From 0debe0a80c9ae53d848bc04c0b46e76473fbe11b Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E7=A5=9E=E4=B9=90?= <6682635@qq.com>
Date: Sat, 7 Mar 2026 20:02:58 +0800
Subject: [PATCH 1/6] =?UTF-8?q?fix:=20=E4=BF=AE=E5=A4=8D=20OpenAI=20WS=20?=
 =?UTF-8?q?=E7=94=A8=E9=87=8F=E7=AA=97=E5=8F=A3=E5=88=B7=E6=96=B0=E4=B8=8E?=
 =?UTF-8?q?=E9=99=90=E9=A2=9D=E7=BA=A0=E5=81=8F?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 .../internal/service/account_test_service.go  |  10 ++
 .../account_test_service_openai_test.go       |  66 +++++++
 .../internal/service/account_usage_service.go |  64 ++++++-
 .../service/account_usage_service_test.go     |  68 +++++++
 backend/internal/service/ratelimit_service.go |  18 ++
 .../service/ratelimit_service_openai_test.go  |  46 +++++
 .../components/account/AccountUsageCell.vue   |  68 ++++++-
 .../components/account/UsageProgressBar.vue   |  60 +------
 .../__tests__/AccountUsageCell.spec.ts        | 167 ++++++++++++++++++
 .../__tests__/accountUsageRefresh.spec.ts     |  39 ++++
 frontend/src/utils/accountUsageRefresh.ts     |  28 +++
 frontend/src/views/admin/AccountsView.vue     |   4 +-
 12 files changed, 568 insertions(+), 70 deletions(-)
 create mode 100644 backend/internal/service/account_test_service_openai_test.go
 create mode 100644 backend/internal/service/account_usage_service_test.go
 create mode 100644 frontend/src/utils/__tests__/accountUsageRefresh.spec.ts
 create mode 100644 frontend/src/utils/accountUsageRefresh.ts

diff --git a/backend/internal/service/account_test_service.go b/backend/internal/service/account_test_service.go
index 9557e175..5b22c645 100644
--- a/backend/internal/service/account_test_service.go
+++ b/backend/internal/service/account_test_service.go
@@ -408,6 +408,16 @@ func (s *AccountTestService) testOpenAIAccountConnection(c *gin.Context, account
 
 	if resp.StatusCode != http.StatusOK {
 		body, _ := io.ReadAll(resp.Body)
+		if isOAuth && s.accountRepo != nil {
+			if updates, err := extractOpenAICodexProbeUpdates(resp); err == nil && len(updates) > 0 {
+				_ = s.accountRepo.UpdateExtra(ctx, account.ID, updates)
+				mergeAccountExtra(account, updates)
+			}
+			if resetAt := (&RateLimitService{}).calculateOpenAI429ResetTime(resp.Header); resetAt != nil {
+				_ = s.accountRepo.SetRateLimited(ctx, account.ID, *resetAt)
+				account.RateLimitResetAt = resetAt
+			}
+		}
 		return s.sendErrorAndEnd(c, fmt.Sprintf("API returned %d: %s", resp.StatusCode, string(body)))
 	}
 
diff --git a/backend/internal/service/account_test_service_openai_test.go b/backend/internal/service/account_test_service_openai_test.go
new file mode 100644
index 00000000..61a755a7
--- /dev/null
+++ b/backend/internal/service/account_test_service_openai_test.go
@@ -0,0 +1,66 @@
+//go:build unit
+
+package service
+
+import (
+	"context"
+	"net/http"
+	"testing"
+	"time"
+
+	"github.com/gin-gonic/gin"
+	"github.com/stretchr/testify/require"
+)
+
+type openAIAccountTestRepo struct {
+	mockAccountRepoForGemini
+	updatedExtra  map[string]any
+	rateLimitedID int64
+	rateLimitedAt *time.Time
+}
+
+func (r *openAIAccountTestRepo) UpdateExtra(_ context.Context, _ int64, updates map[string]any) error {
+	r.updatedExtra = updates
+	return nil
+}
+
+func (r *openAIAccountTestRepo) SetRateLimited(_ context.Context, id int64, resetAt time.Time) error {
+	r.rateLimitedID = id
+	r.rateLimitedAt = &resetAt
+	return nil
+}
+
+func TestAccountTestService_OpenAI429PersistsSnapshotAndRateLimit(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	ctx, _ := newSoraTestContext()
+
+	resp := newJSONResponse(http.StatusTooManyRequests, `{"error":{"type":"usage_limit_reached","message":"limit reached"}}`)
+	resp.Header.Set("x-codex-primary-used-percent", "100")
+	resp.Header.Set("x-codex-primary-reset-after-seconds", "604800")
+	resp.Header.Set("x-codex-primary-window-minutes", "10080")
+	resp.Header.Set("x-codex-secondary-used-percent", "100")
+	resp.Header.Set("x-codex-secondary-reset-after-seconds", "18000")
+	resp.Header.Set("x-codex-secondary-window-minutes", "300")
+
+	repo := &openAIAccountTestRepo{}
+	upstream := &queuedHTTPUpstream{responses: []*http.Response{resp}}
+	svc := &AccountTestService{accountRepo: repo, httpUpstream: upstream}
+	account := &Account{
+		ID:          88,
+		Platform:    PlatformOpenAI,
+		Type:        AccountTypeOAuth,
+		Concurrency: 1,
+		Credentials: map[string]any{"access_token": "test-token"},
+	}
+
+	err := svc.testOpenAIAccountConnection(ctx, account, "gpt-5.4")
+	require.Error(t, err)
+	require.NotEmpty(t, repo.updatedExtra)
+	require.Equal(t, 100.0, repo.updatedExtra["codex_5h_used_percent"])
+	require.Equal(t, int64(88), repo.rateLimitedID)
+	require.NotNil(t, repo.rateLimitedAt)
+	require.NotNil(t, account.RateLimitResetAt)
+	if account.RateLimitResetAt != nil && repo.rateLimitedAt != nil {
+		require.WithinDuration(t, *repo.rateLimitedAt, *account.RateLimitResetAt, time.Second)
+	}
+}
diff --git a/backend/internal/service/account_usage_service.go b/backend/internal/service/account_usage_service.go
index b0a4900d..9bb3aa0b 100644
--- a/backend/internal/service/account_usage_service.go
+++ b/backend/internal/service/account_usage_service.go
@@ -367,7 +367,7 @@ func (s *AccountUsageService) getOpenAIUsage(ctx context.Context, account *Accou
 		usage.SevenDay = progress
 	}
 
