Files
ipx/docs/directory.md
rays cd7a9b8530 Apple Podcasts' top charts into the directory weekly, and its categories checked (#252)
With apple_country set, the directory fetches Apple's top 100 podcasts and the top 50 of each of
its 19 categories for that country (itunes.apple.com/{cc}/rss/toppodcasts), looks up each
show's RSS address (/lookup, 100 ids a call, three seconds apart, as Apple asks for no more than
about twenty a minute), and stores them as the snapshot apple-podcasts with no subscribers. The
refresher fetches them like any feed; the most-subscribed chart is unchanged. Apple's primary
genre goes in as the snapshot's category, used only while a feed names none.

The same weekly run reads Apple's genre tree (ws/genres?id=26) and warns when it differs from
ipx_feed's CATEGORIES; on 2026-10-15 they were the same, all 19 and their subcategories.

A syncs table keeps when it last ran: at start if a week has passed, again weekly, an hour after
a failure. keep_instances spares apple-podcasts, which no token holds, and no token may take
the name. A trace a sync, a span each answer.

CLAUDE.md: builds one cargo command at a time, the load looked at before and after, after a
build chained with a test compile took Tower past 200.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-16 13:21:48 +00:00

9.3 KiB

The directory

ipx-directory is a catalogue of public podcasts and blogs that ipx servers share (#204). An ipx server sends it the public feeds its people subscribe to, and how many of them do. The directory fetches each feed's title, artwork, category, description and latest twenty items every twelve hours, and any server holding a token can browse it. It never downloads a file, and it never stores one's address: every ipx server still fetches its own feeds.

This server's is at https://directory.ipxreader.app (until 2026-10-12 ipx-dir.sdf1.net, #214), the ipx-directory service of the Arcane project ipx, with its database ipx_directory on the databases project's Postgres.

In ipx

An admin sets it up in Admin, Directory: the directory's address, the token its operator gave this server, and whether to share. ipx asks the directory on its people's behalf, so the page never talks to it and never sees the token or a feed's address. With no directory set up, the Directory says so and feeds are added by their address.

ipx keeps the address and token with its other server settings, in its database (settings, directory_url and directory_token), not in config.toml. The token is written, never shown again: the tab says it is set and offers to clear it. To replace one, make a new token on the directory, paste it over the old, then delete the old token's entry from the directory's config.

With sharing on, ipx sends its snapshot when what it would send has changed (checked after every scan, and whenever the catalogue or these settings change), and every half hour regardless (#211): each feed checked at least once, with how many people here subscribe, and the category an admin gave a feed that names none. It never sends an OPML or a Patreon creator itself (their feeds go instead), or a private feed. A feed's settings have a Private switch for a paid feed iPX does not recognise by its address; anyone subscribed may turn it on, and only an admin off.

When this server's token is rwa, its admins administer the directory from iPX (#206). A feed's page in the Directory has Check now and Remove from the directory, which blocks it so no server's snapshot brings it back. Admin, Directory lists the Servers sharing, by token name, with how many feeds and subscribers each gave, and Blocked, by title, with Unblock. Dropping a server's snapshot takes its feeds and counts out until it next shares; to keep a server out, revoke its token. An admin's own subscriptions open as their feeds, not as Directory pages, so removing one of those is done by marking it private, or from another admin's account.

In a web browser

Anyone may browse it at its address, no token and no sign-in (#215): the most subscribed, the categories, every feed A to Z, a search by name, and a page for each feed with its description, its latest items and its address, with a Copy button, to subscribe to in any podcast app or feed reader. The pages only read: nothing in them changes the directory. They are rendered by the directory itself from what the API serves, with the publisher's HTML shown as text, an item's link only when it is http(s), and a Content-Security-Policy that allows the site's own script and stylesheet and the covers, which load from their publishers. /v1 is the API and still needs a token for everything.

Tokens

Everything but /healthz needs a token, sent as Authorization: Bearer <token>. Each one has a name and one of three levels of access:

browse send snapshots remove, block, refresh feeds; drop a server's snapshot
ro yes
rw yes yes
rwa yes yes yes

Tokens are in the config file and nowhere else, and no level of access can make or revoke one over the API. To make one:

docker exec ipx-directory ipx-directory token new <name> --access ro|rw|rwa

It prints the token, to give to whoever it is for, and the lines to add to the config file. The file holds the token's SHA-256, never the token. The directory reads the file again within a minute of a change, so adding a token and revoking one (deleting its entry) need no restart. A revoked token also takes the counts its server sent. A file that does not load (a misspelled access, two tokens with one name) leaves the tokens as they were, and the log says why.

