docs: spec limit semantics, pricing caveats, manifest hedge

I2 (controller ruling): docs/superpowers/specs/2026-08-19-traefik-llmgateway-design.md
described limits as "user overrides group" / "user limit if set, else
group limit, else unlimited" — stale relative to the actual, already-correct
implementation (buildLimitScopes/checkAndCount evaluate every configured
scope and return the first violation). Amended §2 and §7 to state both
scopes are enforced together and the tighter bound wins.

I4 (manifest hedge): .traefik.yml's groups.default was a bare "{}",
the shape empirically reproduced as a decode failure outside this
repo's own (necessarily simplified) yaegi-check harness. Changed to
"providers: []" — same zero semantics, no longer a bare empty object.
manifest_test.go's JSON mirror and drift-check token updated to match;
make yaegi-check still passes (its YAML-subset parser already handles
an inline "[]" under a nested map key).

M9 (pricing provenance): pricing.go's built-in table comment replaced
"verify before release" with a real source note (each provider's own
published list prices, recorded 2026-08, approximate). README's limits
section gained the matching caveat and points to the "pricing" config
override for billing-grade accuracy.

Doc nits: README's Kubernetes section now says the example manifest
applies cleanly once Traefik's CRDs (Middleware, IngressRoute) are
installed, rather than claiming kubectl apply -f works unconditionally.
README's Security notes section no longer claims "no auth-event logging
at all" (see the paired fix: commit) — it now describes what is logged
and that key material never appears.

This commit and its sibling fix: commit are the final wave closing out
the whole-branch review findings for docs/superpowers/sdd/2026-08-20-
traefik-llmgateway.

