From 07fb3377ad5df51ab59b93396ba00fa3dcc02ca1 Mon Sep 17 00:00:00 2001
From: mawkone <mark@getacquired.com>
Date: Fri, 12 Jun 2026 15:18:34 -0700
Subject: [PATCH] fix(preview): restore resilient zombie-killer logic to
 auto-restart suspended previews

---
 .../api/projects/[projectId]/anatomy/route.ts | 76 ++++++++++++++++++-
 1 file changed, 75 insertions(+), 1 deletion(-)

diff --git a/vibn-frontend/app/api/projects/[projectId]/anatomy/route.ts b/vibn-frontend/app/api/projects/[projectId]/anatomy/route.ts
index 57e32c4..917b867 100644
--- a/vibn-frontend/app/api/projects/[projectId]/anatomy/route.ts
+++ b/vibn-frontend/app/api/projects/[projectId]/anatomy/route.ts
@@ -793,7 +793,81 @@ async function loadPreviews(projectId: string): Promise<Preview[]> {
       [projectId],
     );
 
-    return sortDevPreviewsFrontendFirst(rows).map((r) => ({
+    // Filter out zombies: if a server is marked 'running' but the URL returns a 50x
+    // Gateway error or times out, the process died. We mark it stopped so the
+    // UI can trigger an auto-restart.
+    const activePreviews: typeof rows = [];
+
+    await Promise.all(
+      rows.map(async (r) => {
+        if (r.state !== "running") {
+          activePreviews.push(r);
+          return;
+        }
+
+        try {
+          const controller = new AbortController();
+          // We use a short timeout because we don't want to block the anatomy
+          // response. A slow response doesn't mean it's dead (Next.js might
+          // just be compiling) — we ONLY want to catch instant 502/503s from Traefik.
+          const timeout = setTimeout(() => controller.abort(), 2000);
+          const ping = await fetch(r.preview_url, {
+            method: "HEAD",
+            signal: controller.signal,
+          });
+          clearTimeout(timeout);
+
+          // 502/503/504 means Traefik is up but the container isn't answering.
+          // 404 means Traefik doesn't even know about the route.
+          if (
+            ping.status === 502 ||
+            ping.status === 503 ||
+            ping.status === 504 ||
+            ping.status === 404
+          ) {
+            // GRACE PERIOD: If the server was started less than 60 seconds ago,
+            // Traefik might return a 502/504 simply because the Node process hasn't
+            // finished booting and binding to the port yet. Do not kill it!
+            const ageMs = Date.now() - new Date(r.started_at).getTime();
+            if (ageMs < 60_000) {
+              activePreviews.push(r); // Give it the benefit of the doubt
+              return;
+            }
+
+            console.warn(
+              `[anatomy] Preview zombie detected for ${r.preview_url} (HTTP ${ping.status}, age ${Math.round(ageMs / 1000)}s). Marking stopped.`,
+            );
+            await query(
+              `UPDATE fs_dev_servers SET state = 'stopped' WHERE id = $1`,
+              [r.id],
+            ).catch(() => {});
+          } else {
+            activePreviews.push(r);
+          }
+        } catch (e: any) {
+          // If the fetch aborts due to our 2s timeout, the server is just slow
+          // (likely doing a cold Webpack compile). DO NOT mark it as a zombie!
+          // Only kill it if we get a hard DNS/network error that isn't a timeout.
+          if (
+            e.name === "AbortError" ||
+            e.type === "aborted" ||
+            e.message?.includes("timeout")
+          ) {
+            activePreviews.push(r); // Benefit of the doubt — it's thinking
+          } else {
+            console.warn(
+              `[anatomy] Preview zombie detected for ${r.preview_url} (${e.message}). Marking stopped.`,
+            );
+            await query(
+              `UPDATE fs_dev_servers SET state = 'stopped' WHERE id = $1`,
+              [r.id],
+            ).catch(() => {});
+          }
+        }
+      }),
+    );
+
+    return sortDevPreviewsFrontendFirst(activePreviews).map((r) => ({
       id: r.id,
       name: r.name,
       command: r.command ?? undefined,