-	if (usage.FiveHour == nil || usage.SevenDay == nil) && s.shouldProbeOpenAICodexSnapshot(account.ID, now) {
+	if shouldRefreshOpenAICodexSnapshot(account, usage, now) && s.shouldProbeOpenAICodexSnapshot(account.ID, now) {
 		if updates, err := s.probeOpenAICodexSnapshot(ctx, account); err == nil && len(updates) > 0 {
 			mergeAccountExtra(account, updates)
 			if usage.UpdatedAt == nil {
@@ -409,6 +409,40 @@ func (s *AccountUsageService) getOpenAIUsage(ctx context.Context, account *Accou
 	return usage, nil
 }
 
+func shouldRefreshOpenAICodexSnapshot(account *Account, usage *UsageInfo, now time.Time) bool {
+	if account == nil {
+		return false
+	}
+	if usage == nil {
+		return true
+	}
+	if usage.FiveHour == nil || usage.SevenDay == nil {
+		return true
+	}
+	if account.IsRateLimited() {
+		return true
+	}
+	return isOpenAICodexSnapshotStale(account, now)
+}
+
+func isOpenAICodexSnapshotStale(account *Account, now time.Time) bool {
+	if account == nil || !account.IsOpenAIOAuth() || !account.IsOpenAIResponsesWebSocketV2Enabled() {
+		return false
+	}
+	if account.Extra == nil {
+		return true
+	}
+	raw, ok := account.Extra["codex_usage_updated_at"]
+	if !ok {
+		return true
+	}
+	ts, err := parseTime(fmt.Sprint(raw))
+	if err != nil {
+		return true
+	}
+	return now.Sub(ts) >= openAIProbeCacheTTL
+}
+
 func (s *AccountUsageService) shouldProbeOpenAICodexSnapshot(accountID int64, now time.Time) bool {
 	if s == nil || s.cache == nil || accountID <= 0 {
 		return true
@@ -478,20 +512,34 @@ func (s *AccountUsageService) probeOpenAICodexSnapshot(ctx context.Context, acco
 	}
 	defer func() { _ = resp.Body.Close() }()
 
-	if resp.StatusCode < 200 || resp.StatusCode >= 300 {
-		return nil, fmt.Errorf("openai codex probe returned status %d", resp.StatusCode)
+	updates, err := extractOpenAICodexProbeUpdates(resp)
+	if err != nil {
+		return nil, err
+	}
+	if len(updates) > 0 {
+		go func(accountID int64, updates map[string]any) {
+			updateCtx, updateCancel := context.WithTimeout(context.Background(), 5*time.Second)
+			defer updateCancel()
+			_ = s.accountRepo.UpdateExtra(updateCtx, accountID, updates)
+		}(account.ID, updates)
+		return updates, nil
+	}
+	return nil, nil
+}
+
+func extractOpenAICodexProbeUpdates(resp *http.Response) (map[string]any, error) {
+	if resp == nil {
+		return nil, nil
 	}
 	if snapshot := ParseCodexRateLimitHeaders(resp.Header); snapshot != nil {
 		updates := buildCodexUsageExtraUpdates(snapshot, time.Now())
 		if len(updates) > 0 {
-			go func(accountID int64, updates map[string]any) {
-				updateCtx, updateCancel := context.WithTimeout(context.Background(), 5*time.Second)
-				defer updateCancel()
-				_ = s.accountRepo.UpdateExtra(updateCtx, accountID, updates)
-			}(account.ID, updates)
 			return updates, nil
 		}
 	}
+	if resp.StatusCode < 200 || resp.StatusCode >= 300 {
+		return nil, fmt.Errorf("openai codex probe returned status %d", resp.StatusCode)
+	}
 	return nil, nil
 }
 