Gates: go test ./... -count=1 (-race) green, gofmt -l . empty, go vet
clean, make yaegi-check OK, golangci-lint 0 issues.
This commit is contained in:
2026-08-20 14:59:57 +01:00
parent 4b849fd6c7
commit c9ae14adf7
5 changed files with 32 additions and 19 deletions
+2 -1
View File
@@ -21,7 +21,8 @@ testData:
models:
- gpt-test
groups:
default: {}
default:
providers: []
users:
inline:
- name: tester
+16 -8
View File
@@ -358,7 +358,10 @@ kept — a bad edit to the file never breaks already-authenticated traffic.
built-in per-model table (`pricing.go`) or a configured `pricing`
override. An unpriced model costs 0 and logs a once-per-model warning
(capped at 128 distinct unpriced model names per process lifetime, then
one summary warning).
one summary warning). The built-in table is **approximate** — list
prices as each provider published them, recorded 2026-08, and not kept
in sync automatically. Set `pricing` overrides for billing-grade
accuracy.
- **Storage**: `redis` configured wires a hand-rolled, stdlib-only RESP2
client (`resp.go`) — no `go-redis`, which is not Yaegi-interpretable —
against Redis, Valkey, or Dragonfly. Semantics are deliberately
@@ -421,10 +424,14 @@ kept — a bad edit to the file never breaks already-authenticated traffic.
provider response content into a log line; today's only caller
(`writeProviderUpstreamError`) reads that body directly and puts it in
the client-facing response, never in a log.
- **There is no auth-event logging at all** — `auth.go` has zero logging
calls of any kind. A key that fails lookup gets a plain 401 JSON
envelope and nothing else happens; there is no log line, successful or
failed, for a request to appear in.
- **Every authentication attempt is logged**, through one shared helper
(`logAuthEvent`, `logger.go`) called at each of the six routes that
authenticate a request (`llmgateway.go`'s `ServeHTTP`, and
`routes_unified.go`'s `handleModels`). A failed attempt logs the request
method, path, and remote address, and returns the usual 401 JSON
envelope. A successful attempt logs the resolved user's name and the
route. Neither line ever includes the presented API key — only the
identity it resolved to, never the key material itself.
## Known limitations
@@ -470,9 +477,10 @@ kept — a bad edit to the file never breaks already-authenticated traffic.
## Kubernetes
See [`examples/kubernetes.yaml`](examples/kubernetes.yaml) for a complete,
YAML-validated example manifest that `kubectl apply -f` runs as-is: a
`Secret` holding the users file, a `Middleware` custom resource carrying
the plugin's own configuration (validated against the real `Config`
YAML-validated example manifest that `kubectl apply -f` applies cleanly
once Traefik's CRDs (`Middleware`, `IngressRoute`) are installed in the
cluster: a `Secret` holding the users file, a `Middleware` custom resource
carrying the plugin's own configuration (validated against the real `Config`
struct — every field in that example round-trips through `New()`
cleanly), a dummy backing `Deployment`/`Service` (Traefik requires a real
router backend even though this middleware never actually reaches it —
+2 -1
View File
@@ -34,6 +34,7 @@ func testDataConfig(t *testing.T) *Config {
"providers", "groups", "users", // top-level testData keys
"openai", "type: openai", "test-key", "gpt-test", // providers.openai
"default", // groups.default and users.inline[0].group
"providers: []", // groups.default.providers — zero-semantic hedge, not bare {}
"tester", // users.inline[0].name
"test-user-key", // users.inline[0].apiKey
}
@@ -50,7 +51,7 @@ func testDataConfig(t *testing.T) *Config {
cfg := CreateConfig()
// mirror of testData, kept in sync by this test living next to the manifest
blob := `{"providers":{"openai":{"type":"openai","apiKey":"test-key","models":["gpt-test"]}},
"groups":{"default":{}},
"groups":{"default":{"providers":[]}},
"users":{"inline":[{"name":"tester","group":"default","apiKey":"test-user-key"}]}}`
if err := json.Unmarshal([]byte(blob), cfg); err != nil {
t.Fatal(err)
+8 -7
View File
@@ -185,13 +185,14 @@ func (g *Gateway) handleTargetProxy(w http.ResponseWriter, r *http.Request, u *u
// validateTargetURLs checks every configured MCP-server and agent entry at
// construction: the map value must be non-nil, the name must pass
// validateConfigName (same character set and reserved-name rules as a
// provider name — an MCP server or agent is routed at "/mcp/{name}/*" or
// "/a2a/{name}/*", the same shadowing risk applies), and the target URL
// must parse (url.Parse) and use an http or https scheme. A nil value,
// invalid name, or malformed/non-HTTP target is rejected once, here, at
// construction — a constructor error, not a nil-pointer panic or a 502 the
// first time some caller happens to address it.
// validateConfigName (the same character-set, dot-only, and reserved-name
// rules as a provider name — an MCP server or agent is routed at
// "/mcp/{name}/*" or "/a2a/{name}/*", so the same shadowing risk and the
// same "." / ".." directory-traversal-segment risk both apply), and the
// target URL must parse (url.Parse) and use an http or https scheme. A nil
// value, invalid name, or malformed/non-HTTP target is rejected once, here,
// at construction — a constructor error, not a nil-pointer panic or a 502
// the first time some caller happens to address it.
func validateTargetURLs(cfg *Config) error {
for name, tc := range cfg.MCPServers {
if tc == nil {
+4 -2
View File
@@ -10,8 +10,10 @@ import (
const tokensPerMillion = 1_000_000
// builtinPricing is the built-in per-model price table, in USD per 1M
// tokens (InputPerM, OutputPerM). Source: public list prices, 2026-08;
// verify before release.
// tokens (InputPerM, OutputPerM). Source: each provider's own published
// list prices, as recorded 2026-08. Approximate: providers change prices
// without notice, and this table is not refreshed automatically — set the
// middleware config's "pricing" overrides for billing-grade accuracy.
var builtinPricing = map[string]ModelPricing{
"gpt-5": {InputPerM: 1.25, OutputPerM: 10},
"gpt-5-mini": {InputPerM: 0.25, OutputPerM: 2},