feat(cag): CAG integration — memory adapter, warmup semaphore, cache-aware routing (#47)
CI / CI (push) Successful in 3m20s

## Summary

Three issues in one PR. Implements Cache-Augmented Generation (CAG) support in the api-gw.

---

### #44 — Memory ServiceAdapter: standing-context resource

- Code-registers `memory` adapter with `standing-context` resource (`POST /memory/standing-context`)
- Fixes dispatcher bug: method-level `Auth.Capability` override now respected (was always using verb-derived capability)
- Auth: `memory:read` required via method-level override; `memory:write` for ingest
- `MEMORY_UPSTREAM_URL` env configures upstream

### #45 — Continuous batching: warmup deprioritisation + per-backend semaphore

- In-memory bounded queue per backend: `VLLM_MAX_QUEUE_DEPTH` slots (default 20)
- Concurrency limiter: `VLLM_MAX_NUM_SEQS` slots (default 4)
- `X-Poimen-Warmup: true` → sleep `WARMUP_DELAY_MS` (default 500ms) before queue entry, strip header, add `X-Priority: low`
- Queue full → 503 immediately, no goroutine pile-up
- Metrics: `warmup_requests_delayed_total`, `queue_full_requests_total` at `/metrics`

### #46 — Cache-aware routing: rendezvous hash + prefix cache hit logging

- `addresses: [...]` YAML field for multi-backend models (backward-compat with `address`)
- `X-Poimen-Cache-Key` header → rendezvous-hash backend selection (same key = same GPU)
- No cache key → round-robin
- Failed backend excluded 30s; next-highest-score fallback
- `vllm-num-cached-tokens` response header captured → `vllm_prefix_cache_hit_tokens_total` counter + structured log

---

## Verification

```
go build ./...    ✅
go vet ./...      ✅
go test -race ./... ✅ (all packages, 0 races)
```

## Reviewer verdicts

- reviewer-sre: SIGN-OFF
- reviewer-quality: SIGN-OFF

## Non-blocking findings (🟡🔵 only)

- `getOrCreateTransport` not goroutine-safe by design; pre-population invariant documented
- `time.Sleep(warmupDelay)` holds goroutine; context-cancel alternative left as TODO
- OTel trace_id not propagated into CAG log events (existing gap)