diff --git a/backend/internal/service/account_usage_service_test.go b/backend/internal/service/account_usage_service_test.go
new file mode 100644
index 00000000..974d9029
--- /dev/null
+++ b/backend/internal/service/account_usage_service_test.go
@@ -0,0 +1,68 @@
+package service
+
+import (
+	"net/http"
+	"testing"
+	"time"
+)
+
+func TestShouldRefreshOpenAICodexSnapshot(t *testing.T) {
+	t.Parallel()
+
+	rateLimitedUntil := time.Now().Add(5 * time.Minute)
+	now := time.Now()
+	usage := &UsageInfo{
+		FiveHour: &UsageProgress{Utilization: 0},
+		SevenDay: &UsageProgress{Utilization: 0},
+	}
+
+	if !shouldRefreshOpenAICodexSnapshot(&Account{RateLimitResetAt: &rateLimitedUntil}, usage, now) {
+		t.Fatal("expected rate-limited account to force codex snapshot refresh")
+	}
+
+	if shouldRefreshOpenAICodexSnapshot(&Account{}, usage, now) {
+		t.Fatal("expected complete non-rate-limited usage to skip codex snapshot refresh")
+	}
+
+	if !shouldRefreshOpenAICodexSnapshot(&Account{}, &UsageInfo{FiveHour: nil, SevenDay: &UsageProgress{}}, now) {
+		t.Fatal("expected missing 5h snapshot to require refresh")
+	}
+
+	staleAt := now.Add(-(openAIProbeCacheTTL + time.Minute)).Format(time.RFC3339)
+	if !shouldRefreshOpenAICodexSnapshot(&Account{
+		Platform: PlatformOpenAI,
+		Type:     AccountTypeOAuth,
+		Extra: map[string]any{
+			"openai_oauth_responses_websockets_v2_enabled": true,
+			"codex_usage_updated_at":                       staleAt,
+		},
+	}, usage, now) {
+		t.Fatal("expected stale ws snapshot to trigger refresh")
+	}
+}
+
+func TestExtractOpenAICodexProbeUpdatesAccepts429WithCodexHeaders(t *testing.T) {
+	t.Parallel()
+
+	headers := make(http.Header)
+	headers.Set("x-codex-primary-used-percent", "100")
+	headers.Set("x-codex-primary-reset-after-seconds", "604800")
+	headers.Set("x-codex-primary-window-minutes", "10080")
+	headers.Set("x-codex-secondary-used-percent", "100")
+	headers.Set("x-codex-secondary-reset-after-seconds", "18000")
+	headers.Set("x-codex-secondary-window-minutes", "300")
+
+	updates, err := extractOpenAICodexProbeUpdates(&http.Response{StatusCode: http.StatusTooManyRequests, Header: headers})
+	if err != nil {
+		t.Fatalf("extractOpenAICodexProbeUpdates() error = %v", err)
+	}
+	if len(updates) == 0 {
+		t.Fatal("expected codex probe updates from 429 headers")
+	}
+	if got := updates["codex_5h_used_percent"]; got != 100.0 {
+		t.Fatalf("codex_5h_used_percent = %v, want 100", got)
+	}
+	if got := updates["codex_7d_used_percent"]; got != 100.0 {
+		t.Fatalf("codex_7d_used_percent = %v, want 100", got)
+	}
+}
diff --git a/backend/internal/service/ratelimit_service.go b/backend/internal/service/ratelimit_service.go
index f8f3154b..60ad99d0 100644
--- a/backend/internal/service/ratelimit_service.go
+++ b/backend/internal/service/ratelimit_service.go
@@ -615,6 +615,7 @@ func (s *RateLimitService) handleCustomErrorCode(ctx context.Context, account *A
 func (s *RateLimitService) handle429(ctx context.Context, account *Account, headers http.Header, responseBody []byte) {
 	// 1. OpenAI 平台：优先尝试解析 x-codex-* 响应头（用于 rate_limit_exceeded）
 	if account.Platform == PlatformOpenAI {
+		s.persistOpenAICodexSnapshot(ctx, account, headers)
 		if resetAt := s.calculateOpenAI429ResetTime(headers); resetAt != nil {
 			if err := s.accountRepo.SetRateLimited(ctx, account.ID, *resetAt); err != nil {
 				slog.Warn("rate_limit_set_failed", "account_id", account.ID, "error", err)
@@ -878,6 +879,23 @@ func pickSooner(a, b *time.Time) *time.Time {
 	}
 }
 
+func (s *RateLimitService) persistOpenAICodexSnapshot(ctx context.Context, account *Account, headers http.Header) {
+	if s == nil || s.accountRepo == nil || account == nil || headers == nil {
+		return
+	}
+	snapshot := ParseCodexRateLimitHeaders(headers)
+	if snapshot == nil {
+		return
+	}
+	updates := buildCodexUsageExtraUpdates(snapshot, time.Now())
+	if len(updates) == 0 {
+		return
+	}
+	if err := s.accountRepo.UpdateExtra(ctx, account.ID, updates); err != nil {
+		slog.Warn("openai_codex_snapshot_persist_failed", "account_id", account.ID, "error", err)
+	}
+}
+
 // parseOpenAIRateLimitResetTime 解析 OpenAI 格式的 429 响应，返回重置时间的 Unix 时间戳
 // OpenAI 的 usage_limit_reached 错误格式：
 //
diff --git a/backend/internal/service/ratelimit_service_openai_test.go b/backend/internal/service/ratelimit_service_openai_test.go
index 00902068..94b9a170 100644
--- a/backend/internal/service/ratelimit_service_openai_test.go
+++ b/backend/internal/service/ratelimit_service_openai_test.go
@@ -1,6 +1,7 @@
 package service
 
 import (
+	"context"
 	"net/http"
 	"testing"
 	"time"
@@ -141,6 +142,51 @@ func TestCalculateOpenAI429ResetTime_ReversedWindowOrder(t *testing.T) {
 	}
 }
 
+type openAI429SnapshotRepo struct {
+	mockAccountRepoForGemini
+	rateLimitedID int64
+	updatedExtra  map[string]any
+}
+
+func (r *openAI429SnapshotRepo) SetRateLimited(_ context.Context, id int64, _ time.Time) error {
+	r.rateLimitedID = id
+	return nil
+}
+
+func (r *openAI429SnapshotRepo) UpdateExtra(_ context.Context, _ int64, updates map[string]any) error {
+	r.updatedExtra = updates
+	return nil
+}
+
+func TestHandle429_OpenAIPersistsCodexSnapshotImmediately(t *testing.T) {
+	repo := &openAI429SnapshotRepo{}
+	svc := NewRateLimitService(repo, nil, nil, nil, nil)
+	account := &Account{ID: 123, Platform: PlatformOpenAI, Type: AccountTypeOAuth}
+
+	headers := http.Header{}
+	headers.Set("x-codex-primary-used-percent", "100")
+	headers.Set("x-codex-primary-reset-after-seconds", "604800")
+	headers.Set("x-codex-primary-window-minutes", "10080")
+	headers.Set("x-codex-secondary-used-percent", "100")
+	headers.Set("x-codex-secondary-reset-after-seconds", "18000")
+	headers.Set("x-codex-secondary-window-minutes", "300")
+
+	svc.handle429(context.Background(), account, headers, nil)
+
+	if repo.rateLimitedID != account.ID {
+		t.Fatalf("rateLimitedID = %d, want %d", repo.rateLimitedID, account.ID)
+	}
+	if len(repo.updatedExtra) == 0 {
+		t.Fatal("expected codex snapshot to be persisted on 429")
+	}
+	if got := repo.updatedExtra["codex_5h_used_percent"]; got != 100.0 {
+		t.Fatalf("codex_5h_used_percent = %v, want 100", got)
+	}
+	if got := repo.updatedExtra["codex_7d_used_percent"]; got != 100.0 {
+		t.Fatalf("codex_7d_used_percent = %v, want 100", got)
+	}
+}
+
 func TestNormalizedCodexLimits(t *testing.T) {
 	// Test the Normalize() method directly
 	pUsed := 100.0
diff --git a/frontend/src/components/account/AccountUsageCell.vue b/frontend/src/components/account/AccountUsageCell.vue
index 20b4b629..44c8e209 100644
--- a/frontend/src/components/account/AccountUsageCell.vue
+++ b/frontend/src/components/account/AccountUsageCell.vue
@@ -69,9 +69,39 @@
       <div v-else class="text-xs text-gray-400">-</div>
     </template>
 
