fix(cli): stop chrome-headless-shell download hangs from stalling forever

@puppeteer/browsers' download path has no AbortController/timeout anywhere in its chain, and its proxy support is inert (proxy-agent is only an optional peer dependency hyperframes doesn't install) — so on a network that requires a proxy, the download attempts a direct connection and hangs at "Downloading Chrome... 0%" forever instead of failing. Add a stall watchdog (resets on every progress tick, not a total-duration cap, so a slow-but-live download is unaffected) around the install() call, and name HTTP(S)_PROXY as the likely cause in the existing HYPERFRAMES_BROWSER_PATH remediation hint when one is configured. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
2026-09-14 18:01:20 +08:00 · 2026-09-11 19:28:33 +00:00
parent ea7e1dbd0b
commit 4ed25995b1
2 changed files with 290 additions and 9 deletions
@@ -135,7 +135,9 @@ function installPuppeteerBrowsersMock(
    browserPlatform?: string;
    installedInHfCacheError?: Error;
    installResult?: { executablePath: string };
-    installImpl?: () => Promise<{ executablePath: string }>;
+    installImpl?: (installArgs: {
+      downloadProgressCallback?: (downloaded: number, total: number) => void;
+    }) => Promise<{ executablePath: string }>;
  } = {},
 ) {
  vi.doMock("@puppeteer/browsers", () => ({
@@ -909,6 +911,121 @@ describe("installWithCorruptArchiveRecovery", () => {
  });
 });

