OGProxy : Other Memory Saturation Root Cause & Fix

undefined

@cagatay sorry for late response. This looks too be one of the widgets, and I’m wondering if you have custom code that targets this.

Had this recently happened after an upgrade or other changes?

undefined

Hi, the active users are not showing correctly. Instead of the usernames, I’m seeing code. What should I do to fix this?

undefined

@DownPW Possibly, but the source files were also missing, and @cagatay has a custom theme developed by me.

undefined

OK
Likely deleted by mistake from the NodeBB ACP headers.

undefined

@DownPW the FA library was missing with no paths defined. I’ve set these in the headers and uploaded the source, including necessary shim files which resolves the issue.

undefined

i do not know Mark resolved.

undefined

so what is the problem of @cagatay mark @phenomlab ?

undefined

thank you Mark.

undefined

OGProxy : Other Memory Saturation Root Cause & Fix

OGProxy was periodically saturating server RAM and swap (up to ~4 GB of
arrayBuffers, swap fully consumed), causing multi-minute service degradation.
After tracing through several misleading leads, the root cause was identified:
OGProxy was downloading entire file-host link bodies into memory when trying
to generate previews.

On a file-sharing forum, links to file hosts (1fichier, etc.) are everywhere.
When OGProxy received a URL like https://1fichier.com/?xxxx, it attempted to
“preview” it, but that URL is a direct file download
(Content-Type: application/octet-stream, Content-Length: 20.6 GB). OGProxy
pulled the file into memory. Critically, neither open-graph-scraper’s
downloadLimit nor an AbortController stopped this, verified by
reproduction: arrayBuffers climbed ~120 MB/s past 4 GB while the abort
timeout was ignored.

Diagnostic path (for reference)

We instrumented the process with a /debug/mem endpoint exposing
process.memoryUsage() + cache size, plus a 30-second sampling trace. This let
us correlate memory spikes with nginx access logs. The trace showed
arrayBuffers jumping from 0 → 457 → 3669 MB in ~5 minutes, correlated via
nginx log to a single GET on a 1fichier link. The cache, EventEmitter
listeners, and image links were all ruled out as primary causes (cache stayed
at <30 entries during the spike; heapUsed stayed low; only arrayBuffers
leaked).

A representative slice of the trace at the moment of the spike:

11:24:39  arrayBuffers=0     rss=161
11:25:09  arrayBuffers=457   rss=427   <- jump in one 30s sample
11:25:39  arrayBuffers=884
11:26:09  arrayBuffers=1437
...
11:30:09  arrayBuffers=3669

No OGProxy fail log line appeared during the spike window, the offending
request neither failed nor completed; it was an in-progress, never-ending
download. The nginx access log for that minute pointed at the 1fichier GET.

Root cause

open-graph-scraper (ogs) performs its own internal fetch, and for these URLs:

The downloadLimit option does not reliably abort the body download on
streamed / chunked responses or on hosts that serve large
application/octet-stream payloads.
An AbortController passed via fetchOptions.signal does not propagate
to the underlying stream read in a way that stops the transfer in time.

Result: a single large file-host link could pull multiple GB into
arrayBuffers before anything intervened.

The fix: bounded streaming fetch

The structural problem is that ogs() controls the fetch and we don’t control
body consumption. The fix moves the fetch into our own code so we control every
byte read:

boundedFetch(url, maxBytes, timeoutMs) performs the HTTP fetch itself,
then:
- Re-checks the final host for SSRF after redirects.
- Rejects any non-text/html / application/xhtml Content-Type
  before reading the body (aborts immediately).
- Reads the body chunk-by-chunk via resp.body.getReader(), tracking total
  bytes, and hard-aborts at 5 MB regardless of what the server claims.
The retrieved HTML is then handed to ogs for parsing only: ogs({ html }).

This makes the protection structural rather than cooperative: no file host
can leak memory regardless of whether it honors HEAD, serves chunked, or
misreports headers.

Important ogs constraint

You must call ogs({ html }) alone. Passing { html, url } together
throws:

Must specify either `url` or `html`, not both

Because url is omitted, ogs cannot resolve relative og:image paths. This is
fine here: the ACP client already resolves relative image paths itself
(isFullPath() + host + imageUrl), so no client-side change was required.

