// CRITICALimport type { Hono } from "hono";import { performance } from "node:perf_hooks";import type { AppContext } from "../types/context";import { getGpuInfo } from "../services/gpu"; /** * Register monitoring routes. * @param app - Hono app. * @param context - App context. */export const registerMonitoringRoutes = (app: Hono, context: AppContext): void => {  app.get("/metrics", async (_ctx) => {    const current = await context.processManager.findInferenceProcess(context.config.inference_port);    if (current) {      context.metrics.updateActiveModel(current.model_path, current.backend, current.served_model_name);    } else {      context.metrics.updateActiveModel();    }     const gpus = getGpuInfo();    context.metrics.updateGpuMetrics(gpus.map((gpu) => ({ ...gpu })));    context.metrics.updateSseMetrics(context.eventManager.getStats());     const content = await context.metricsRegistry.getMetrics();    return new Response(content, {      headers: { "Content-Type": context.metricsRegistry.contentType },    });  });   app.get("/peak-metrics", async (ctx) => {    const modelId = ctx.req.query("model_id");    if (modelId) {      const result = context.stores.peakMetricsStore.get(modelId);      return ctx.json(result ?? { error: "No metrics for this model" });    }    return ctx.json({ metrics: context.stores.peakMetricsStore.getAll() });  });   app.get("/lifetime-metrics", async (ctx) => {    const data = context.stores.lifetimeMetricsStore.getAll();    const uptimeHours = (data["uptime_seconds"] ?? 0) / 3600;    const energyKwh = (data["energy_wh"] ?? 0) / 1000;    const tokens = data["tokens_total"] ?? 0;    const kwhPerMillion = tokens > 0 ? energyKwh / (tokens / 1_000_000) : 0;    const gpus = getGpuInfo();    const currentPower = gpus.reduce((sum, gpu) => sum + gpu.power_draw, 0);     return ctx.json({      tokens_total: Math.floor(data["tokens_total"] ?? 0),      requests_total: Math.floor(data["requests_total"] ?? 0),      energy_wh: data["energy_wh"] ?? 0,      energy_kwh: energyKwh,      uptime_seconds: data["uptime_seconds"] ?? 0,      uptime_hours: uptimeHours,      first_started_at: data["first_started_at"] ?? 0,      kwh_per_million_tokens: kwhPerMillion,      current_power_watts: currentPower,    });  });   app.post("/benchmark", async (ctx) => {    const promptTokens = Number(ctx.req.query("prompt_tokens") ?? 1000);    const maxTokens = Number(ctx.req.query("max_tokens") ?? 100);    const current = await context.processManager.findInferenceProcess(context.config.inference_port);    if (!current) {      return ctx.json({ error: "No model running" });    }    const modelId = current.served_model_name ?? current.model_path?.split("/").pop() ?? "unknown";    const prompt = `Please count: ${Array.from({ length: Math.floor(promptTokens / 2) }).map((_, index) => index.toString()).join(" ")}`;     try {      const start = performance.now();      const response = await fetch(`http://localhost:${context.config.inference_port}/v1/chat/completions`, {        method: "POST",        headers: { "Content-Type": "application/json" },        body: JSON.stringify({          model: modelId,          messages: [{ role: "user", content: prompt }],          max_tokens: maxTokens,          stream: false,        }),      });      const totalTime = (performance.now() - start) / 1000;      if (!response.ok) {        return ctx.json({ error: `Request failed: ${response.status}` });      }      const data = (await response.json()) as { usage?: Record<string, number> };      const usage = data.usage ?? {};      const promptTokensActual = usage["prompt_tokens"] ?? 0;      const completionTokens = usage["completion_tokens"] ?? 0;       if (completionTokens > 0 && promptTokensActual > 0) {        // Calculate generation throughput from total time        // Note: This includes prefill time so it's a conservative estimate        // Real-time metrics collector tracks actual generation throughput more accurately        const generationTps = completionTokens / totalTime;         // Don't fake prefill - it requires TTFT measurement from streaming        const result = context.stores.peakMetricsStore.updateIfBetter(          modelId,          undefined, // prefill requires proper TTFT measurement          generationTps,          undefined, // TTFT requires streaming measurement        );        context.stores.peakMetricsStore.addTokens(modelId, completionTokens, 1);         return ctx.json({          success: true,          model_id: modelId,          benchmark: {            prompt_tokens: promptTokensActual,            completion_tokens: completionTokens,            total_time_s: Math.round(totalTime * 100) / 100,            generation_tps: Math.round(generationTps * 10) / 10,          },          peak_metrics: result,        });      }      return ctx.json({ error: "No tokens in response" });    } catch (error) {      return ctx.json({ error: String(error) });    }  });};