-    <!-- OpenAI OAuth accounts: show Codex usage from extra field -->
+    <!-- OpenAI OAuth accounts: prefer fresh usage query for active rate-limited rows -->
     <template v-else-if="account.platform === 'openai' && account.type === 'oauth'">
-      <div v-if="hasCodexUsage" class="space-y-1">
+      <div v-if="preferFetchedOpenAIUsage" class="space-y-1">
+        <UsageProgressBar
+          v-if="usageInfo?.five_hour"
+          label="5h"
+          :utilization="usageInfo.five_hour.utilization"
+          :resets-at="usageInfo.five_hour.resets_at"
+          :window-stats="usageInfo.five_hour.window_stats"
+          color="indigo"
+        />
+        <UsageProgressBar
+          v-if="usageInfo?.seven_day"
+          label="7d"
+          :utilization="usageInfo.seven_day.utilization"
+          :resets-at="usageInfo.seven_day.resets_at"
+          :window-stats="usageInfo.seven_day.window_stats"
+          color="emerald"
+        />
+      </div>
+      <div v-else-if="isActiveOpenAIRateLimited && loading" class="space-y-1.5">
+        <div class="flex items-center gap-1">
+          <div class="h-3 w-[32px] animate-pulse rounded bg-gray-200 dark:bg-gray-700"></div>
+          <div class="h-1.5 w-8 animate-pulse rounded-full bg-gray-200 dark:bg-gray-700"></div>
+          <div class="h-3 w-[32px] animate-pulse rounded bg-gray-200 dark:bg-gray-700"></div>
+        </div>
+        <div class="flex items-center gap-1">
+          <div class="h-3 w-[32px] animate-pulse rounded bg-gray-200 dark:bg-gray-700"></div>
+          <div class="h-1.5 w-8 animate-pulse rounded-full bg-gray-200 dark:bg-gray-700"></div>
+          <div class="h-3 w-[32px] animate-pulse rounded bg-gray-200 dark:bg-gray-700"></div>
+        </div>
+      </div>
+      <div v-else-if="hasCodexUsage" class="space-y-1">
         <!-- 5h Window -->
         <UsageProgressBar
           v-if="codex5hUsedPercent !== null"
@@ -308,10 +338,11 @@
 </template>
 
 <script setup lang="ts">
-import { ref, computed, onMounted } from 'vue'
+import { ref, computed, onMounted, watch } from 'vue'
 import { useI18n } from 'vue-i18n'
 import { adminAPI } from '@/api/admin'
 import type { Account, AccountUsageInfo, GeminiCredentials, WindowStats } from '@/types'
+import { buildOpenAIUsageRefreshKey } from '@/utils/accountUsageRefresh'
 import { resolveCodexUsageWindow } from '@/utils/codexUsage'
 import UsageProgressBar from './UsageProgressBar.vue'
 import AccountQuotaInfo from './AccountQuotaInfo.vue'
@@ -373,6 +404,26 @@ const hasOpenAIUsageFallback = computed(() => {
   return !!usageInfo.value?.five_hour || !!usageInfo.value?.seven_day
 })
 
+const isActiveOpenAIRateLimited = computed(() => {
+  if (props.account.platform !== 'openai' || props.account.type !== 'oauth') return false
+  if (!props.account.rate_limit_reset_at) return false
+  const resetAt = Date.parse(props.account.rate_limit_reset_at)
+  return !Number.isNaN(resetAt) && resetAt > Date.now()
+})
+
+const preferFetchedOpenAIUsage = computed(() => {
+  return isActiveOpenAIRateLimited.value && hasOpenAIUsageFallback.value
+})
+
+const openAIUsageRefreshKey = computed(() => buildOpenAIUsageRefreshKey(props.account))
+
+const shouldAutoLoadUsageOnMount = computed(() => {
+  if (props.account.platform === 'openai' && props.account.type === 'oauth') {
+    return isActiveOpenAIRateLimited.value || !hasCodexUsage.value
+  }
+  return shouldFetchUsage.value
+})
+
 const codex5hUsedPercent = computed(() => codex5hWindow.value.usedPercent)
 const codex5hResetAt = computed(() => codex5hWindow.value.resetAt)
 const codex7dUsedPercent = computed(() => codex7dWindow.value.usedPercent)
@@ -749,6 +800,17 @@ const loadUsage = async () => {
 }
 
