feat(agent): friendly 504 page for paused/timing-out upstreams

[runleveldev]

Summary

Adds a friendly 504 error page for frozen/paused or OOMing-but-alive containers.

When a container is frozen (cgroup.freeze) or OOMing but still alive, it keeps its TCP socket open but never responds. NGINX then hits proxy_read_timeout and returns a bare 504 Gateway Timeout — distinct from a stopped/crashed container, which refuses the connection and yields 502. Previously only 502.html existed, so a paused/slow upstream produced an unstyled 504.

Part of #431 (memory-exhaustion thrash). Closes #435.

Changes

• error-pages/504.html — new page ("Service Paused") explaining the app is paused/under heavy load and should return shortly. Styled to match the existing error pages (dark gradient, grid overlay, glass card, JS-injected hostname); uses an amber/gold gradient to read distinctly from 502 (blue/purple) and 503 (orange/red).
• agent/templates/nginx.conf.ejs — wires error_page 504 @504 following the established @502 pattern in every server block that already handles 502 (default HTTPS server, per-service blocks, bare-domain docs server), and adds location /504.html { } to the internal error-pages socket server.
• error-pages/502.html — copy revised to connection-refused semantics ("the upstream refused the connection… likely stopped or has crashed") so 502 vs 504 read sensibly for the memory-thrash case.

No packaging changes needed: agent/Makefile already globs error-pages/* into /opt/opensource-server/error-pages/.

Acceptance criteria

• A frozen or timing-out upstream renders the friendly 504 page, not a bare gateway error.
• 502 still renders for connection-refused/stopped containers.
• Pages are consistent with existing error-pages/ styling.

Verification

Rendered nginx.conf.ejs via EJS across service permutations (no services, plain HTTP, auth+server, auth-no-server, external domains) — all render successfully with the correct number of error_page 504/location @504 blocks and the socket-server location /504.html. nginx -t was not run here (nginx isn't installed in this environment); the agent already runs a config test with rollback on apply, and the 504 wiring mirrors the proven 502 blocks verbatim.

[Copilot]

Pull request overview

Adds a dedicated friendly 504 Gateway Timeout error page and wires it into the agent-generated NGINX config so paused/frozen or timing-out upstream containers render a styled, explanatory page (consistent with existing 502/503 handling and styling).

Changes:

• Introduces error-pages/504.html (“Service Paused”) styled to match existing error pages.
• Updates agent/templates/nginx.conf.ejs to route 504s through the internal error-pages socket server (mirroring the existing @502 pattern).
• Refines error-pages/502.html copy to better align with “connection refused / stopped” semantics.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

File	Description
error-pages/504.html	New friendly 504 page with consistent styling and explanatory troubleshooting text.
error-pages/502.html	Updates 502 message copy to distinguish it from 504 timeout scenarios.
agent/templates/nginx.conf.ejs	Adds error_page 504 @504 handling and serves /504.html via the internal error page server.