROMPTS[variant], variant };
}

// Log qual variante foi usada
logger.info({
  userId,
  promptVariant: prompt.variant,
  promptVersion: prompt.version,
  tokens: result.usage,
  latencyMs: elapsed,
}, 'AB test request completed');

// Depois: query para comparar performance
// SELECT prompt_variant, AVG(latency_ms), AVG(cost_usd),
//   SUM(CASE WHEN status='error' THEN 1 ELSE 0 END) / COUNT(*) as error_rate
// FROM usage_log WHERE created_at > NOW() - INTERVAL '7 days'
// GROUP BY prompt_variant