 onMounted(() => {
+  if (!shouldAutoLoadUsageOnMount.value) return
   loadUsage()
 })
+
+watch(openAIUsageRefreshKey, (nextKey, prevKey) => {
+  if (!prevKey || nextKey === prevKey) return
+  if (props.account.platform !== 'openai' || props.account.type !== 'oauth') return
+  if (!isActiveOpenAIRateLimited.value && hasCodexUsage.value) return
+
+  loadUsage().catch((e) => {
+    console.error('Failed to refresh OpenAI usage:', e)
+  })
+})
 </script>
diff --git a/frontend/src/components/account/UsageProgressBar.vue b/frontend/src/components/account/UsageProgressBar.vue
index 93844295..ea6c71f3 100644
--- a/frontend/src/components/account/UsageProgressBar.vue
+++ b/frontend/src/components/account/UsageProgressBar.vue
@@ -1,30 +1,5 @@
 <template>
   <div>
-    <!-- Window stats row (above progress bar, left-right aligned with progress bar) -->
-    <div
-      v-if="windowStats"
-      class="mb-0.5 flex items-center justify-between"
-      :title="statsTitle || t('admin.accounts.usageWindow.statsTitle')"
-    >
-      <div
-        class="flex cursor-help items-center gap-1.5 text-[9px] text-gray-500 dark:text-gray-400"
-      >
-        <span class="rounded bg-gray-100 px-1.5 py-0.5 dark:bg-gray-800">
-          {{ formatRequests }} req
-        </span>
-        <span class="rounded bg-gray-100 px-1.5 py-0.5 dark:bg-gray-800">
-          {{ formatTokens }}
-        </span>
-        <span class="rounded bg-gray-100 px-1.5 py-0.5 dark:bg-gray-800"> A ${{ formatAccountCost }} </span>
-        <span
-          v-if="windowStats?.user_cost != null"
-          class="rounded bg-gray-100 px-1.5 py-0.5 dark:bg-gray-800"
-        >
-          U ${{ formatUserCost }}
-        </span>
-      </div>
-    </div>
-
     <!-- Progress bar row -->
     <div class="flex items-center gap-1">
       <!-- Label badge (fixed width for alignment) -->
@@ -57,7 +32,6 @@
 
 <script setup lang="ts">
 import { computed } from 'vue'
-import { useI18n } from 'vue-i18n'
 import type { WindowStats } from '@/types'
 
 const props = defineProps<{
@@ -66,11 +40,8 @@ const props = defineProps<{
   resetsAt?: string | null
   color: 'indigo' | 'emerald' | 'purple' | 'amber'
   windowStats?: WindowStats | null
-  statsTitle?: string
 }>()
 
-const { t } = useI18n()
-
 // Label background colors
 const labelClass = computed(() => {
   const colors = {
@@ -117,12 +88,12 @@ const displayPercent = computed(() => {
 
 // Format reset time
 const formatResetTime = computed(() => {
-  if (!props.resetsAt) return t('common.notAvailable')
+  if (!props.resetsAt) return '-'
   const date = new Date(props.resetsAt)
   const now = new Date()
   const diffMs = date.getTime() - now.getTime()
 
-  if (diffMs <= 0) return t('common.now')
+  if (diffMs <= 0) return '现在'
 
   const diffHours = Math.floor(diffMs / (1000 * 60 * 60))
   const diffMins = Math.floor((diffMs % (1000 * 60 * 60)) / (1000 * 60))
@@ -137,31 +108,4 @@ const formatResetTime = computed(() => {
   }
 })
 
-// Format window stats
-const formatRequests = computed(() => {
-  if (!props.windowStats) return ''
-  const r = props.windowStats.requests
-  if (r >= 1000000) return `${(r / 1000000).toFixed(1)}M`
-  if (r >= 1000) return `${(r / 1000).toFixed(1)}K`
-  return r.toString()
-})
-
-const formatTokens = computed(() => {
-  if (!props.windowStats) return ''
-  const t = props.windowStats.tokens
-  if (t >= 1000000000) return `${(t / 1000000000).toFixed(1)}B`
-  if (t >= 1000000) return `${(t / 1000000).toFixed(1)}M`
-  if (t >= 1000) return `${(t / 1000).toFixed(1)}K`
-  return t.toString()
-})
-
-const formatAccountCost = computed(() => {
-  if (!props.windowStats) return '0.00'
-  return props.windowStats.cost.toFixed(2)
-})
-
-const formatUserCost = computed(() => {
-  if (!props.windowStats || props.windowStats.user_cost == null) return '0.00'
-  return props.windowStats.user_cost.toFixed(2)
-})
 </script>
diff --git a/frontend/src/components/account/__tests__/AccountUsageCell.spec.ts b/frontend/src/components/account/__tests__/AccountUsageCell.spec.ts
index 12c512e3..400eafbb 100644
--- a/frontend/src/components/account/__tests__/AccountUsageCell.spec.ts
+++ b/frontend/src/components/account/__tests__/AccountUsageCell.spec.ts
@@ -68,6 +68,40 @@ describe('AccountUsageCell', () => {
     expect(wrapper.text()).toContain('admin.accounts.usageWindow.gemini3Image|70|2026-03-01T09:00:00Z')
   })
 