+// PRINFRA-682: a chrome-headless-shell download that never gets a response
+// (e.g. a network that only egresses through a proxy `@puppeteer/browsers`
+// silently fails to use — see `proxyDownloadStallHint`) hangs at "Downloading
+// Chrome... 0%" forever, with no timeout anywhere in the install() call
+// chain. This guards the watchdog that makes that fail loud instead.
+describe("withDownloadStallGuard", () => {
+  afterEach(() => {
+    vi.useRealTimers();
+  });
+
+  it("resolves normally when the run completes before any stall", async () => {
+    const { withDownloadStallGuard } = await import("./manager.js");
+    const result = await withDownloadStallGuard(
+      async (onProgress) => {
+        onProgress(50, 100);
+        onProgress(100, 100);
+        return "done";
+      },
+      { timeoutMs: 1_000 },
+    );
+    expect(result).toBe("done");
+  });
+
+  it("rejects if no progress arrives before the timeout (connection never responds)", async () => {
+    vi.useFakeTimers();
+    const { withDownloadStallGuard } = await import("./manager.js");
+
+    // A run that never resolves and never reports progress — the "0% forever"
+    // symptom: no response ever comes back at all.
+    const guarded = withDownloadStallGuard(() => new Promise<void>(() => {}), {
+      timeoutMs: 1_000,
+    });
+    const outcome = guarded.then(
+      () => "resolved",
+      (err: unknown) => (err instanceof Error ? err.message : String(err)),
+    );
+
+    await vi.advanceTimersByTimeAsync(1_000);
+
+    await expect(outcome).resolves.toContain("stalled");
+  });
+
+  it("resets the watchdog on every progress tick, so a slow-but-live download never trips it", async () => {
+    vi.useFakeTimers();
+    const { withDownloadStallGuard } = await import("./manager.js");
+
+    const result = withDownloadStallGuard(
+      async (onProgress) => {
+        for (let i = 1; i <= 5; i += 1) {
+          // Each tick arrives just under the timeout — only safe if the
+          // watchdog resets rather than measuring total elapsed duration.
+          await new Promise((resolve) => setTimeout(resolve, 900));
+          onProgress(i * 20, 100);
+        }
+        return "done";
+      },
+      { timeoutMs: 1_000 },
+    );
+
+    await vi.advanceTimersByTimeAsync(900 * 5 + 10);
+
+    await expect(result).resolves.toBe("done");
+  });
+
+  it("still trips if progress stops arriving partway through (mid-download stall)", async () => {
+    vi.useFakeTimers();
+    const { withDownloadStallGuard } = await import("./manager.js");
+
+    const guarded = withDownloadStallGuard(
+      (onProgress) =>
+        new Promise<void>((resolve) => {
+          onProgress(10, 100);
+          // ...then nothing further ever arrives — resolve() is never called.
+          void resolve;
+        }),
+      { timeoutMs: 1_000 },
+    );
+    const outcome = guarded.then(
+      () => "resolved",
+      (err: unknown) => (err instanceof Error ? err.message : String(err)),
+    );
+
+    await vi.advanceTimersByTimeAsync(1_000);
+
+    await expect(outcome).resolves.toContain("stalled");
+  });
+
+  it("propagates a genuine rejection unchanged, not a stall error", async () => {
+    const { withDownloadStallGuard } = await import("./manager.js");
+    await expect(
+      withDownloadStallGuard(
+        async () => {
+          throw new Error("ENOTFOUND");
+        },
+        { timeoutMs: 1_000 },
+      ),
+    ).rejects.toThrow("ENOTFOUND");
+  });
+
+  it("forwards progress ticks to the caller's onProgress callback", async () => {
+    const { withDownloadStallGuard } = await import("./manager.js");
+    const onProgress = vi.fn();
+    await withDownloadStallGuard(
+      async (tick) => {
+        tick(10, 100);
+        tick(100, 100);
+        return "done";
+      },
+      { timeoutMs: 1_000, onProgress },
+    );
+    expect(onProgress).toHaveBeenNthCalledWith(1, 10, 100);
+    expect(onProgress).toHaveBeenNthCalledWith(2, 100, 100);
+  });
+});
+
 // Sibling failure mode to #2078 (SIGTRAP at launch): the field feedback in
 // #hyperframes-cli-feedback ts 1784055194.202169 (darwin/arm64, HF CLI 0.7.57)
 // hit `All providers failed for chrome-headless-shell 152.0.7928.2` at download
@@ -1007,6 +1124,67 @@ describe("downloadBrowser — install failure surfaces HYPERFRAMES_BROWSER_PATH
      expect((caught as Error).cause).toBe(originalError);
    },
  );
+
+  it("appends a proxy hint when HTTP(S)_PROXY is set", async () => {
+    process.env["HTTPS_PROXY"] = "http://127.0.0.1:18080";
+    installFsMocks({ existing: new Set([CACHE_ROOT]) });
+    installPuppeteerBrowsersMock({
+      installedInHfCache: [],
+      installImpl: async () => {
+        throw new Error("connect ETIMEDOUT");
+      },
+    });
+
+    const { ensureBrowser } = await import("./manager.js");
+    await expect(ensureBrowser()).rejects.toThrow(/HTTPS_PROXY.*is set/s);
+
+    delete process.env["HTTPS_PROXY"];
+  });
+
+  it("omits the proxy hint when no proxy env var is set", async () => {
+    installFsMocks({ existing: new Set([CACHE_ROOT]) });
+    installPuppeteerBrowsersMock({
+      installedInHfCache: [],
+      installImpl: async () => {
+        throw new Error("connect ETIMEDOUT");
+      },
+    });
+
+    const { ensureBrowser } = await import("./manager.js");
+    let caught: unknown;
+    try {
+      await ensureBrowser();
+    } catch (err) {
+      caught = err;
+    }
+    expect((caught as Error).message).not.toContain("HTTPS_PROXY");
+  });
+
+  it("PRINFRA-682: a download that never responds surfaces a stall error with the HYPERFRAMES_BROWSER_PATH hint instead of hanging forever", async () => {
+    vi.useFakeTimers();
+    installFsMocks({ existing: new Set([CACHE_ROOT]) });
+    installPuppeteerBrowsersMock({
+      installedInHfCache: [],
+      // Mirrors the real bug: install() never resolves, rejects, or reports
+      // progress — exactly what a proxy-only network's silently-bypassed
+      // direct connection attempt looks like from the caller's side.
+      installImpl: () => new Promise(() => {}),
+    });
+
+    const { ensureBrowser } = await import("./manager.js");
+    const attempt = ensureBrowser();
+    const outcome = attempt.then(
+      () => "resolved",
+      (err: unknown) => (err instanceof Error ? err.message : String(err)),
+    );
+
+    await vi.advanceTimersByTimeAsync(90_000);
+
+    const message = await outcome;
+    expect(message).toContain("Download stalled");
+    expect(message).toContain("HYPERFRAMES_BROWSER_PATH");
+    vi.useRealTimers();
+  });
 });