Other hardening applied in the same pass

Cache: replaced memory-cache (which creates a per-entry setTimeout
that retains the cached object, a secondary leak) with a plain Map using
lazy expiry + a single sweep interval. Stored value is slimmed via
slimResult(): only error + result + HTML truncated at </head>
(preserves <title>, drops the multi-MB body and the undici response
object). Cap 300 entries, 30 min TTL, 10 min negative-cache TTL.
Negative cache: failed/rejected URLs are cached to prevent re-scrape
hammering from the client.
SSRF guards (three layers): static host/IP blocklist (private ranges,
loopback, link-local, CGNAT, IPv6 ULA/link-local), DNS resolution check, and
post-redirect re-validation of the final host. (Also backed at the OS level
by systemd IPAddressDeny on the unit.)
AbortController + clearTimeout in finally to stop the earlier
MaxListenersExceededWarning listener leak on timed-out requests.
nginx rate limit: limit_req_zone (10 r/s, burst 50, nodelay, returns
429) on the /ogproxy location. The API key is necessarily exposed
client-side (it ships in the ACP JS), so it provides no real protection on its
own; the rate limit is the actual abuse mitigation.
systemd guard rail: MemoryMax=512M / MemoryHigh=400M so OGProxy can
never take the whole box down again, this was the silent hero that kept the
server alive throughout diagnosis.

Validation

Test URL	Expected	Result
`https://1fichier.com/?xxxx` (20.6 GB)	reject, no body read	415, `arrayBuffers` stays 0
Direct image (`pbs.twimg.com/...jpg`)	reject on content-type	415
`https://github.com`	full preview	200, OG title/image/description, HTML truncated at `</head>`

Process idles at ~100 MB RSS; under load heapUsed oscillates and returns to
baseline (no step-up accumulation).

Reproduction of the bounded fetch against the 20.6 GB link, confirming zero
body is pulled:

arrayBuffers AVANT: 0 MB
  pendant: 0 MB
Resultat 1fichier: REJETE: non-HTML content-type: application/octet-stream
arrayBuffers APRES: 0 MB

Note on dependencies

Reproduced on open-graph-scraper 6.1.0 / undici 5.22.1 / Node 24. The
unreliable downloadLimit behavior may be version-specific; a newer undici
might handle aborts on large streams better. The bounded-fetch approach is
robust regardless of the underlying library version, so it is the recommended
long-term fix.

Appendix A: Full `server.js`

const express = require('express');
const ogs = require('open-graph-scraper');
const cors = require('cors');
const { URL } = require('url');
const dns = require('dns').promises;
const net = require('net');

require('events').EventEmitter.defaultMaxListeners = 50;

const app = express();
const port = 2000;

const apiKey = process.env.OGPROXY_API_KEY || '<API_KEY>';

const REQUEST_TIMEOUT = 12000;
const MAX_CONTENT_BYTES = 5 * 1024 * 1024;    // 5 MB hard cap on body
const CACHE_TTL_MS = 30 * 60 * 1000;
const FAIL_CACHE_TTL_MS = 10 * 60 * 1000;
const CACHE_MAX_ENTRIES = 300;
const MAX_REDIRECTS = 3;

// --- Map cache (lazy expiry, no per-entry timers) ---
const cacheStore = new Map();
function cacheGet(key) {
  const e = cacheStore.get(key);
  if (!e) return null;
  if (Date.now() > e.expires) { cacheStore.delete(key); return null; }
  return e.value;
}
function cacheSet(key, value, ttl) {
  if (cacheStore.size >= CACHE_MAX_ENTRIES) {
    cacheStore.delete(cacheStore.keys().next().value);
  }
  cacheStore.set(key, { value, expires: Date.now() + ttl });
}
setInterval(() => {
  const now = Date.now();
  for (const [k, e] of cacheStore) if (now > e.expires) cacheStore.delete(k);
}, 60 * 1000).unref();

function slimResult(results) {
  if (!results || typeof results !== 'object') return results;
  let slimHtml = '';
  if (typeof results.html === 'string') {
    const headEnd = results.html.search(/<\/head>/i);
    slimHtml = headEnd !== -1 ? results.html.slice(0, headEnd + 7) : results.html.slice(0, 8192);
  }
  return { error: results.error, result: results.result, html: slimHtml };
}

function isBlockedIp(ip) {
  if (!ip) return true;
  if (net.isIPv4(ip)) {
    const p = ip.split('.').map(Number);
    if (p[0] === 10) return true;
    if (p[0] === 127) return true;
    if (p[0] === 0) return true;
    if (p[0] === 169 && p[1] === 254) return true;
    if (p[0] === 192 && p[1] === 168) return true;
    if (p[0] === 172 && p[1] >= 16 && p[1] <= 31) return true;
    if (p[0] === 100 && p[1] >= 64 && p[1] <= 127) return true;
    return false;
  }
  if (net.isIPv6(ip)) {
    const v = ip.toLowerCase();
    if (v === '::1') return true;
    if (v.startsWith('fc') || v.startsWith('fd')) return true;
    if (v.startsWith('fe80')) return true;
    if (v.startsWith('::ffff:')) return isBlockedIp(v.split(':').pop());
    return false;
  }
  return true;
}

