import { fetchLocal } from "../../http/local-fetch"; export type EngineScrape = {  status: number;  metrics: Record<string, number>;  modelName: string | null;  hasVllm: boolean;  hasSglang: boolean;}; export const scrapeEngineMetrics = async (  port: number,  timeoutMs: number,): Promise<EngineScrape> => {  const scrape: EngineScrape = {    status: 0,    metrics: {},    modelName: null,    hasVllm: false,    hasSglang: false,  };  try {    const response = await fetchLocal(port, "/metrics", { timeoutMs });    scrape.status = response.status;    if (response.status !== 200) return scrape;    const text = await response.text();    for (const line of text.split("\n")) {      if (line.startsWith("#") || line.trim().length === 0) continue;      if (!scrape.hasVllm && line.startsWith("vllm:")) scrape.hasVllm = true;      if (!scrape.hasSglang && line.startsWith("sglang:")) scrape.hasSglang = true;      if (!scrape.modelName) {        const label = line.match(/(?:served_model_name|model_name)="([^"]+)"/);        if (label?.[1]) scrape.modelName = label[1];      }      const match = line.match(/^([a-zA-Z_:][a-zA-Z0-9_:]*)\{?[^}]*\}?\s+([\d.eE+-]+)$/);      if (!match?.[1] || !match[2]) continue;      const value = Number(match[2]);      if (Number.isFinite(value)) scrape.metrics[match[1]] = value;    }    return scrape;  } catch {    return scrape;  }}; export type EngineMetricNames = {  promptTokens: string[];  generationTokens: string[];  promptThroughput: string[];  generationThroughput: string[];  runningRequests: string[];  pendingRequests: string[];  kvCacheUsage: string[];  ttftSum: string;  ttftCount: string;}; export const VLLM_METRIC_NAMES: EngineMetricNames = {  promptTokens: ["vllm:prompt_tokens_total"],  generationTokens: ["vllm:generation_tokens_total"],  promptThroughput: ["vllm:prompt_throughput", "vllm:prefill_throughput"],  generationThroughput: ["vllm:gen_throughput", "vllm:generation_throughput"],  runningRequests: ["vllm:num_requests_running"],  pendingRequests: ["vllm:num_requests_waiting"],  kvCacheUsage: ["vllm:kv_cache_usage_perc"],  ttftSum: "vllm:time_to_first_token_seconds_sum",  ttftCount: "vllm:time_to_first_token_seconds_count",}; export const SGLANG_METRIC_NAMES: EngineMetricNames = {  promptTokens: ["sglang:prompt_tokens_total", "sglang:prefill_tokens_total"],  generationTokens: [    "sglang:generation_tokens_total",    "sglang:completion_tokens_total",    "sglang:gen_tokens_total",  ],  promptThroughput: ["sglang:prompt_throughput", "sglang:prefill_throughput"],  generationThroughput: ["sglang:gen_throughput", "sglang:generation_throughput"],  runningRequests: ["sglang:num_running_reqs", "sglang:num_requests_running"],  pendingRequests: [    "sglang:num_queue_reqs",    "sglang:num_pending_reqs",    "sglang:num_requests_waiting",  ],  kvCacheUsage: ["sglang:token_usage", "sglang:kv_cache_usage_perc"],  ttftSum: "sglang:time_to_first_token_seconds_sum",  ttftCount: "sglang:time_to_first_token_seconds_count",};