The "I can't approve a request, the toast just says unknown" bug was
four problems compounded:
1. Lidarr config let enabled=true save with default_quality_profile_id=0
and default_root_folder_path=''. Approve then sent invalid POST bodies
to Lidarr, which 5xx'd.
2. lidarr.client.post() discarded Lidarr's response body on error, so
we couldn't tell why Lidarr 5xx'd from server logs.
3. handleApproveRequest's error switch didn't map ErrServerError or
ErrLookupFailed — both fell through to a generic 500 server_error.
4. apiFetch only parsed {error: {code, message}} envelopes, but admin
endpoints write {error: 'code_string'}. Every admin error toast
rendered as 'unknown'.
Fixes:
- internal/lidarr/client.go: capture up to 512 bytes of Lidarr's response
body when it returns 4xx/5xx; include in the wrapped error so server
logs show what Lidarr actually said instead of just the status bucket.
- internal/lidarrrequests/service.go: new ErrDefaultsIncomplete fires
before the Lidarr call when QP=0 or root_folder=''. Stops the bad
POST entirely.
- internal/api/admin_requests.go: handleApproveRequest now maps
ErrDefaultsIncomplete -> 'lidarr_defaults_incomplete' (400),
ErrServerError -> 'lidarr_server_error' (502),
ErrLookupFailed -> 'lidarr_rejected' (502).
- internal/api/admin_lidarr.go: handlePutLidarrConfig now requires
QP + root folder to be set whenever enabled=true.
- web/src/lib/api/client.ts: apiFetch handles both error envelope shapes
so admin error codes propagate to toasts.
- web/src/routes/admin/integrations/+page.svelte: auto-default to the
first quality profile and first root folder Lidarr returns when the
operator hasn't picked one yet — saves a click for typical
one-profile/one-folder home setups.
- web/src/routes/admin/requests/+page.svelte: friendly toast copy for
the new error codes.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two debuggability gaps surfaced by the /api/admin route shadowing
investigation:
1. handleTestLidarrConnection swallowed client.Ping's error — only the
bucket code (lidarr_unreachable) reached the response, the underlying
*net.DNSError / *net.OpError / TLS error never reached the logs.
Operators couldn't tell a typo'd hostname from a wrong port from a
refused TLS handshake. Now logged at Warn (expected-when-misconfigured)
with the base_url for diagnostic context. The api_key is never logged.
2. The chi router had no access-log middleware — every 4xx/5xx was silent,
making it impossible to tell whether a request even reached the server.
chi's middleware.Logger writes to the standard log package not slog,
so a small slog wrapper handles it instead. /healthz is skipped (the
compose healthcheck hits it every 5s; would be ~17k log lines/day of
noise). Severity is keyed off response status: 5xx -> Error,
4xx -> Warn, else Info — so 4xx/5xx surface even when the operator's
logger level is set above Info.
Tests cover both the severity routing and the /healthz skip.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Five admin-only handlers under RequireAdmin middleware: GET/PUT config
(api_key masked, empty key on PUT preserves saved), POST test (always
200, maps Lidarr errors to stable codes), GET quality-profiles, GET
root-folders. 10 HTTP integration tests, all green.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
bvandeusen