Two bugs under one roof.
The visible one: a publisher uploading anything over 10 MB on the live
site got "Network error — Unexpected token '<'" in the browser console.
Cause: the Caddy block I wrote for prometheus.x had
`request_body { max_size 2GB }` nested inside the handle block, where it
is a syntax Caddy parses but does not apply, so Caddy's own 10 MB
default was still the gate. Any oversize POST got Caddy's HTML 413 page,
which the browser's JSON parser choked on at the leading `<`.
Fix: moved `request_body { max_size … }` to the top of the prometheus.x
site block (outside `handle`), where Caddy actually honours it. The live
Caddyfile on p2pexchange is already patched + reloaded; a 50 MB upload
verified through the end-to-end path (Caddy → service → clamd → signed
scan report).
The scope one: 2 GiB was more than Prometheus should offer to publishers
at launch. Mobile apps are usually ~100 MB and even most desktop
installers come in well under 200 MiB. Dropping to 200 MiB:
- Fits every Android APK and most desktop installers.
- Shrinks the operator's Sia storage bill (fewer accidental Steam-size
uploads).
- Keeps the "publishers of larger binaries self-host and submit a
manifest with the external URL + SHA" escape hatch — the scanner
still fetches the URL and checks the hash, so a 5 GB game isn't
locked out, Prometheus just doesn't eat its storage.
Changes:
service/uploads.js MAX_FILE_BYTES default: 2 GiB → 200 MiB
service/pages.js /submit step-1 subtitle: "any size up to
2 GiB" → "up to 200 MiB"
deploy/prometheus-x.conf (unused angie fallback) 2g → 200m
deploy/apps-silentmode-st.conf (unused angie fallback) 2g → 200m
PROMETHEUS_MAX_UPLOAD_BYTES env still overrides per-instance.
|
||
|---|---|---|
| deploy | ||
| public | ||
| scripts | ||
| service | ||
| .gitignore | ||
| package-lock.json | ||
| package.json | ||
| README.md | ||
Prometheus
The service behind prometheus.x — a standalone website that accepts app
listings, scans the binaries, generates APS1 records for publishers to sign,
and serves the browse pages.
The architecture decisions and the reasoning behind them live at
_coordination/sessions/prometheus-x-architecture.md.
Read that before touching the code; it answers the questions this directory
otherwise invites ("why isn't this on Sia?", "why isn't the admin on another
subdomain?", "why Node?").
service/ Node HTTP server, templates, scanner, storage, payments
public/ Static assets served under /static/
content/ Prose content in a form the templates render into pages
Run it locally
cd Prometheus
npm install # none yet; placeholder for when deps show up
node service/server.js
Opens on http://127.0.0.1:3700 by default. Override with PORT=xxxx.
Layout
The server is one node:http listener behind a trivial map-based router.
Pages are plain functions that return {title, description, body} and
layout() wraps them in the common document shell. Static files under
public/ are served directly.
There's no framework and no templating engine on purpose — a site this small does not earn a dependency surface, and the response-rendering path is short enough to read at a glance.