function isBlockedHost(hostname) {
  if (!hostname) return true;
  const h = hostname.toLowerCase();
  return (
    h === 'localhost' || h.endsWith('.localhost') ||
    h.endsWith('.internal') || h.endsWith('.local') ||
    (net.isIP(h) && isBlockedIp(h))
  );
}

async function resolvesToPublicIp(hostname) {
  try {
    const records = await dns.lookup(hostname, { all: true });
    if (!records || records.length === 0) return false;
    return records.every(r => !isBlockedIp(r.address));
  } catch (e) {
    return false;
  }
}

// Bounded streaming fetch: reads the body chunk by chunk and aborts hard at maxBytes.
// Rejects non-HTML content-types before reading any body. Structural protection
// against file hosts (1fichier, etc.) - independent of what the server claims.
async function boundedFetch(url, maxBytes, timeoutMs) {
  const controller = new AbortController();
  const timer = setTimeout(() => controller.abort(), timeoutMs);
  try {
    const resp = await fetch(url, {
      redirect: 'follow',
      signal: controller.signal,
      headers: {
        'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/126.0.0.0 Safari/537.36',
        'Accept': 'text/html,application/xhtml+xml,application/xml;q=0.9,image/avif,image/webp,*/*;q=0.8',
        'Accept-Language': 'fr-FR,fr;q=0.9,en;q=0.8',
      },
    });

    // Re-check final host after redirects (anti-SSRF)
    try {
      const finalHost = new URL(resp.url || url).hostname;
      if (isBlockedHost(finalHost) || !(await resolvesToPublicIp(finalHost))) {
        controller.abort();
        return { ok: false, reason: 'redirect to forbidden host' };
      }
    } catch (e) { /* ignore */ }

    const ctype = (resp.headers.get('content-type') || '').toLowerCase();
    if (ctype && !ctype.includes('text/html') && !ctype.includes('application/xhtml')) {
      controller.abort(); // not HTML: read nothing
      return { ok: false, reason: `non-HTML content-type: ${ctype.split(';')[0]}` };
    }

    if (!resp.body) {
      return { ok: false, reason: 'no response body' };
    }

    const reader = resp.body.getReader();
    const chunks = [];
    let total = 0;
    while (true) {
      const { done, value } = await reader.read();
      if (done) break;
      total += value.length;
      if (total > maxBytes) {
        controller.abort(); // hard cap reached: stop downloading
        return { ok: false, reason: `body exceeded ${maxBytes} bytes` };
      }
      chunks.push(value);
    }
    const html = Buffer.concat(chunks).toString('utf8');
    return { ok: true, html };
  } catch (e) {
    return { ok: false, reason: (e && e.name === 'AbortError') ? 'timeout/abort' : (e && e.message) || 'fetch error' };
  } finally {
    clearTimeout(timer);
  }
}

app.use(cors({ origin: 'https://YOUR_DOMAIN.EXT' }));

app.get('/debug/mem', (req, res) => {
  const m = process.memoryUsage();
  res.json({
    rss_mb: Math.round(m.rss / 1048576),
    heapUsed_mb: Math.round(m.heapUsed / 1048576),
    external_mb: Math.round(m.external / 1048576),
    arrayBuffers_mb: Math.round(m.arrayBuffers / 1048576),
    cache_entries: cacheStore.size,
  });
});

app.get('/ogproxy', async (req, res) => {
  let { url } = req.query;
  const requestApiKey = req.headers['x-api-key'];

  if (requestApiKey !== apiKey) return res.status(401).send('Unauthorized');
  if (!url || typeof url !== 'string') return res.status(400).send('Missing URL parameter');
  if (!url.startsWith('http')) {
    try { url = new URL(url, `${req.protocol}://${req.get('host')}`).href; }
    catch (e) { return res.status(400).send('Invalid URL'); }
  }

  let parsedUrl;
  try { parsedUrl = new URL(url); }
  catch (e) { console.warn(`OGProxy reject [${url}]: invalid URL`); return res.status(400).send('Invalid URL'); }

  if (!['http:', 'https:'].includes(parsedUrl.protocol)) {
    return res.status(400).send('Invalid protocol');
  }
  if (isBlockedHost(parsedUrl.hostname)) {
    console.warn(`OGProxy reject [${url}]: forbidden host (static guard)`);
    return res.status(403).send('Forbidden host');
  }

  const cached = cacheGet(url);
  if (cached) {
    if (cached.__ogproxyFail === true) return res.status(500).send('Error scraping Open Graph data (cached)');
    return res.json(cached);
  }

  if (!(await resolvesToPublicIp(parsedUrl.hostname))) {
    console.warn(`OGProxy reject [${url}]: resolves to private IP / DNS fail (SSRF)`);
    cacheSet(url, { __ogproxyFail: true }, FAIL_CACHE_TTL_MS);
    return res.status(403).send('Forbidden host');
  }

  if (cacheStore.size >= CACHE_MAX_ENTRIES) {
    cacheStore.delete(cacheStore.keys().next().value);
  }

