genie-api ignores `[services.api].url` — hardcodes listen on `127.0.0.1:3080`

[kiannidev]

Summary

genie-api always binds to 127.0.0.1:3080, while health probes, genie-ctl, and the dashboard Services row read [services.api].url from geniepod.toml. If an operator changes only the configured URL (e.g. for a second local instance on port 4080), probes and CLI target the new port but genie-api still listens on :3080, producing false DOWN signals and a broken dashboard.

This is the same class of config drift as #121 ([core].port vs [services.core].url), but for the dashboard service itself.

Duplicate check

Searched open/closed issues and PRs (May 2026). Related work fixed probe/proxy paths only, not the API server bind address:

Item	Notes
#90 / PR #102	genie-ctl / dashboard proxies to genie-core now use Config::core_http_addr().
#103–#127	genie-ctl / dashboard probe genie-api and HA from [services.*].url.
No existing issue	Nothing filed for genie-api listen address ignoring config.

Steps to reproduce

1. Set a non-default API URL while leaving the binary unchanged:

   [services.api]
   url = "http://127.0.0.1:4080/api/status"
   systemd_unit = "genie-api.service"

2. Start the stack:

   GENIEPOD_CONFIG=/path/to/geniepod.toml ./target/release/genie-api
   GENIEPOD_CONFIG=/path/to/geniepod.toml ./target/release/genie-health

3. Observe bind vs config:

   # Server still listens on 3080 (hardcoded in main.rs)
   ss -ltnp | grep -E '3080|4080'
   
   # Probes target 4080 per TOML
   curl -sf http://127.0.0.1:4080/api/status   # fails (connection refused)
   curl -sf http://127.0.0.1:3080/api/status   # succeeds

4. genie-ctl health / dashboard Services row for api report unhealthy while the process is up on the wrong port.

Expected behavior

• Single source of truth for where genie-api listens, aligned with [services.api].url (or explicit [api].port / [api].bind_host mirroring [core]).
• Health monitor, dashboard latency row, and genie-ctl agree with the actual listen socket.
• Default config (:3080) unchanged.

Actual behavior

crates/genie-api/src/main.rs hardcodes the bind address:

let bind_addr = "127.0.0.1:3080";

Meanwhile genie-core correctly uses config.core.port and config.core.bind_host at startup.

Suggested fix

1. Add ApiConfig (or reuse parsed host/port from [services.api].url) in genie-common, e.g.:

   pub fn api_http_addr(&self) -> String { /* parse services.api.url or [api] section */ }
   pub fn api_status_url(&self) -> String { format!("http://{}/api/status", self.api_http_addr()) }

2. genie-api: http::serve(&config.api_http_addr(), config) instead of a literal.

3. Optional: At Config::load(), tracing::warn! if services.api.url port disagrees with the derived listen port (same pattern proposed for [services.core].url drifts from [core].port — health monitor and dashboard probe wrong port #121).

4. Tests: genie-common unit test — URL http://127.0.0.1:4080/api/status → bind addr 127.0.0.1:4080.

Acceptance criteria

• With [services.api].url = "http://127.0.0.1:4080/api/status", genie-api listens on 127.0.0.1:4080.
• genie-health, dashboard /api/services api row, and genie-ctl health all agree when the service is healthy.
• Default TOML (:3080) has no regression.

Hardware / version

• Hardware: Non-Jetson dev host (logic-only)
• Version: main @ clone time (2026-05-22)

Related

• #121 — same drift pattern for genie-core health URL
• #99 / PR #102 — core proxy addressing (already fixed)