turbo: dedup downloads by content hash, not just filename

Same video gets uploaded to turbo.cr under different IDs and resolves to
different filenames, so the existsSync(filename) check can't catch
content-duplicates. Switched to the same signature the gallery scanner
uses — md5 of the first 64KB plus exact byte-size match — and apply it
during the download stream so we abort once a same-content existing file
is detected. Avoids re-downloading content the user already has (or has
deliberately deleted via the duplicate scanner).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

server/scrapers/turbo.js

@@ -1,8 +1,47 @@
-import { writeFileSync, existsSync } from 'fs';
+import { writeFileSync, existsSync, readdirSync, statSync, createReadStream, unlinkSync } from 'fs';
 import { join, basename } from 'path';
+import { createHash } from 'crypto';
 import { fsCreateSession, fsDestroySession, fsGet } from '../flaresolverr.js';
 import { upsertMediaFile } from '../db.js';
 
+// Match the duplicate scanner in gallery.js — md5 of first 64KB + exact size.
+const HASH_BYTES = 65536;
+
+function hashFirst64kSync(filePath) {
+  return new Promise((resolve, reject) => {
+    const hash = createHash('md5');
+    const s = createReadStream(filePath, { start: 0, end: HASH_BYTES - 1 });
+    s.on('data', (c) => hash.update(c));
+    s.on('end', () => resolve(hash.digest('hex')));
+    s.on('error', reject);
+  });
+}
+
+// Build size -> [{filename, path, hash:null}] index for the folder. Hashes are
+// computed lazily only when a size collision is found.
+function buildSizeIndex(folderPath) {
+  const idx = new Map();
+  let entries;
+  try { entries = readdirSync(folderPath); } catch { return idx; }
+  for (const name of entries) {
+    if (name.startsWith('.')) continue;
+    const p = join(folderPath, name);
+    try {
+      const st = statSync(p);
+      if (!st.isFile()) continue;
+      if (!idx.has(st.size)) idx.set(st.size, []);
+      idx.get(st.size).push({ filename: name, path: p, hash: null });
+    } catch {}
+  }
+  return idx;
+}
+
+async function ensureCandidateHash(c) {
+  if (c.hash != null) return c.hash;
+  try { c.hash = await hashFirst64kSync(c.path); } catch { c.hash = ''; }
+  return c.hash;
+}
+
 const TURBO_HOST_RE = /^https?:\/\/(?:www\.)?turbo\.\w+\//i;
 const TURBO_BASE = 'https://turbo.cr';
 const DEFAULT_UA = 'Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/120.0.0.0 Safari/537.36';
@@ -36,16 +75,67 @@ function turboFilename(mp4Url, fallbackId) {
   return (fallbackId || 'turbo') + '.mp4';
 }
 
-async function downloadVideo(url, dest, ua) {
+// Stream-download a video. After the first 64KB has been received, compute its
+// md5 and check it against existing files of identical Content-Length in
+// sizeIndex. If a match is found, the stream is aborted and no file is written
+// — that's the same dedup signature the gallery scanner uses, so this prevents
+// re-downloading content the user has already kept (or deleted by hand).
+async function downloadVideo(url, dest, ua, sizeIndex) {
   try {
     const r = await fetch(url, {
       headers: { 'User-Agent': ua, 'Referer': TURBO_BASE + '/' },
-      signal: AbortSignal.timeout(600000), // 10 min for big videos
+      signal: AbortSignal.timeout(600000),
     });
     if (!r.ok) return { ok: false, status: r.status };
-    const buf = Buffer.from(await r.arrayBuffer());
+
+    const totalSize = parseInt(r.headers.get('content-length') || '0', 10);
+    const reader = r.body.getReader();
+    const chunks = [];
+    let totalRead = 0;
+    let firstHash = null;
+
+    while (true) {
+      const { done, value } = await reader.read();
+      if (done) break;
+      chunks.push(value);
+      totalRead += value.length;
+
+      if (firstHash === null && totalRead >= HASH_BYTES) {
+        const head = Buffer.concat(chunks).subarray(0, HASH_BYTES);
+        firstHash = createHash('md5').update(head).digest('hex');
+
+        if (totalSize > 0 && sizeIndex.has(totalSize)) {
+          for (const cand of sizeIndex.get(totalSize)) {
+            const ch = await ensureCandidateHash(cand);
+            if (ch && ch === firstHash) {
+              try { await reader.cancel(); } catch {}
+              return { ok: true, dupe: true, dupeOf: cand.filename, size: totalSize };
+            }
+          }
+        }
+      }
+    }
+
+    const buf = Buffer.concat(chunks);
     if (buf.length < 10000) return { ok: false, reason: 'too small ' + buf.length };
+
+    if (firstHash === null && buf.length > 0) {
+      firstHash = createHash('md5').update(buf.subarray(0, Math.min(buf.length, HASH_BYTES))).digest('hex');
+    }
+
+    // One more dedup pass with the actual final size (some servers omit Content-Length)
+    if (firstHash && sizeIndex.has(buf.length)) {
+      for (const cand of sizeIndex.get(buf.length)) {
+        const ch = await ensureCandidateHash(cand);
+        if (ch && ch === firstHash) {
+          return { ok: true, dupe: true, dupeOf: cand.filename, size: buf.length };
+        }
+      }
+    }
+
     writeFileSync(dest, buf);
+    if (!sizeIndex.has(buf.length)) sizeIndex.set(buf.length, []);
+    sizeIndex.get(buf.length).push({ filename: basename(dest), path: dest, hash: firstHash });
     return { ok: true, size: buf.length };
   } catch (e) {
     return { ok: false, error: e.message };
@@ -104,6 +194,10 @@ export async function downloadTurbo(url, outputDir, logFn, userAgent, fsSession)
       return 0;
     }
 
+    // Build size index once per call so repeated downloads in the same job
+    // share the cache (and the lazy hash cache).
+    const sizeIndex = buildSizeIndex(outputDir);
+
     let count = 0;
     for (const id of embedIds) {
       const embedUrl = `${TURBO_BASE}/embed/${id}`;
@@ -119,8 +213,10 @@ export async function downloadTurbo(url, outputDir, logFn, userAgent, fsSession)
           logFn(`turbo: already have ${filename}`);
           continue;
         }
-        const dl = await downloadVideo(mp4, dest, ua);
+        const dl = await downloadVideo(mp4, dest, ua, sizeIndex);
-        if (dl.ok) {
+        if (dl.ok && dl.dupe) {
+          logFn(`turbo: skip ${filename} — same content as ${dl.dupeOf}`);
+        } else if (dl.ok) {
           try { upsertMediaFile(folderName, filename, 'video', dl.size, Date.now(), null); } catch {}
           logFn(`Downloaded: ${filename} (${(dl.size / (1024 * 1024)).toFixed(1)} MB) [video]`);
           count++;