The config file

/config/directory.toml in the container (DIRECTORY_CONFIG, or --config), on Tower /mnt/fast/appdata/ipx-directory/directory.toml:

bind = "0.0.0.0:8100"
# Prometheus metrics, on a port of their own and not the one the tunnel publishes. Empty, or
# left out, serves none.
metrics_bind = "0.0.0.0:9188"
# Apple Podcasts' top charts for this country, weekly (see below). Empty, or left out: none.
apple_country = "ca"
# The environment variable holding the database URL: postgres://... or sqlite://...
database_url_env = "DIRECTORY_DATABASE_URL"

[[tokens]]
name = "tower"
access = "rwa"
token_sha256 = "…"

DIRECTORY_LOG_FORMAT=json logs one JSON object a line, for Loki, as ipx's IPX_LOG_FORMAT does. Each request is a line with its method, path, status, duration_ms and the token's name.

With OTEL_EXPORTER_OTLP_ENDPOINT set (http://host:4318, OTLP over HTTP), it sends traces: a span a request, a child of the caller's when the request carries a traceparent, as ipx's do, and a trace a refresher round with a span a feed (#213). OTEL_RESOURCE_ATTRIBUTES tags them, as production does with deployment.environment.name=production. Its version is in crates/directory/Cargo.toml, logged at start and answered by ipx-directory --version; its changelog is crates/directory/CHANGELOG.md.

Apple Podcasts

With apple_country set, the directory takes Apple Podcasts' top charts for that country once a week (#252): the top 100 overall and the top 50 of each of Apple's categories, about a thousand shows. Apple's lookup API gives each one's RSS address, and they go in as a snapshot of their own, apple-podcasts, with no subscribers: fetched, sorted into categories and listed like any other feed, while the most subscribed stay what the servers sharing here subscribe to. A show Apple drops from its charts leaves the directory at the next sync, unless a server shares it. An admin can drop the snapshot as any server's (DELETE /v1/instances/apple-podcasts); the next sync brings it back. No token may be named apple-podcasts.

The same sync compares Apple's list of categories with the one ipx keeps (CATEGORIES in crates/feed) and logs a warning naming what differs, for whoever checks the log to update it. It runs at start when the last sync is a week old, and a failed one tries again an hour later. Each sync is a trace (apple_sync, a span for each of Apple's answers), and a line Apple Podcasts synced says how many shows came and went in.

Running one

services:
  ipx-directory:
    image: ipx-directory          # docker build -f Dockerfile.directory -t ipx-directory .
    restart: unless-stopped
    environment:
      DIRECTORY_DATABASE_URL: "postgres://ipx_directory:password@db-host:5432/ipx_directory"
    ports:
      - "8100:8100"
    volumes:
      - ./directory:/config:ro    # directory.toml

What it will not hold

A private feed is never meant to reach it. ipx leaves out a feed marked private, one with a username or password, and one whose address ipx_feed::url_looks_private catches: a paid-feed host, or a key in the query. The directory checks every address it is sent with the same function and refuses what it catches, so one server's miss does not publish someone's paid feed to every other. What both miss, an rwa token can remove: DELETE /v1/feeds/{id} removes the feed and blocks it, so no snapshot brings it back.

API

Route Access
GET /healthz none ok when it can read its database
GET /v1/me ro the token's name and access
GET /v1/feeds ro every feed A to Z: id, url, title, image, category, subcategory, podcast, subscribers
GET /v1/popular?limit=10 ro the most subscribed, at most 100
GET /v1/feeds/{id} ro one feed, with description, failing (why its last check failed, in words) and its latest twenty items (guid, title, link, published, duration, description), never their files
POST /v1/snapshot rw [{"url", "subscribers"}], at most 20,000: replaces everything this token sent before; answers what it accepted and how many it refused as private, blocked or invalid
DELETE /v1/feeds/{id} rwa remove and block
GET /v1/blocked, DELETE /v1/blocked/{id} rwa what is blocked; unblock
GET /v1/instances rwa each token holding a snapshot: its name, feeds and subscribers
DELETE /v1/instances/{name} rwa drop one token's snapshot, its counts with it
POST /v1/feeds/{id}/refresh rwa fetch the feed now, answering with its page

A feed's id is the first 64 bits of the SHA-256 of its address, normalised (scheme and host in lower case, no default port or fragment), so every server arrives at the same id. subscribers is the sum over every token's latest snapshot. A feed is listed once the directory has read it as a feed, so an OPML or an address that never worked is not, and goes when no snapshot has it any more. Descriptions are the feed's own HTML, unsanitized: whoever shows them sanitizes them, as ipx does.