fix(auth): handle OAuth model alias in retry logic and refine Qwen quota handling

internal/runtime/executor/qwen_executor.go

@@ -32,16 +32,6 @@ const (
 
 var qwenDefaultSystemMessage = []byte(`{"role":"system","content":[{"type":"text","text":"","cache_control":{"type":"ephemeral"}}]}`)
 
-// qwenBeijingLoc caches the Beijing timezone to avoid repeated LoadLocation syscalls.
-var qwenBeijingLoc = func() *time.Location {
-	loc, err := time.LoadLocation("Asia/Shanghai")
-	if err != nil || loc == nil {
-		log.Warnf("qwen: failed to load Asia/Shanghai timezone: %v, using fixed UTC+8", err)
-		return time.FixedZone("CST", 8*3600)
-	}
-	return loc
-}()
-
 // qwenQuotaCodes is a package-level set of error codes that indicate quota exhaustion.
 var qwenQuotaCodes = map[string]struct{}{
 	"insufficient_quota": {},
@@ -156,22 +146,13 @@ func wrapQwenError(ctx context.Context, httpCode int, body []byte) (errCode int,
 	// Qwen returns 403 for quota errors, 429 for rate limits
 	if (httpCode == http.StatusForbidden || httpCode == http.StatusTooManyRequests) && isQwenQuotaError(body) {
 		errCode = http.StatusTooManyRequests // Map to 429 to trigger quota logic
-		cooldown := timeUntilNextDay()
+		// Do not force an excessively long retry-after (e.g. until tomorrow), otherwise
-		retryAfter = &cooldown
+		// the global request-retry scheduler may skip retries due to max-retry-interval.
-		helps.LogWithRequestID(ctx).Warnf("qwen quota exceeded (http %d -> %d), cooling down until tomorrow (%v)", httpCode, errCode, cooldown)
+		helps.LogWithRequestID(ctx).Warnf("qwen quota exceeded (http %d -> %d)", httpCode, errCode)
 	}
 	return errCode, retryAfter
 }
 
-// timeUntilNextDay returns duration until midnight Beijing time (UTC+8).
-// Qwen's daily quota resets at 00:00 Beijing time.
-func timeUntilNextDay() time.Duration {
-	now := time.Now()
-	nowLocal := now.In(qwenBeijingLoc)
-	tomorrow := time.Date(nowLocal.Year(), nowLocal.Month(), nowLocal.Day()+1, 0, 0, 0, 0, qwenBeijingLoc)
-	return tomorrow.Sub(now)
-}
-
 // ensureQwenSystemMessage ensures the request has a single system message at the beginning.
 // It always injects the default system prompt and merges any user-provided system messages
 // into the injected system message content to satisfy Qwen's strict message ordering rules.

internal/runtime/executor/qwen_executor_test.go

@@ -1,6 +1,8 @@
 package executor
 
 import (
+	"context"
+	"net/http"
 	"testing"
 
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/thinking"
@@ -152,3 +154,25 @@ func TestEnsureQwenSystemMessage_MergesMultipleSystemMessages(t *testing.T) {
 		t.Fatalf("messages[0].content[2].text = %q, want %q", parts[2].Get("text").String(), "B")
 	}
 }
+
+func TestWrapQwenError_InsufficientQuotaDoesNotSetRetryAfter(t *testing.T) {
+	body := []byte(`{"error":{"code":"insufficient_quota","message":"You exceeded your current quota","type":"insufficient_quota"}}`)
+	code, retryAfter := wrapQwenError(context.Background(), http.StatusTooManyRequests, body)
+	if code != http.StatusTooManyRequests {
+		t.Fatalf("wrapQwenError status = %d, want %d", code, http.StatusTooManyRequests)
+	}
+	if retryAfter != nil {
+		t.Fatalf("wrapQwenError retryAfter = %v, want nil", *retryAfter)
+	}
+}
+
+func TestWrapQwenError_Maps403QuotaTo429WithoutRetryAfter(t *testing.T) {
+	body := []byte(`{"error":{"code":"insufficient_quota","message":"You exceeded your current quota","type":"insufficient_quota"}}`)
+	code, retryAfter := wrapQwenError(context.Background(), http.StatusForbidden, body)
+	if code != http.StatusTooManyRequests {
+		t.Fatalf("wrapQwenError status = %d, want %d", code, http.StatusTooManyRequests)
+	}
+	if retryAfter != nil {
+		t.Fatalf("wrapQwenError retryAfter = %v, want nil", *retryAfter)
+	}
+}

sdk/cliproxy/auth/conductor.go

@@ -1830,7 +1830,11 @@ func (m *Manager) closestCooldownWait(providers []string, model string, attempt
 		if attempt >= effectiveRetry {
 			continue
 		}
-		blocked, reason, next := isAuthBlockedForModel(auth, model, now)
+		checkModel := model
+		if strings.TrimSpace(model) != "" {
+			checkModel = m.selectionModelForAuth(auth, model)
+		}
+		blocked, reason, next := isAuthBlockedForModel(auth, checkModel, now)
 		if !blocked || next.IsZero() || reason == blockReasonDisabled {
 			continue
 		}

sdk/cliproxy/auth/conductor_overrides_test.go

@@ -8,6 +8,7 @@ import (
 	"time"
 
 	"github.com/google/uuid"
+	internalconfig "github.com/router-for-me/CLIProxyAPI/v6/internal/config"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/registry"
 	cliproxyexecutor "github.com/router-for-me/CLIProxyAPI/v6/sdk/cliproxy/executor"
 )
@@ -64,6 +65,49 @@ func TestManager_ShouldRetryAfterError_RespectsAuthRequestRetryOverride(t *testi
 	}
 }
 
+func TestManager_ShouldRetryAfterError_UsesOAuthModelAliasForCooldown(t *testing.T) {
+	m := NewManager(nil, nil, nil)
+	m.SetRetryConfig(3, 30*time.Second, 0)
+	m.SetOAuthModelAlias(map[string][]internalconfig.OAuthModelAlias{
+		"qwen": {
+			{Name: "qwen3.6-plus", Alias: "coder-model"},
+		},
+	})
+
+	routeModel := "coder-model"
+	upstreamModel := "qwen3.6-plus"
+	next := time.Now().Add(5 * time.Second)
+
+	auth := &Auth{
+		ID:       "auth-1",
+		Provider: "qwen",
+		ModelStates: map[string]*ModelState{
+			upstreamModel: {
+				Unavailable:    true,
+				Status:         StatusError,
+				NextRetryAfter: next,
+				Quota: QuotaState{
+					Exceeded:      true,
+					Reason:        "quota",
+					NextRecoverAt: next,
+				},
+			},
+		},
+	}
+	if _, errRegister := m.Register(context.Background(), auth); errRegister != nil {
+		t.Fatalf("register auth: %v", errRegister)
+	}
+
+	_, _, maxWait := m.retrySettings()
+	wait, shouldRetry := m.shouldRetryAfterError(&Error{HTTPStatus: 429, Message: "quota"}, 0, []string{"qwen"}, routeModel, maxWait)
+	if !shouldRetry {
+		t.Fatalf("expected shouldRetry=true, got false (wait=%v)", wait)
+	}
+	if wait <= 0 {
+		t.Fatalf("expected wait > 0, got %v", wait)
+	}
+}
+
 type credentialRetryLimitExecutor struct {
 	id string