+
+  it('OpenAI OAuth 有现成快照且未限额时不会首屏请求 usage', async () => {
+    const wrapper = mount(AccountUsageCell, {
+      props: {
+        account: {
+          id: 2001,
+          platform: 'openai',
+          type: 'oauth',
+          extra: {
+            codex_5h_used_percent: 12,
+            codex_5h_reset_at: '2099-03-07T12:00:00Z',
+            codex_7d_used_percent: 34,
+            codex_7d_reset_at: '2099-03-13T12:00:00Z'
+          }
+        } as any
+      },
+      global: {
+        stubs: {
+          UsageProgressBar: {
+            props: ['label', 'utilization', 'resetsAt', 'windowStats', 'color'],
+            template: '<div class="usage-bar">{{ label }}|{{ utilization }}</div>'
+          },
+          AccountQuotaInfo: true
+        }
+      }
+    })
+
+    await flushPromises()
+
+    expect(getUsage).not.toHaveBeenCalled()
+    expect(wrapper.text()).toContain('5h|12')
+    expect(wrapper.text()).toContain('7d|34')
+  })
+
   it('OpenAI OAuth 在无 codex 快照时会回退显示 usage 接口窗口', async () => {
 	getUsage.mockResolvedValue({
 	  five_hour: {
@@ -122,4 +156,137 @@ describe('AccountUsageCell', () => {
 	expect(wrapper.text()).toContain('5h|0|27700')
 	expect(wrapper.text()).toContain('7d|0|27700')
   })
+
+  it('OpenAI OAuth 在行数据刷新但仍无 codex 快照时会重新拉取 usage', async () => {
+	getUsage
+	  .mockResolvedValueOnce({
+	    five_hour: {
+	      utilization: 0,
+	      resets_at: null,
+	      remaining_seconds: 0,
+	      window_stats: {
+	        requests: 1,
+	        tokens: 100,
+	        cost: 0.01,
+	        standard_cost: 0.01,
+	        user_cost: 0.01
+	      }
+	    },
+	    seven_day: null
+	  })
+	  .mockResolvedValueOnce({
+	    five_hour: {
+	      utilization: 0,
+	      resets_at: null,
+	      remaining_seconds: 0,
+	      window_stats: {
+	        requests: 2,
+	        tokens: 200,
+	        cost: 0.02,
+	        standard_cost: 0.02,
+	        user_cost: 0.02
+	      }
+	    },
+	    seven_day: null
+	  })
+
+	const wrapper = mount(AccountUsageCell, {
+	  props: {
+	    account: {
+	      id: 2003,
+	      platform: 'openai',
+	      type: 'oauth',
+	      updated_at: '2026-03-07T10:00:00Z',
+	      extra: {}
+	    } as any
+	  },
+	  global: {
+	    stubs: {
+	      UsageProgressBar: {
+	        props: ['label', 'utilization', 'resetsAt', 'windowStats', 'color'],
+	        template: '<div class="usage-bar">{{ label }}|{{ utilization }}|{{ windowStats?.tokens }}</div>'
+	      },
+	      AccountQuotaInfo: true
+	    }
+	  }
+	})
+
+	await flushPromises()
+	expect(wrapper.text()).toContain('5h|0|100')
+	expect(getUsage).toHaveBeenCalledTimes(1)
+
+	await wrapper.setProps({
+	  account: {
+	    id: 2003,
+	    platform: 'openai',
+	    type: 'oauth',
+	    updated_at: '2026-03-07T10:01:00Z',
+	    extra: {}
+	  }
+	})
+
+	await flushPromises()
+	expect(getUsage).toHaveBeenCalledTimes(2)
+	expect(wrapper.text()).toContain('5h|0|200')
+  })
+
+  it('OpenAI OAuth 已限额时首屏优先展示重新查询后的 usage，而不是旧 codex 快照', async () => {
+	getUsage.mockResolvedValue({
+	  five_hour: {
+	    utilization: 100,
+	    resets_at: '2026-03-07T12:00:00Z',
+	    remaining_seconds: 3600,
+	    window_stats: {
+	      requests: 211,
+	      tokens: 106540000,
+	      cost: 38.13,
+	      standard_cost: 38.13,
+	      user_cost: 38.13
+	    }
+	  },
+	  seven_day: {
+	    utilization: 100,
+	    resets_at: '2026-03-13T12:00:00Z',
+	    remaining_seconds: 3600,
+	    window_stats: {
+	      requests: 211,
+	      tokens: 106540000,
+	      cost: 38.13,
+	      standard_cost: 38.13,
+	      user_cost: 38.13
+	    }
+	  }
+	})
+
+	const wrapper = mount(AccountUsageCell, {
+	  props: {
+	    account: {
+	      id: 2004,
+	      platform: 'openai',
+	      type: 'oauth',
+	      rate_limit_reset_at: '2099-03-07T12:00:00Z',
+	      extra: {
+	        codex_5h_used_percent: 0,
+	        codex_7d_used_percent: 0
+	      }
+	    } as any
+	  },
+	  global: {
+	    stubs: {
+	      UsageProgressBar: {
+	        props: ['label', 'utilization', 'resetsAt', 'windowStats', 'color'],
+	        template: '<div class="usage-bar">{{ label }}|{{ utilization }}|{{ windowStats?.tokens }}</div>'
+	      },
+	      AccountQuotaInfo: true
+	    }
+	  }
+	})
+
+	await flushPromises()
+
+	expect(getUsage).toHaveBeenCalledWith(2004)
+	expect(wrapper.text()).toContain('5h|100|106540000')
+	expect(wrapper.text()).toContain('7d|100|106540000')
+	expect(wrapper.text()).not.toContain('5h|0|')
+  })
 })
diff --git a/frontend/src/utils/__tests__/accountUsageRefresh.spec.ts b/frontend/src/utils/__tests__/accountUsageRefresh.spec.ts
new file mode 100644
index 00000000..ae13d690
--- /dev/null
+++ b/frontend/src/utils/__tests__/accountUsageRefresh.spec.ts
@@ -0,0 +1,39 @@
+import { describe, expect, it } from 'vitest'
+import { buildOpenAIUsageRefreshKey } from '../accountUsageRefresh'
+
+describe('buildOpenAIUsageRefreshKey', () => {
+  it('会在 codex 快照变化时生成不同 key', () => {
+    const base = {
+      id: 1,
+      platform: 'openai',
+      type: 'oauth',
+      updated_at: '2026-03-07T10:00:00Z',
+      extra: {
+        codex_usage_updated_at: '2026-03-07T10:00:00Z',
+        codex_5h_used_percent: 0,
+        codex_7d_used_percent: 0
+      }
+    } as any
+
+    const next = {
+      ...base,
+      extra: {
+        ...base.extra,
+        codex_usage_updated_at: '2026-03-07T10:01:00Z',
+        codex_5h_used_percent: 100
+      }
+    }
+
+    expect(buildOpenAIUsageRefreshKey(base)).not.toBe(buildOpenAIUsageRefreshKey(next))
+  })
+
+  it('非 OpenAI OAuth 账号返回空 key', () => {
+    expect(buildOpenAIUsageRefreshKey({
+      id: 2,
+      platform: 'anthropic',
+      type: 'oauth',
+      updated_at: '2026-03-07T10:00:00Z',
+      extra: {}
+    } as any)).toBe('')
+  })
+})
diff --git a/frontend/src/utils/accountUsageRefresh.ts b/frontend/src/utils/accountUsageRefresh.ts
new file mode 100644
index 00000000..219ac57f
--- /dev/null
+++ b/frontend/src/utils/accountUsageRefresh.ts
@@ -0,0 +1,28 @@
+import type { Account } from '@/types'
+
+const normalizeUsageRefreshValue = (value: unknown): string => {
+  if (value == null) return ''
+  return String(value)
+}
+
+export const buildOpenAIUsageRefreshKey = (account: Pick<Account, 'id' | 'platform' | 'type' | 'updated_at' | 'rate_limit_reset_at' | 'extra'>): string => {
+  if (account.platform !== 'openai' || account.type !== 'oauth') {
+    return ''
+  }
+
+  const extra = account.extra ?? {}
+  return [
+    account.id,
+    account.updated_at,
+    account.rate_limit_reset_at,
+    extra.codex_usage_updated_at,
+    extra.codex_5h_used_percent,
+    extra.codex_5h_reset_at,
+    extra.codex_5h_reset_after_seconds,
+    extra.codex_5h_window_minutes,
+    extra.codex_7d_used_percent,
+    extra.codex_7d_reset_at,
+    extra.codex_7d_reset_after_seconds,
+    extra.codex_7d_window_minutes
+  ].map(normalizeUsageRefreshValue).join('|')
+}
diff --git a/frontend/src/views/admin/AccountsView.vue b/frontend/src/views/admin/AccountsView.vue
index 0173ea0a..65686e6d 100644
--- a/frontend/src/views/admin/AccountsView.vue
+++ b/frontend/src/views/admin/AccountsView.vue
@@ -309,6 +309,7 @@ import AccountCapacityCell from '@/components/account/AccountCapacityCell.vue'
 import PlatformTypeBadge from '@/components/common/PlatformTypeBadge.vue'
 import Icon from '@/components/icons/Icon.vue'
 import ErrorPassthroughRulesModal from '@/components/admin/ErrorPassthroughRulesModal.vue'
+import { buildOpenAIUsageRefreshKey } from '@/utils/accountUsageRefresh'
 import { formatDateTime, formatRelativeTime } from '@/utils/format'
 import type { Account, AccountPlatform, AccountType, Proxy, AdminGroup, WindowStats, ClaudeModel } from '@/types'
 
