claude/kochwas
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix(search): Filter zuverlässiger durch allowTruncate
All checks were successful
Build & Publish Docker Image / build-and-push (push) Successful in 1m16s
Details

Browse Source 
Vorher warf fetchText einen Fehler, sobald eine Seite >512 KB war —
bei modernen Rezeptseiten (eingebettete Bundles, base64-Bilder) läuft
das praktisch immer voll. Der Catch-Block hat dann hasRecipe auf NULL
gelassen, und der Treffer ging ungefiltert durch.

Neue FetchOptions.allowTruncate: true → wir bekommen die ersten 512 KB
(das reicht für <head> mit og:image und JSON-LD) statt eines Throws.
Timeout auf 8s erhöht, weil der Pi manchmal langsamer ist.

Migration 008 räumt alte NULL-has_recipe-Einträge aus dem Cache, damit
sie beim nächsten Search frisch klassifiziert werden statt weitere
30 Tage falsch gecached zu bleiben.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
hsiegeln
2026-04-17 22:33:55 +02:00
parent d3c9bc5619
commit 0992e51a5d
4 changed files with 61 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
src/lib/server/db/migrations/008_thumbnail_cache_drop_unknown.sql Normal file
View File

@@ -0,0 +1,7 @@
-- Bei Migration 007 war `allowTruncate` in fetchText noch nicht implementiert,
-- weshalb Seiten >512 KB einen Fehler warfen und hasRecipe als NULL (unbekannt)
-- gespeichert wurde. Diese Einträge würden weitere 30 Tage nicht revalidiert
-- und Treffer ohne schema.org/Recipe-Markup fälschlich durchlassen. Wir
-- räumen sie jetzt einmalig ab, damit sie beim nächsten Fetch korrekt
-- klassifiziert werden. Ein reines Cache-Flush, keine User-Daten betroffen.
DELETE FROM thumbnail_cache WHERE has_recipe IS NULL;

38
src/lib/server/http.ts
View File

@@ -3,9 +3,16 @@ export type FetchOptions = {
timeoutMs?: number;
userAgent?: string;
extraHeaders?: Record<string, string>;
/**
* When true, return the data read up to `maxBytes` instead of throwing.
* Useful when we only care about the page head (og:image, JSON-LD) — most
* recipe sites are >1 MB today because of inlined bundles, but the head is
* usually well under 512 KB.
*/
allowTruncate?: boolean;
};
const DEFAULTS: Required<Omit<FetchOptions, 'extraHeaders'>> = {
const DEFAULTS: Required<Omit<FetchOptions, 'extraHeaders' | 'allowTruncate'>> = {
maxBytes: 10 * 1024 * 1024,
timeoutMs: 10_000,
userAgent: 'Kochwas/0.1'
@@ -25,16 +32,23 @@ function assertSafeUrl(url: string): void {
async function readBody(
response: Response,
maxBytes: number
): Promise<{ data: Uint8Array; total: number }> {
maxBytes: number,
allowTruncate: boolean
): Promise<{ data: Uint8Array; total: number; truncated: boolean }> {
const reader = response.body?.getReader();
if (!reader) {
const buf = new Uint8Array(await response.arrayBuffer());
if (buf.byteLength > maxBytes) throw new Error(`Response exceeds ${maxBytes} bytes`);
return { data: buf, total: buf.byteLength };
if (buf.byteLength > maxBytes) {
if (allowTruncate) {
return { data: buf.slice(0, maxBytes), total: maxBytes, truncated: true };
}
throw new Error(`Response exceeds ${maxBytes} bytes`);
}
return { data: buf, total: buf.byteLength, truncated: false };
}
const chunks: Uint8Array[] = [];
let total = 0;
let truncated = false;
for (;;) {
const { value, done } = await reader.read();
if (done) break;
@@ -42,6 +56,14 @@ async function readBody(
total += value.byteLength;
if (total > maxBytes) {
await reader.cancel();
if (allowTruncate) {
// keep what we have up to the chunk boundary; good enough for HTML head
const keep = value.byteLength - (total - maxBytes);
if (keep > 0) chunks.push(value.slice(0, keep));
total = maxBytes;
truncated = true;
break;
}
throw new Error(`Response exceeds ${maxBytes} bytes`);
}
chunks.push(value);
@@ -53,7 +75,7 @@ async function readBody(
merged.set(c, offset);
offset += c.byteLength;
}
return { data: merged, total };
return { data: merged, total, truncated };
}
async function doFetch(url: string, opts: FetchOptions): Promise<Response> {
@@ -82,7 +104,7 @@ async function doFetch(url: string, opts: FetchOptions): Promise<Response> {
export async function fetchText(url: string, opts: FetchOptions = {}): Promise<string> {
const maxBytes = opts.maxBytes ?? DEFAULTS.maxBytes;
const res = await doFetch(url, opts);
const { data } = await readBody(res, maxBytes);
const { data } = await readBody(res, maxBytes, opts.allowTruncate ?? false);
return new TextDecoder('utf-8').decode(data);
}
@@ -92,6 +114,6 @@ export async function fetchBuffer(
): Promise<{ data: Uint8Array; contentType: string | null }> {
const maxBytes = opts.maxBytes ?? DEFAULTS.maxBytes;
const res = await doFetch(url, opts);
const { data } = await readBody(res, maxBytes);
const { data } = await readBody(res, maxBytes, opts.allowTruncate ?? false);
return { data, contentType: res.headers.get('content-type') };
}

11
src/lib/server/search/searxng.ts
View File

@@ -223,7 +223,16 @@ async function enrichPageMeta(
if (cached) return cached;
let meta: PageMeta = { image: null, hasRecipe: null };
try {
const html = await fetchText(url, { timeoutMs: 4_000, maxBytes: 512 * 1024 });
// allowTruncate: moderne Rezeptseiten sind oft >1 MB (eingebettete
// Bundles, base64-Bilder). Das og:image und JSON-LD steht praktisch
// immer im <head>, was locker in die ersten 512 KB passt. Früher
// warf fetchText auf Überschreitung und hasRecipe blieb NULL, sodass
// Nicht-Rezept-Seiten fälschlich durchgingen.
const html = await fetchText(url, {
timeoutMs: 8_000,
maxBytes: 512 * 1024,
allowTruncate: true
});
meta = {
image: extractPageImage(html, url),
hasRecipe: hasRecipeJsonLd(html) ? 1 : 0

14
tests/integration/http.test.ts
View File

@@ -45,6 +45,20 @@ describe('fetchText', () => {
});
await expect(fetchText(`${baseUrl}/`, { timeoutMs: 150 })).rejects.toThrow();
});
it('allowTruncate returns first maxBytes instead of throwing', async () => {
const head = '<html><head><title>hi</title></head>';
const filler = 'x'.repeat(2000);
server.on('request', (_req, res) => {
res.writeHead(200, { 'content-type': 'text/html' });
res.end(head + filler);
});
const text = await fetchText(`${baseUrl}/`, { maxBytes: 100, allowTruncate: true });
// First 100 bytes of body — should contain the <head> opening at least
expect(text.length).toBeLessThanOrEqual(2048); // chunk boundary may overshoot exact bytes slightly
expect(text).toContain('<html>');
expect(text).toContain('<head>');
});
});
describe('fetchBuffer', () => {