From e919d667325048a03fca4ebefda786a68b685d35 Mon Sep 17 00:00:00 2001
From: Muse Sweep <muse-sweep@developing-today.dev>
Date: Wed, 21 Oct 2026 15:04:50 -0500
Subject: [PATCH] fix(lab-ops-hub): give the Worker-fallback crawler a fetch
 timeout (#500)

crawlFallback's fetcher passed no AbortSignal, so a validated public
server that stalls, or slow-drips a body under the 2 MB cap, hung the
crawl indefinitely. The container crawler times out at 20s
(app.py _OPENER.open(timeout=20)) and the container dispatch already
uses AbortSignal.timeout(120s); the fallback now mirrors the
container's 20s per hop. The signal stays attached to the response,
so it also bounds the readResponseTextCapped body read.
---
 pkgs/lab-ops-hub/src/index.ts | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)

diff --git a/pkgs/lab-ops-hub/src/index.ts b/pkgs/lab-ops-hub/src/index.ts
index 7950b345d..ff03656b7 100644
--- a/pkgs/lab-ops-hub/src/index.ts
+++ b/pkgs/lab-ops-hub/src/index.ts
@@ -161,8 +161,18 @@ async function crawlFallback(url: string) {
   // Redirects are followed manually with every target re-validated (#460):
   // fetch's default "follow" mode would silently chase a 3xx from a public
   // URL to a private/loopback/link-local target the initial check refused.
+  // Every hop gets the container crawler's 20s timeout (app.py's
+  // _OPENER.open(timeout=20)): without a signal, a validated public server
+  // that accepts the connection and then stalls — or slow-drips a body
+  // under the byte cap — hangs the crawl until platform limits kill it
+  // (#500). The signal stays attached to the response, so it also bounds
+  // the readResponseTextCapped body read below.
   const res = await fetchFollowingValidatedRedirects(url, (u) =>
-    fetch(u, { redirect: "manual", headers: { "user-agent": "LabOpsHub/0.1 (+personal crawler)" } }),
+    fetch(u, {
+      redirect: "manual",
+      headers: { "user-agent": "LabOpsHub/0.1 (+personal crawler)" },
+      signal: AbortSignal.timeout(20_000),
+    }),
   );
   // Cap the body read at the container's 2,000,000 bytes (#481): res.text()
   // would buffer a huge page whole in the Worker's memory.