fix(deploy): reach storage over HTTP/1.1, and own storage CORS in the proxy
CI / verify (pull_request) Successful in 50s
CI / infrastructure-smoke (pull_request) Successful in 13s
CI / migration-smoke (pull_request) Successful in 1m47s
CI / image-build (pull_request) Successful in 18s
CI / deploy-smoke (same-origin) (pull_request) Successful in 47s
CI / deploy-smoke (split) (pull_request) Successful in 1m2s

Every reload of a cached files. image failed in the browser with
ERR_HTTP2_PROTOCOL_ERROR. A browser revalidates a cached signed URL
with If-None-Match. Caddy reached SeaweedFS over HTTP/2, and there
SeaweedFS answers with a 304 declaring content-type: application/xml
and content-length: 327 while sending no body. Caddy then logged
"aborting with incomplete response ... unexpected EOF" and reset the
stream. Over HTTP/1.1 SeaweedFS's 304 is clean. Measured layer by layer:
DevTools, curl through Caddy, and curl direct over HTTP/1.1 and HTTP/2.

The storage upstream is now HTTP/1.1 only (`versions 1.1`). Browsers
still reach the proxy over HTTP/2 and HTTP/3.

Delivery-layer CORS is the deployment's, so it no longer depends on
which S3 backend is behind it. SeaweedFS's default reflected any origin
and advertised PUT, POST and DELETE. Now the proxy:
- grants CORS only to the origins that read objects with fetch(): the
  Ekklesia origin, or, when split, the API origin (displays) and the
  Dashboard origin (a module's preview caches images through fetch());
- allows GET and HEAD only, with Vary: Origin;
- answers preflights itself, so they never reach storage;
- strips the object store's own CORS headers.

The deploy README documents the policy, its reasons, and what a
self-run proxy must do.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
This commit is contained in:
ekklesia
2026-10-06 01:19:17 +00:00
committed by The_River
co-authored by Claude Opus 5.5
parent dcbd7613ee
commit e1ac1da416
3 changed files with 135 additions and 25 deletions
+29 -1
View File
@@ -178,6 +178,29 @@ the topology and warns when the two hostnames are not siblings.
The Dashboard learns the API's origin at run time: the API writes its identity base URL into the page
it serves (`<meta name="ekklesia-api-origin">`). Nothing about the topology is built into the image.
### Storage delivery: CORS is the deployment's
Browsers fetch objects directly from the storage hostname through signed, expiring URLs (ADR 0031).
**Which pages may read those bytes with `fetch()` is decided by the delivery layer, the proxy, and
never by the object store.** That policy therefore stays the same with SeaweedFS, R2, or any other
S3 backend behind it, and an installation that changes backends changes nothing about CORS.
The reference proxy's storage block:
- grants CORS only to the origins whose pages read objects with `fetch()`: the Ekklesia origin in the
same-origin topology; the API origin (displays) and the Dashboard origin (a module's preview) in
the split topology. An `<img>` needs no CORS and is unaffected;
- allows `GET` and `HEAD` only, and no request header beyond the CORS-safelisted ones, always with
`Vary: Origin`;
- answers preflight `OPTIONS` requests itself, so they never reach storage;
- removes the object store's own CORS headers. SeaweedFS's default reflects any origin and
advertises `PUT`, `POST`, and `DELETE`.
The proxy reaches storage over **HTTP/1.1 only**. Over HTTP/2, SeaweedFS answers a conditional
request with a `304` that declares a `Content-Length` it never sends. The proxy then resets the
browser's stream, and every reload of a cached image fails with `ERR_HTTP2_PROTOCOL_ERROR`.
Browsers still reach the proxy itself over HTTP/2 and HTTP/3.
### Behind an existing reverse proxy
Run the base alone and attach `api` and `object-storage` to your proxy's network with an overlay of
@@ -213,6 +236,8 @@ Then make your proxy do what `Caddyfile.example` does:
- forward the storage hostname to `https://object-storage:8443`, verifying it against
`/etc/ekklesia/tls/ca.crt` with server name `object-storage`, and **pass the original `Host`
through unchanged**, or every signed URL fails (ADR 0031).
- reach `object-storage` over **HTTP/1.1** only, and own the storage hostname's CORS as described in
[Storage delivery](#storage-delivery-cors-is-the-deployments).
The proxy's address must be fixed and must never be handed to anything else. With Docker, give its
network an explicit subnet, limit `ip_range` to part of it, and fix the proxy's address outside that
@@ -288,7 +313,10 @@ deployment:
platform-only image and a disposable PostgreSQL and SeaweedFS that exist only for that run. It
follows the first boot above and checks the following through the proxy:
- readiness and the Dashboard;
- a signed object URL fetched through the storage hostname;
- a signed object URL fetched through the storage hostname, and the same image loaded twice over
HTTP/2, the second time as a conditional request that must be a clean `304`;
- storage CORS: exactly the allowed origins, `GET` and `HEAD` only, preflights answered by the
proxy, and none of the object store's own CORS headers;
- the client-address chain, both from an ordinary peer and from the tunnel connector's address;
- initial-owner sign-in and sign-out;
- that the API's database connection is TLS 1.3, and that it is refused by a foreign CA and by a