  // Bounded fetch: download the body ourselves, capped at 5 MB, HTML-only.
  const fetched = await boundedFetch(url, MAX_CONTENT_BYTES, REQUEST_TIMEOUT);
  if (!fetched.ok) {
    console.error(`OGProxy reject [${url}]: ${fetched.reason}`);
    cacheSet(url, { __ogproxyFail: true }, FAIL_CACHE_TTL_MS);
    const code = (fetched.reason.startsWith('non-HTML') || fetched.reason.startsWith('body exceeded')) ? 415 : 500;
    return res.status(code).send('Unable to preview this URL');
  }

  try {
    // Parse the already-fetched HTML (no second fetch). Client resolves relative image paths itself.
    const results = await ogs({ html: fetched.html });
    const slim = slimResult(results);
    cacheSet(url, slim, CACHE_TTL_MS);
    return res.json(slim);
  } catch (error) {
    const reason = (error && error.result && error.result.error) || (error && error.message) || 'unknown';
    console.error(`OGProxy fail [${url}]: ${reason}`);
    cacheSet(url, { __ogproxyFail: true }, FAIL_CACHE_TTL_MS);
    return res.status(500).send('Error scraping Open Graph data');
  }
});

app.listen(port, () => {
  console.log(`OGProxy server listening on port ${port}`);
});

Note: /debug/mem is a temporary diagnostic endpoint. Remove it once the
deployment is confirmed stable in production.

Appendix B: nginx rate limit

Zone definition, placed in /etc/nginx/conf.d/ogproxy-ratelimit.conf (included
at the http level; survives vhost regeneration by the panel):

# Rate limit zone for OGProxy - 10 MB shared memory (~160k IPs tracked)
# 10 requests/second sustained per IP
limit_req_zone $binary_remote_addr zone=ogproxy_limit:10m rate=10r/s;

Application, inside the reverse-proxy location / of the OGProxy vhost:

location / {
    limit_req zone=ogproxy_limit burst=50 nodelay;
    limit_req_status 429;

    proxy_set_header Host $host;
    proxy_pass http://127.0.0.1:2000;
    proxy_redirect off;
    proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
    proxy_set_header X-Api-Key $http_x_api_key;
}

burst=50 absorbs the legitimate burst when a user opens a link-heavy topic
(the client fires many preview requests at once); sustained hammering beyond
that is rejected with 429.

Appendix C : systemd unit guard rails

Key directives on ogproxy.service:

[Service]
MemoryHigh=400M
MemoryMax=512M
Restart=always
RestartSec=3

# SSRF egress guard (OS-level backstop to the in-app checks)
IPAddressAllow=127.0.0.1 127.0.0.53 127.0.0.54
IPAddressDeny=10.0.0.0/8 172.16.0.0/12 192.168.0.0/16 169.254.0.0/16 100.64.0.0/10 fc00::/7 fe80::/10

127.0.0.1 must stay allowed because nginx reverse-proxies to OGProxy over
loopback; blocking all loopback breaks the nginx -> ogproxy hop (504s).

undefined

@cagatay Just looking at this now. Do you have a FA Pro 6 license, or are you trying to use the free version? It’s not immediately clear, and the directory structure on your server is a mess to be honest. Not sure what has happened here.

EDIT - I have it working. DM me.

Verified

Posts

OGProxy : Other Memory Saturation Root Cause & Fix

Diagnostic path (for reference)

Root cause

The fix: bounded streaming fetch

Important ogs constraint

Other hardening applied in the same pass

Validation

Note on dependencies

Appendix A: Full `server.js`

Appendix B: nginx rate limit

Appendix C : systemd unit guard rails

Member List

Verified

Posts

OGProxy : Other Memory Saturation Root Cause & Fix

Diagnostic path (for reference)

Root cause

The fix: bounded streaming fetch

Important ogs constraint

Other hardening applied in the same pass

Validation

Note on dependencies

Appendix A: Full server.js

Appendix B: nginx rate limit

Appendix C : systemd unit guard rails

Member List

Individual Categories

Appendix A: Full `server.js`