fix(rate-limits): treat empty policies as unlimited

apps/api/internal/runner/limits.go

@@ -83,11 +83,18 @@ func (s *Service) rateLimitReservations(ctx context.Context, user *auth.User, ca
 
 func effectiveRateLimitPolicy(candidate store.RuntimeModelCandidate) map[string]any {
 	policy := candidate.PlatformRateLimitPolicy
-	if hasRules(candidate.RuntimeRateLimitPolicy) {
+	if strings.TrimSpace(candidate.RuntimePolicySetID) != "" {
+		policy = candidate.RuntimeRateLimitPolicy
+	} else if hasRules(candidate.RuntimeRateLimitPolicy) {
 		policy = mergeMap(policy, candidate.RuntimeRateLimitPolicy)
 	}
-	if nested, ok := candidate.RuntimePolicyOverride["rateLimitPolicy"].(map[string]any); ok && len(nested) > 0 {
-		policy = mergeMap(policy, nested)
+	if _, hasOverride := candidate.RuntimePolicyOverride["rateLimitPolicy"]; hasOverride {
+		nested, _ := candidate.RuntimePolicyOverride["rateLimitPolicy"].(map[string]any)
+		if len(nested) == 0 {
+			policy = nil
+		} else {
+			policy = mergeMap(policy, nested)
+		}
 	}
 	if hasRules(candidate.ModelRateLimitPolicy) {
 		policy = mergeMap(policy, candidate.ModelRateLimitPolicy)
@@ -123,6 +130,9 @@ func reservationsFromPolicy(scopeType string, scopeKey string, scopeName string,
 		rule, _ := rawRule.(map[string]any)
 		metric := strings.TrimSpace(stringFromMap(rule, "metric"))
 		limit := floatFromAny(rule["limit"])
+		if metric == "" || limit <= 0 {
+			continue
+		}
 		amount := 1.0
 		if strings.HasPrefix(metric, "tpm") {
 			amount = float64(estimatedTokens)

apps/api/internal/runner/limits_test.go

@@ -4,6 +4,7 @@ import (
 	"testing"
 
 	"github.com/easyai/easyai-ai-gateway/apps/api/internal/clients"
+	"github.com/easyai/easyai-ai-gateway/apps/api/internal/store"
 )
 
 func TestTokenUsageAmountsUsesActualUsageForTPM(t *testing.T) {
@@ -27,3 +28,39 @@ func TestTokenUsageAmountsFallsBackToInputOutputTotal(t *testing.T) {
 		t.Fatalf("expected total token fallback 8, got %v", got["tpm_total"])
 	}
 }
+
+func TestEffectiveRateLimitPolicyTreatsEmptyRuntimePolicyAsUnlimited(t *testing.T) {
+	policy := effectiveRateLimitPolicy(store.RuntimeModelCandidate{
+		PlatformRateLimitPolicy: map[string]any{"rules": []any{
+			map[string]any{"metric": "rpm", "limit": 500},
+		}},
+		RuntimePolicySetID:     "runtime-policy-1",
+		RuntimeRateLimitPolicy: map[string]any{"rules": []any{}},
+	})
+
+	if hasRules(policy) {
+		t.Fatalf("expected empty runtime policy to clear inherited limits, got %+v", policy)
+	}
+}
+
+func TestReservationsFromPolicySkipsNonPositiveLimits(t *testing.T) {
+	reservations := reservationsFromPolicy(
+		"platform_model",
+		"model-1",
+		"Model 1",
+		nil,
+		map[string]any{"rules": []any{
+			map[string]any{"metric": "rpm", "limit": -1},
+			map[string]any{"metric": "tpm_total", "limit": 0},
+			map[string]any{"metric": "concurrent", "limit": 2},
+		}},
+		map[string]any{"prompt": "hello"},
+	)
+
+	if len(reservations) != 1 {
+		t.Fatalf("expected only the positive concurrent rule to reserve, got %+v", reservations)
+	}
+	if reservations[0].Metric != "concurrent" || reservations[0].Limit != 2 {
+		t.Fatalf("expected concurrent reservation with limit 2, got %+v", reservations[0])
+	}
+}