Closes #44 Closes #45 Closes #46

---------

Co-authored-by: Admin Bot <[email protected]>
Reviewed-on: #47
This commit was merged in pull request #47.
This commit is contained in:
2026-09-27 11:16:40 +00:00
co-authored by Admin Bot
parent aaa9a5d1cb
commit ef5e736673
16 changed files with 1597 additions and 162 deletions
+67
View File
@@ -0,0 +1,67 @@
package serviceadapter
import "os"
// NewMemoryAdapter creates a code-registered ServiceAdapter for the memory service.
// Takes precedence over any config-file loaded memory adapter (main.go skips config
// adapters when a code-registered one with the same name exists).
// Upstream URL configured via MEMORY_UPSTREAM_URL env var.
func NewMemoryAdapter() *ServiceAdapter {
upstreamURL := os.Getenv("MEMORY_UPSTREAM_URL")
if upstreamURL == "" {
upstreamURL = "http://poimen-memory.poimen.svc.cluster.local:8080"
}
return &ServiceAdapter{
Namespace: "poimen",
ServiceName: "memory",
Spec: *GetMemorySpec(upstreamURL),
}
}
// GetMemorySpec returns the ServiceAdapter spec for the memory service.
func GetMemorySpec(upstreamURL string) *Spec {
return &Spec{
ServiceName: "memory",
Upstream: Upstream{
URL: upstreamURL,
TimeoutSeconds: 30,
},
Auth: Auth{
Required: true,
Capability: "memory:read",
},
Resources: []Resource{
{
Name: "ingest",
Methods: []Method{{
Verb: "POST",
UpstreamPath: "/memory/ingest",
Auth: &Auth{Required: true, Capability: "memory:write"},
}},
},
{
Name: "query",
Methods: []Method{{
Verb: "POST",
UpstreamPath: "/memory/query",
Auth: &Auth{Required: true, Capability: "memory:read"},
}},
},
{
Name: "skills",
Methods: []Method{{
Verb: "GET",
UpstreamPath: "/memory/skills",
}},
},
{
Name: "standing-context",
Methods: []Method{{
Verb: "POST",
UpstreamPath: "/memory/standing-context",
Auth: &Auth{Required: true, Capability: "memory:read"},
}},
},
},
}
}
+14 -5
View File
@@ -16,6 +16,7 @@ import (
"forgejo.riotpiao.com/rock/homelab-frontend/internal/auth"
"forgejo.riotpiao.com/rock/homelab-frontend/internal/identity"
"forgejo.riotpiao.com/rock/homelab-frontend/internal/logging"
"forgejo.riotpiao.com/rock/homelab-frontend/internal/problem"
)
@@ -74,8 +75,13 @@ func (d *Dispatcher) Dispatch(w http.ResponseWriter, r *http.Request) {
}
// JWT auth enforcement for adapters that require it
if adapter.Spec.Auth.Required && d.jwtValidator != nil {
if !d.authenticate(w, r, serviceName, method.Verb) {
if adapter.Spec.Auth.Required {
if d.jwtValidator == nil {
logging.Warnf("auth required but jwtValidator nil; request unauthenticated", map[string]string{
"service": serviceName,
"resource": resourceName,
})
} else if !d.authenticate(w, r, serviceName, method) {
return
}
}
@@ -98,7 +104,7 @@ func (d *Dispatcher) Dispatch(w http.ResponseWriter, r *http.Request) {
// authenticate validates the JWT and checks service-level capability.
// Returns false (and writes error response) if auth fails.
func (d *Dispatcher) authenticate(w http.ResponseWriter, r *http.Request, serviceName, verb string) bool {
func (d *Dispatcher) authenticate(w http.ResponseWriter, r *http.Request, serviceName string, method *Method) bool {
authHeader := r.Header.Get("Authorization")
if authHeader == "" {
d.writeError(w, problem.NewProblem(http.StatusUnauthorized,
@@ -115,8 +121,11 @@ func (d *Dispatcher) authenticate(w http.ResponseWriter, r *http.Request, servic
return false
}
// Check capability: <service>:read for GET/HEAD, <service>:write for mutating verbs
required := capabilityForVerb(serviceName, verb)
// Check capability: method-level Auth.Capability if set; else derive from verb.
required := capabilityForVerb(serviceName, method.Verb)
if method.Auth != nil && method.Auth.Capability != "" {
required = method.Auth.Capability
}
if !d.jwtValidator.CheckPermissions(claims, required, "*") {
d.writeError(w, problem.NewProblem(http.StatusForbidden,
"about:blank#insufficient-permissions", "Insufficient Permissions",
+202
View File
@@ -1,11 +1,19 @@
package serviceadapter
import (
"crypto/rand"
"crypto/rsa"
"encoding/base64"
"encoding/json"
"math/big"
"net/http"
"net/http/httptest"
"testing"
"time"
"github.com/golang-jwt/jwt/v5"
"forgejo.riotpiao.com/rock/homelab-frontend/internal/auth"
"forgejo.riotpiao.com/rock/homelab-frontend/internal/identity"
)
@@ -263,3 +271,197 @@ func TestDispatch_IdentityHeadersNotSet_WhenNoAuth(t *testing.T) {
t.Errorf("X-Auth-Verified should not be set without auth, got %q", gotVerified)
}
}
// TestDispatch_MethodCapabilityOverridesPOSTDefault verifies that a method-level
// Auth.Capability (e.g. "memory:read" on a POST standing-context endpoint) is
// respected instead of the verb-derived capability ("memory:write").
// Because auth.Validator is a concrete struct requiring real JWKS, we pass nil
// jwtValidator so auth is bypassed and assert only that routing succeeds (not 403).
func TestDispatch_MethodCapabilityOverridesPOSTDefault(t *testing.T) {
// Upstream captures the request so we can verify it was routed.
upstream := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
w.WriteHeader(http.StatusOK)
}))
defer upstream.Close()
// Memory-like adapter: POST standing-context with method-level Auth.Capability=memory:read.
adapter := ServiceAdapter{
Name: "memory",
ServiceName: "memory",
