- HttpScope (AsyncLocal) tags each outbound request with a purpose and a trigger:
- purpose is set by the caller: actor, key, object, webfinger, context, nodeinfo;
- trigger is set by the job kind, by "verify" during inbox verification, or defaults
to "request".
- FederationHttp records every JSON, media and stream fetch: status, time, bytes, hops,
and an outcome of ok, refused or failed, with a reason: disallowed, remembered,
bad-redirect, too-many-redirects, content-type, too-large, bad-json, private-address,
timeout, network, or the status. A fetch a reader caused (trigger "request") is only
counted per server per day.
- Link previews record a 'preview' event: card, no-card or failed.
- The media proxy counts cache hits.
- TrafficMeter counts the client API per endpoint group, method and status class. It
counts our served documents (actor, outbox, collection, object, activity, licence,
webfinger, nodeinfo) by kind, status and whether signed, per day and never per server,
and never names a circle's collections.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ELjqpznMFMNrJoJUj6K5p2
- The media proxy streams ranged requests from the origin (passing the range on, never caching), downloads and caches
whole files otherwise, and serves cached files with range support. A PeerTube video is never fetched whole for one
viewer, and clients still never contact the remote host.
- PeerTube's fragmented MP4 files inside an HLS entry are read as variants, so HLS-only instances play too.
- A remote Video or Audio post becomes one playable Mastodon attachment: the best MP4 up to 720p that carries both
sound and picture, with its poster and duration; the card is kept only when nothing is playable.
- nginx: /media/proxy/ with proxy_buffering off and a 600 s read timeout (applied on the box, with a backup).
Checked live: a GoToSocial image through the proxy answers 206 with exactly the asked range when streamed, 200 when
cached, and 206 with the right Content-Range from the cache.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012CzABvBkbcFqoHdmi8b9WB
- A public post that links to a page gets a preview card: from the post's own data first (FEP-8967 Link preview, the
object's image, title and summary); otherwise the server reads the page once, 0-60 s after the post arrives, never
when someone reads it. OpenGraph and Twitter tags give title, description and image; one LinkPreview per address
is cached for 7 days and shared by the whole server, so a fetch never points at a persona.
- Lemmy link posts, which carry no title or description, get them filled in.
- The page fetch uses the guarded client (public addresses, three redirects, HTML only, first 512 KB).
- Federation:FetchLinkPreviews switches page fetching off.
- Local public posts get cards too.
Checked live: a link to a GoToSocial profile page becomes a card with its title, description and proxied image.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012CzABvBkbcFqoHdmi8b9WB
- When a sender's key cannot be fetched because its server timed out or answered 5xx, the inbox answers 503 with
Retry-After: 300 instead of 401, so Mastodon 4.7 retries rather than switching to RFC 9421 signatures we do not
verify yet. The fetcher's failure cache now remembers whether a failure was temporary.
- Follow, Like, Block, Accept, Reject and Undo ids are deliberately not dereferenceable: serving them would publish
who follows, likes and blocks whom. They are always sent with their object embedded.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012CzABvBkbcFqoHdmi8b9WB
- /api/v1/media and /api/v2/media (and GET/PUT /api/v1/media/:id):
images go through libvips (NetVips, its native build bundled):
autorotated, every kind of metadata dropped (EXIF, GPS, XMP, IPTC,
comments), capped at 4096 px, with a 640 px preview and a blurhash
(own encoder, the reference algorithm); animated GIFs are re-encoded;
video and audio are remuxed by ffmpeg with -map_metadata -1, never
re-encoded, and a video gets a still preview. Files get random names
under /var/lib/privapub/media, outside the web root deploys replace, and
are served at /media/files with nosniff and a sandbox CSP.
- media_ids on create and edit (four at most, the persona's own, each used
once); notes carry them as Document attachments with alt text, blurhash,
focalPoint and size; inbound attachments were already kept.
- avatar and header uploads in update_credentials, cropped to 400x400 and
1500x500, federated with Update{Person}.
- Remote media reaches clients only through /media/proxy/{hmac}/{url},
fetched by the guarded client (no SVG, 40 MB cap) and cached outside the
served root, trimmed to 5 GB; foreign avatars and headers use it too, so
a client never contacts another server.
- MediaJanitor deletes uploads left unattached for a day.
- nginx accepts 100 MB bodies on the upload endpoints only (applied on Max).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012CzABvBkbcFqoHdmi8b9WB
DomainBlock (domain, severity, reject-media, public and private comment)
covers the domain and its subdomains. Admins manage them under
/clientapi/admin/domainblocks/{list,insert,delete}; the set is kept in
memory, reloaded on every change and at most five minutes stale.
A suspended domain is refused by FederationHttp.IsAllowed, so nothing is
fetched from it and no job delivers to it, and the inbox drops its
activities with a 202 before fetching any key. Reject-media strips the
attachments of posts from that domain. Silence is recorded for the
timelines and notifications that arrive in P1.2.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012CzABvBkbcFqoHdmi8b9WB
Infrastructure/Http adds the client the roadmap's S3 and S4 ask for:
- the connect callback resolves the name itself and refuses loopback,
private, link-local, CGNAT, documentation, multicast, ULA, NAT64, 6to4,
Teredo and IPv4-mapped/compatible forms, then connects to the vetted
address, so DNS rebinding cannot swap it afterwards;
- redirects are followed by hand, at most three, each one re-checked;
- bodies are capped at 1 MB after decompression, only JSON media types are
read, every request has a 15 s budget, and a refused URL is not asked
again for five minutes.
Actor and WebFinger fetches and inbox deliveries all use it. Test networks
can switch on Federation:AllowPrivateNetworks/AllowPlainHttp; startup
refuses both in Production.
PrivaPub.Tests (xUnit v3) starts with the address table and the fetcher's
limits against an in-process peer; build.yml and deploy.yml run it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012CzABvBkbcFqoHdmi8b9WB
thepraandClaude Opus 5.5