 // Regression guard for HF#2103: `hyperframes render` hung forever on macOS
@@ -44,6 +44,21 @@ const CACHE_DIR = join(homedir(), ".cache", "hyperframes", "chrome");
 // too or it silently picks system Chrome over a perfectly good headless-shell.
 const PUPPETEER_CACHE_DIR = join(homedir(), ".cache", "puppeteer", "chrome-headless-shell");

+// `@puppeteer/browsers`' downloadFile() (lib/httpUtil.js) opens a bare
+// node:http(s) request with no AbortController/timeout anywhere in the chain
+// (confirmed by source read of the resolved 3.x package) — a connection that
+// never gets a response (e.g. a network that only egresses through an
+// HTTP(S)_PROXY, which downloadFile's own proxy-agent support silently no-ops
+// on: `proxy-agent` is only an *optional peer* dependency, so the package's
+// dynamic import of it throws, is caught, and the request falls back to a
+// bare agent that never reads proxy env vars) just hangs at "Downloading
+// Chrome... 0%" forever. See PRINFRA-682. This is a stall watchdog, not a
+// total-download-duration cap: it resets on every progress tick, so a
+// slow-but-actually-progressing download (large binary over a slow line)
+// never trips it — only a connection that stops producing any bytes for this
+// long does.
+const DOWNLOAD_STALL_TIMEOUT_MS = 45_000;
+
 // `@puppeteer/browsers`' install() has no concurrency guard of its own — two
 // CLI invocations that both miss the cache at the same time both extract into
 // the same target directory simultaneously. A killed/interrupted extraction
@@ -707,6 +722,62 @@ export async function installWithCorruptArchiveRecovery<T>(
  }
 }

+/**
+ * Race an install-style call against a "no progress" watchdog instead of a
+ * total-duration cap. The timer (re)arms on every progress tick — and once
+ * before the first one, since a connection that never gets a response never
+ * ticks at all — so a download that is merely slow keeps resetting it
+ * indefinitely, while one that has truly stalled (no bytes, ever, or bytes
+ * that stop arriving mid-stream) trips it after `timeoutMs` of silence.
+ *
+ * This cannot abort the underlying socket (`@puppeteer/browsers` exposes no
+ * `AbortSignal`), so a stalled connection lingers until the OS reclaims it —
+ * but the CLI's root exit handler calls `process.exit()` unconditionally once
+ * a command settles (see `cli.ts`'s `registerRootExitRequester`), so
+ * rejecting here still unblocks the process instead of leaving it hung.
+ */
+export function withDownloadStallGuard<T>(
+  run: (onProgress: (downloadedBytes: number, totalBytes: number) => void) => Promise<T>,
+  options: {
+    timeoutMs: number;
+    onProgress?: (downloadedBytes: number, totalBytes: number) => void;
+  },
+): Promise<T> {
+  return new Promise((resolve, reject) => {
+    let settled = false;
+    let timer: ReturnType<typeof setTimeout>;
+    const armTimer = () => {
+      clearTimeout(timer);
+      timer = setTimeout(() => {
+        if (settled) return;
+        settled = true;
+        reject(
+          new Error(`Download stalled: no progress for ${Math.round(options.timeoutMs / 1000)}s.`),
+        );
+      }, options.timeoutMs);
+    };
+    armTimer();
+    run((downloadedBytes, totalBytes) => {
+      if (settled) return;
+      armTimer();
+      options.onProgress?.(downloadedBytes, totalBytes);
+    }).then(
+      (result) => {
+        if (settled) return;
+        settled = true;
+        clearTimeout(timer);
+        resolve(result);
+      },
+      (err: unknown) => {
+        if (settled) return;
+        settled = true;
+        clearTimeout(timer);
+        reject(err);
+      },
+    );
+  });
+}
+
 /**
 * When `@puppeteer/browsers`' install() rejects for any reason the corrupt-
 * archive recovery path can't handle (all CDN providers rejected — the
@@ -731,6 +802,33 @@ function browserPathHintForPlatform(): string {
  return "/usr/bin/google-chrome";
 }

