http: improve performance of shouldUseProxy

Uzlopak · Uzlopak · commit ef99bf3f3360 · 2025-08-10T01:37:21.000+02:00
diff --git a/benchmark/http/proxy-should-use-proxy.js b/benchmark/http/proxy-should-use-proxy.js
@@ -0,0 +1,60 @@
+'use strict';
+
+const common = require('../common');
+const assert = require('assert');
+
+// Benchmark configuration
+const bench = common.createBenchmark(main, {
+  hostname: [
+    '127.0.0.1',
+    'localhost',
+    'www.example.com',
+    'example.com',
+    'myexample.com',
+  ],
+  no_proxy: [
+    '',
+    '*',
+    '127.0.0.1',
+    'example.com',
+    '.example.com',
+    '*.example.com',
+  ],
+  n: [1e6],
+}, {
+  flags: ['--expose-internals'],
+});
+
+function main({ hostname, no_proxy, n }) {
+  const { parseProxyConfigFromEnv } = require('internal/http');
+
+  const protocol = 'https:';
+  const env = {
+    no_proxy,
+    https_proxy: `https://www.example.proxy`,
+  };
+  const proxyConfig = parseProxyConfigFromEnv(env, protocol);
+
+  // Warm up.
+  const length = 1024;
+  const array = [];
+  for (let i = 0; i < length; ++i) {
+    array.push(proxyConfig.shouldUseProxy(hostname));
+  }
+
+  // // Benchmark
+  bench.start();
+
+  for (let i = 0; i < n; ++i) {
+    const index = i % length;
+    array[index] = proxyConfig.shouldUseProxy(hostname);
+  }
+
+  bench.end(n);
+
+  // Verify the entries to prevent dead code elimination from making
+  // the benchmark invalid.
+  for (let i = 0; i < length; ++i) {
+    assert.strictEqual(typeof array[i], 'boolean');
+  }
+}
diff --git a/lib/internal/http.js b/lib/internal/http.js
@@ -95,6 +95,19 @@ function ipToInt(ip) {
 // When the proxy protocol is HTTPS, the modified request needs to be sent after
 // TLS handshake with the proxy server. Same goes to the HTTPS request tunnel establishment.
 
+/**
+ * @callback ProxyBypassMatchFn
+ * @param {string} host - Host to match against the bypass list.
+ * @param {string} [hostWithPort] - Host with port to match against the bypass list.
+ * @returns {boolean} - True if the host should be bypassed, false otherwise.
+ */
+
+/**
+ * @typedef {object} ProxyConnectionOptions
+ * @property {string} host - Hostname of the proxy server.
+ * @property {number} port - Port of the proxy server.
+ */
+
 /**
  * Represents the proxy configuration for an agent. The built-in http and https agent
  * implementation have one of this when they are configured to use a proxy.
@@ -105,9 +118,28 @@ function ipToInt(ip) {
  * @property {string} protocol - Protocol of the proxy server, e.g. 'http:' or 'https:'.
  * @property {string|undefined} auth - proxy-authorization header value, if username or password is provided.
  * @property {Array<string>} bypassList - List of hosts to bypass the proxy.
- * @property {object} proxyConnectionOptions - Options for connecting to the proxy server.
+ * @property {ProxyConnectionOptions} proxyConnectionOptions - Options for connecting to the proxy server.
  */
 class ProxyConfig {
+  /** @type {Array<string>} */
+  #bypassList = [];
+  /** @type {Array<ProxyBypassMatchFn>} */
+  #bypassMatchFns = [];
+
+  /** @type {ProxyConnectionOptions} */
+  get proxyConnectionOptions() {
+    return {
+      host: this.hostname,
+      port: this.port,
+    };
+  }
+
+  /**
+   * @param {string} proxyUrl - The URL of the proxy server, e.g. 'http://localhost:8080'.
+   * @param {boolean} [keepAlive] - Whether to keep the connection alive.
+   *   This is not used in the current implementation but can be used in the future.
+   * @param {string} [noProxyList] - Comma-separated list of hosts to bypass the proxy.
+   */
   constructor(proxyUrl, keepAlive, noProxyList) {
     const { host, hostname, port, protocol, username, password } = new URL(proxyUrl);
     this.href = proxyUrl; // Full URL of the proxy server.
@@ -121,59 +153,94 @@ class ProxyConfig {
       const auth = `${decodeURIComponent(username)}:${decodeURIComponent(password)}`;
       this.auth = `Basic ${Buffer.from(auth).toString('base64')}`;
     }
+
     if (noProxyList) {
-      this.bypassList = noProxyList.split(',').map((entry) => entry.trim().toLowerCase());
-    } else {
-      this.bypassList = []; // No bypass list provided.
+      this.#bypassList = noProxyList
+        .split(',')
+        .map((entry) => entry.trim().toLowerCase());
     }
-    this.proxyConnectionOptions = {
-      host: this.hostname,
-      port: this.port,
-    };
-  }
 