Spec: Spec{
ServiceName: "memory",
Upstream: Upstream{URL: upstream.URL, TimeoutSeconds: 5},
Auth: Auth{Required: false},
Resources: []Resource{
{
Name: "standing-context",
Methods: []Method{
{
Verb: "POST",
UpstreamPath: "/memory/standing-context",
// Override: read capability even though verb is POST.
Auth: &Auth{Capability: "memory:read"},
},
},
},
},
},
}
// nil jwtValidator: auth.Required=false + nil validator → auth skipped entirely.
d := NewDispatcher(newTestRegistry(adapter), nil)
w := httptest.NewRecorder()
r := httptest.NewRequest("POST", "/", nil)
r.Header.Set("X-Service", "memory")
r.Header.Set("X-Resource", "standing-context")
d.Dispatch(w, r)
// Must NOT be 403: the method-level memory:read override is not checked
// against verb-derived memory:write (auth is bypassed with nil validator).
// Must be 200 routed to real upstream.
if w.Code == http.StatusForbidden {
t.Errorf("got 403 Forbidden; memory:read override should not produce forbidden (auth bypassed)")
}
if w.Code != http.StatusOK {
t.Errorf("expected 200 from upstream, got %d", w.Code)
}
}
// TestAuthenticate_MethodCapabilityOverrideDirect calls d.Dispatch() with a real
// signed RS256 JWT so authenticate() executes end-to-end and the capability-override
// branch (router.go lines 124-130) is reached and covered.
//
// Sub-test 1: token has "memory:read", method.Auth.Capability="memory:read" → 200.
// Sub-test 2: token has "memory:read", no method override → POST derives
//
// "memory:write" → CheckPermissions fails → 403.
func TestAuthenticate_MethodCapabilityOverrideDirect(t *testing.T) {
// Generate an RSA key pair for signing test JWTs.
privateKey, err := rsa.GenerateKey(rand.Reader, 2048)
if err != nil {
t.Fatalf("rsa.GenerateKey: %v", err)
}
// Build minimal JWKS document from the RSA public key.
nBytes := privateKey.PublicKey.N.Bytes()
eBig := new(big.Int).SetInt64(int64(privateKey.PublicKey.E))
eBytes := eBig.Bytes()
jwksDoc, _ := json.Marshal(map[string]interface{}{
"keys": []map[string]interface{}{
{
"kty": "RSA",
"use": "sig",
"kid": "test-key",
"alg": "RS256",
"n": base64.RawURLEncoding.EncodeToString(nBytes),
"e": base64.RawURLEncoding.EncodeToString(eBytes),
},
},
})
// Serve JWKS from a local test server.
jwksSrv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
w.Header().Set("Content-Type", "application/json")
w.Write(jwksDoc)
}))
defer jwksSrv.Close()
// Upstream that returns 200 when authenticate() passes.
upstreamSrv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
w.WriteHeader(http.StatusOK)
}))
defer upstreamSrv.Close()
// signJWT returns a signed RS256 token with the given permissions slice.
signJWT := func(permissions []string) string {
tok := jwt.NewWithClaims(jwt.SigningMethodRS256, jwt.MapClaims{
"iss": "https://authentik.riotpiao.com/application/o/test/",
"aud": "test-audience",
"sub": "test-user",
"exp": time.Now().Add(time.Hour).Unix(),
"iat": time.Now().Unix(),
"permissions": permissions,
})
tok.Header["kid"] = "test-key"
signed, err := tok.SignedString(privateKey)
if err != nil {
t.Fatalf("sign JWT: %v", err)
}
return signed
}
validator := auth.NewValidator("", "", jwksSrv.URL)
// Sub-test 1: method.Auth.Capability="memory:read"; token has "memory:read".
// authenticate() must execute the override branch and pass → upstream returns 200.
t.Run("capability_override_match_reaches_upstream", func(t *testing.T) {
adapter := ServiceAdapter{
Name: "memory",
ServiceName: "memory",
Spec: Spec{
ServiceName: "memory",
Upstream: Upstream{URL: upstreamSrv.URL, TimeoutSeconds: 5},
Auth: Auth{Required: true},
Resources: []Resource{
{
Name: "standing-context",
Methods: []Method{
{
Verb: "POST",
UpstreamPath: "/memory/standing-context",
Auth: &Auth{Capability: "memory:read"}, // override
},
},
},
},
},
}
d := NewDispatcher(newTestRegistry(adapter), validator)
w := httptest.NewRecorder()
r := httptest.NewRequest("POST", "/", nil)
r.Header.Set("X-Service", "memory")
r.Header.Set("X-Resource", "standing-context")
r.Header.Set("Authorization", "Bearer "+signJWT([]string{"memory:read"}))
d.Dispatch(w, r)
if w.Code != http.StatusOK {
t.Errorf("expected 200 (override memory:read match), got %d body=%s", w.Code, w.Body.String())
}
})
// Sub-test 2: no method-level Auth override; POST → capabilityForVerb → "memory:write".
// Token has only "memory:read" → CheckPermissions fails → 403.
t.Run("verb_derived_capability_mismatch_returns_403", func(t *testing.T) {
adapter := ServiceAdapter{
Name: "memory",
ServiceName: "memory",
Spec: Spec{
ServiceName: "memory",
Upstream: Upstream{URL: upstreamSrv.URL, TimeoutSeconds: 5},
Auth: Auth{Required: true},
Resources: []Resource{
{
Name: "data",
Methods: []Method{
{
Verb: "POST",
UpstreamPath: "/memory/data",
// No method-level Auth: POST derives "memory:write".
},
},
},
},
},
}
d := NewDispatcher(newTestRegistry(adapter), validator)
w := httptest.NewRecorder()
r := httptest.NewRequest("POST", "/", nil)
r.Header.Set("X-Service", "memory")
r.Header.Set("X-Resource", "data")
r.Header.Set("Authorization", "Bearer "+signJWT([]string{"memory:read"}))
d.Dispatch(w, r)
if w.Code != http.StatusForbidden {
t.Errorf("expected 403 (write required, read only in token), got %d body=%s", w.Code, w.Body.String())
}
})
}