+/**
+ * `@puppeteer/browsers`' download request does attempt real proxy support
+ * (a dynamic import of the `proxy-agent` package), but that package is only
+ * an *optional peer* dependency that a normal hyperframes install never
+ * pulls in — so the import throws, is caught silently, and the request falls back
+ * to a bare node:http(s) agent that never reads HTTP_PROXY/HTTPS_PROXY on
+ * its own. On a network that only egresses through that proxy, the direct
+ * connection attempt doesn't fail fast; it stalls, which is what
+ * `withDownloadStallGuard` above catches. Naming the proxy as the likely
+ * cause (when one is actually configured) turns "download stalled, no idea
+ * why" into an actionable diagnosis instead of just repeating the same
+ * escape hatch with no context.
+ */
+function proxyDownloadStallHint(): string {
+  const proxyConfigured = ["HTTPS_PROXY", "HTTP_PROXY", "https_proxy", "http_proxy"].some((key) =>
+    Boolean(process.env[key]?.trim()),
+  );
+  if (!proxyConfigured) return "";
+  return (
+    `\n\nHTTP_PROXY/HTTPS_PROXY is set in this environment, but chrome-headless-shell's ` +
+    `download does not honor it (the proxy support it depends on is an optional ` +
+    `dependency hyperframes does not install) — on a network that requires the proxy for ` +
+    `internet access, the download attempts a direct connection and stalls instead of ` +
+    `failing fast. Use HYPERFRAMES_BROWSER_PATH above to skip the download entirely.`
+  );
+}
+
 function wrapDownloadFailureWithBrowserPathHint(cause: unknown): Error {
  const original = normalizeErrorMessage(cause);
  const example = browserPathHintForPlatform();
@@ -741,7 +839,8 @@ function wrapDownloadFailureWithBrowserPathHint(cause: unknown): Error {
    `Then re-run your command. Any Chrome build works for the screenshot ` +
    `capture path; install a real chrome-headless-shell later if you need the ` +
    `perf-optimized BeginFrame path. Alternatively, run inside the hyperframes ` +
-    `Docker image which ships a compatible headless-shell.`;
+    `Docker image which ships a compatible headless-shell.` +
+    proxyDownloadStallHint();
  return new Error(message, { cause: cause instanceof Error ? cause : undefined });
 }

@@ -758,13 +857,17 @@ async function downloadBrowser(options?: EnsureBrowserOptions): Promise<BrowserR
  }

  const runInstall = () =>
-    install({
-      cacheDir: CACHE_DIR,
-      browser: Browser.CHROMEHEADLESSSHELL,
-      buildId: CHROME_VERSION,
-      platform,
-      downloadProgressCallback: options?.onProgress,
-    });
+    withDownloadStallGuard(
+      (onProgress) =>
+        install({
+          cacheDir: CACHE_DIR,
+          browser: Browser.CHROMEHEADLESSSHELL,
+          buildId: CHROME_VERSION,
+          platform,
+          downloadProgressCallback: onProgress,
+        }),
+      { timeoutMs: DOWNLOAD_STALL_TIMEOUT_MS, onProgress: options?.onProgress },
+    );

  let installed;
  try {