@@ -651,7 +652,8 @@ const shouldReplaceAutoRefreshRow = (current: Account, next: Account) => {
     current.status !== next.status ||
     current.rate_limit_reset_at !== next.rate_limit_reset_at ||
     current.overload_until !== next.overload_until ||
-    current.temp_unschedulable_until !== next.temp_unschedulable_until
+    current.temp_unschedulable_until !== next.temp_unschedulable_until ||
+    buildOpenAIUsageRefreshKey(current) !== buildOpenAIUsageRefreshKey(next)
   )
 }
 

From 101ef0cf6213ebf693677d677c2a2ef2f3f917f0 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E7=A5=9E=E4=B9=90?= <6682635@qq.com>
Date: Sat, 7 Mar 2026 20:59:17 +0800
Subject: [PATCH 2/6] =?UTF-8?q?fix:=20=E9=99=90=E6=B5=81=E8=B4=A6=E5=8F=B7?=
 =?UTF-8?q?=E8=87=AA=E5=8A=A8=E9=80=80=E5=87=BA=E8=B0=83=E5=BA=A6=E5=B9=B6?=
 =?UTF-8?q?=E4=BC=98=E5=8C=96=E6=8F=90=E7=A4=BA=E6=96=87=E6=A1=88?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 backend/internal/repository/account_repo.go   |  2 +
 .../service/openai_account_scheduler.go       | 39 ++++++----
 .../service/openai_account_scheduler_test.go  | 72 ++++++++++++++++++
 .../service/openai_gateway_service.go         | 74 +++++++++++++------
 .../service/openai_ws_account_sticky_test.go  | 37 ++++++++++
 .../internal/service/openai_ws_forwarder.go   |  2 +-
 .../account/AccountStatusIndicator.vue        | 19 +++--
 frontend/src/i18n/locales/en.ts               |  3 +-
 frontend/src/i18n/locales/zh.ts               |  3 +-
 9 files changed, 204 insertions(+), 47 deletions(-)

diff --git a/backend/internal/repository/account_repo.go b/backend/internal/repository/account_repo.go
index ffbfd466..0e56ebd1 100644
--- a/backend/internal/repository/account_repo.go
+++ b/backend/internal/repository/account_repo.go
@@ -925,6 +925,7 @@ func (r *accountRepository) SetRateLimited(ctx context.Context, id int64, resetA
 	if err := enqueueSchedulerOutbox(ctx, r.sql, service.SchedulerOutboxEventAccountChanged, &id, nil, nil); err != nil {
 		logger.LegacyPrintf("repository.account", "[SchedulerOutbox] enqueue rate limit failed: account=%d err=%v", id, err)
 	}
+	r.syncSchedulerAccountSnapshot(ctx, id)
 	return nil
 }
 