-  // See: https://about.gitlab.com/blog/we-need-to-talk-no-proxy
-  // TODO(joyeecheung): share code with undici.
-  shouldUseProxy(hostname, port) {
-    const bypassList = this.bypassList;
-    if (this.bypassList.length === 0) {
-      return true; // No bypass list, always use the proxy.
+    if (this.#bypassList.length === 0) {
+      this.shouldUseProxy = () => true; // No bypass list, always use the proxy.
+    } else if (this.#bypassList.includes('*')) {
+      this.shouldUseProxy = () => false; // '*' in the bypass list means to bypass all hosts.
+    } else {
+      this.#buildBypassMatchFns();
+      // Use the bypass match functions to determine if the proxy should be used.
+      this.shouldUseProxy = this.#match.bind(this);
     }
+  }
 
-    const host = hostname.toLowerCase();
-    const hostWithPort = port ? `${host}:${port}` : host;
-
-    for (let i = 0; i < bypassList.length; i++) {
-      const entry = bypassList[i];
-
-      if (entry === '*') return false;  // * bypasses all hosts.
-      if (entry === host || entry === hostWithPort) return false;  // Matching host and host:port
-
-      // Follow curl's behavior: strip leading dot before matching suffixes.
-      if (entry.startsWith('.')) {
-        const suffix = entry.substring(1);
-        if (host.endsWith(suffix)) return false;
+  #buildBypassMatchFns(bypassList = this.#bypassList) {
+    this.#bypassMatchFns = [];
+
+    for (const entry of this.#bypassList) {
+      if (
+        // Handle wildcard entries like *.example.com
+        entry.startsWith('*.') ||
+        // Follow curl's behavior: strip leading dot before matching suffixes.
+        entry.startsWith('.')
+      ) {
+        const suffix = entry.split('');
+        suffix.shift(); // Remove the leading dot or asterisk.
+        const suffixLength = suffix.length;
+        if (suffixLength === 0) {
+          // If the suffix is empty, it means to match all hosts.
+          this.#bypassMatchFns.push(() => true);
+          continue;
+        }
+        this.#bypassMatchFns.push((host) => {
+          const hostLength = host.length;
+          const offset = hostLength - suffixLength;
+          if (offset < 0) return false; // Host is shorter than the suffix.
+          for (let i = 0; i < suffixLength; i++) {
+            if (host[offset + i] !== suffix[i]) {
+              return false;
+            }
+          }
+          return true;
+        });
+        continue;
       }
 
-      // Handle wildcards like *.example.com
-      if (entry.startsWith('*.') && host.endsWith(entry.substring(1))) return false;
-
       // Handle IP ranges (simple format like 192.168.1.0-192.168.1.255)
       // TODO(joyeecheung): support IPv6.
-      if (entry.includes('-') && isIPv4(host)) {
-        let { 0: startIP, 1: endIP } = entry.split('-');
-        startIP = startIP.trim();
-        endIP = endIP.trim();
-        if (startIP && endIP && isIPv4(startIP) && isIPv4(endIP)) {
-          const hostInt = ipToInt(host);
-          const startInt = ipToInt(startIP);
-          const endInt = ipToInt(endIP);
-          if (hostInt >= startInt && hostInt <= endInt) return false;
-        }
+      const { 0: startIP, 1: endIP } = entry.split('-').map((ip) => ip.trim());
+      if (entry.includes('-') && startIP && endIP && isIPv4(startIP) && isIPv4(endIP)) {
+        const startInt = ipToInt(startIP);
+        const endInt = ipToInt(endIP);
+        this.#bypassMatchFns.push((host) => {
+          if (isIPv4(host)) {
+            const hostInt = ipToInt(host);
+            return hostInt >= startInt && hostInt <= endInt;
+          }
+          return false;
+        });
+        continue;
       }
 
-      // It might be useful to support CIDR notation, but it's not so widely supported
-      // in other tools as a de-facto standard to follow, so we don't implement it for now.
+      // Handle simple host or IP entries
+      this.#bypassMatchFns.push((host, hostWithPort) => {
+        return (host === entry || hostWithPort === entry);
+      });
+    }
+  }
+
+  get bypassList() {
+    // Return a copy of the bypass list to prevent external modification.
+    return [...this.#bypassList];
+  }
+
+  // See: https://about.gitlab.com/blog/we-need-to-talk-no-proxy
+  // TODO(joyeecheung): share code with undici.
+  #match(hostname, port) {
+    const host = hostname.toLowerCase();
+    const hostWithPort = port ? `${host}:${port}` : host;
+
+    for (const bypassMatchFn of this.#bypassMatchFns) {
+      if (bypassMatchFn(host, hostWithPort)) {
+        return false; // If any bypass function matches, do not use the proxy.
+      }
     }
 
     return true; // If no matches found, use the proxy.