@@ -1040,6 +1041,7 @@ func (r *accountRepository) ClearRateLimit(ctx context.Context, id int64) error
 	if err := enqueueSchedulerOutbox(ctx, r.sql, service.SchedulerOutboxEventAccountChanged, &id, nil, nil); err != nil {
 		logger.LegacyPrintf("repository.account", "[SchedulerOutbox] enqueue clear rate limit failed: account=%d err=%v", id, err)
 	}
+	r.syncSchedulerAccountSnapshot(ctx, id)
 	return nil
 }
 
diff --git a/backend/internal/service/openai_account_scheduler.go b/backend/internal/service/openai_account_scheduler.go
index cf4bc26e..0fcf450b 100644
--- a/backend/internal/service/openai_account_scheduler.go
+++ b/backend/internal/service/openai_account_scheduler.go
@@ -319,7 +319,7 @@ func (s *defaultOpenAIAccountScheduler) selectBySessionHash(
 		_ = s.service.deleteStickySessionAccountID(ctx, req.GroupID, sessionHash)
 		return nil, nil
 	}
-	if shouldClearStickySession(account, req.RequestedModel) || !account.IsOpenAI() {
+	if shouldClearStickySession(account, req.RequestedModel) || !account.IsOpenAI() || !account.IsSchedulable() {
 		_ = s.service.deleteStickySessionAccountID(ctx, req.GroupID, sessionHash)
 		return nil, nil
 	}
@@ -687,16 +687,20 @@ func (s *defaultOpenAIAccountScheduler) selectByLoadBalance(
 
 	for i := 0; i < len(selectionOrder); i++ {
 		candidate := selectionOrder[i]
-		result, acquireErr := s.service.tryAcquireAccountSlot(ctx, candidate.account.ID, candidate.account.Concurrency)
+		fresh := s.service.resolveFreshSchedulableOpenAIAccount(ctx, candidate.account, req.RequestedModel)
+		if fresh == nil || !s.isAccountTransportCompatible(fresh, req.RequiredTransport) {
+			continue
+		}
+		result, acquireErr := s.service.tryAcquireAccountSlot(ctx, fresh.ID, fresh.Concurrency)
 		if acquireErr != nil {
 			return nil, len(candidates), topK, loadSkew, acquireErr
 		}
 		if result != nil && result.Acquired {
 			if req.SessionHash != "" {
-				_ = s.service.BindStickySession(ctx, req.GroupID, req.SessionHash, candidate.account.ID)
+				_ = s.service.BindStickySession(ctx, req.GroupID, req.SessionHash, fresh.ID)
 			}
 			return &AccountSelectionResult{
-				Account:     candidate.account,
+				Account:     fresh,
 				Acquired:    true,
 				ReleaseFunc: result.ReleaseFunc,
 			}, len(candidates), topK, loadSkew, nil
@@ -705,16 +709,23 @@ func (s *defaultOpenAIAccountScheduler) selectByLoadBalance(
 
 	cfg := s.service.schedulingConfig()
 	// WaitPlan.MaxConcurrency 使用 Concurrency（非 EffectiveLoadFactor），因为 WaitPlan 控制的是 Redis 实际并发槽位等待。
-	candidate := selectionOrder[0]
-	return &AccountSelectionResult{
-		Account: candidate.account,
-		WaitPlan: &AccountWaitPlan{
-			AccountID:      candidate.account.ID,
-			MaxConcurrency: candidate.account.Concurrency,
-			Timeout:        cfg.FallbackWaitTimeout,
-			MaxWaiting:     cfg.FallbackMaxWaiting,
-		},
-	}, len(candidates), topK, loadSkew, nil
+	for _, candidate := range selectionOrder {
+		fresh := s.service.resolveFreshSchedulableOpenAIAccount(ctx, candidate.account, req.RequestedModel)
+		if fresh == nil || !s.isAccountTransportCompatible(fresh, req.RequiredTransport) {
+			continue
+		}
+		return &AccountSelectionResult{
+			Account: fresh,
+			WaitPlan: &AccountWaitPlan{
+				AccountID:      fresh.ID,
+				MaxConcurrency: fresh.Concurrency,
+				Timeout:        cfg.FallbackWaitTimeout,
+				MaxWaiting:     cfg.FallbackMaxWaiting,
+			},
+		}, len(candidates), topK, loadSkew, nil
+	}
+
+	return nil, len(candidates), topK, loadSkew, errors.New("no available accounts")
 }
 
 func (s *defaultOpenAIAccountScheduler) isAccountTransportCompatible(account *Account, requiredTransport OpenAIUpstreamTransport) bool {
diff --git a/backend/internal/service/openai_account_scheduler_test.go b/backend/internal/service/openai_account_scheduler_test.go
index 7f6f1b66..977c4ee8 100644
--- a/backend/internal/service/openai_account_scheduler_test.go
+++ b/backend/internal/service/openai_account_scheduler_test.go
@@ -12,6 +12,78 @@ import (
 	"github.com/stretchr/testify/require"
 )
 
+type openAISnapshotCacheStub struct {
+	SchedulerCache
+	snapshotAccounts []*Account
+	accountsByID     map[int64]*Account
+}
+
+func (s *openAISnapshotCacheStub) GetSnapshot(ctx context.Context, bucket SchedulerBucket) ([]*Account, bool, error) {
+	if len(s.snapshotAccounts) == 0 {
+		return nil, false, nil
+	}
+	out := make([]*Account, 0, len(s.snapshotAccounts))
+	for _, account := range s.snapshotAccounts {
+		if account == nil {
+			continue
+		}
+		cloned := *account
+		out = append(out, &cloned)
+	}
+	return out, true, nil
+}
+
+func (s *openAISnapshotCacheStub) GetAccount(ctx context.Context, accountID int64) (*Account, error) {
+	if s.accountsByID == nil {
+		return nil, nil
+	}
+	account := s.accountsByID[accountID]
+	if account == nil {
+		return nil, nil
+	}
+	cloned := *account
+	return &cloned, nil
+}
+
+func TestOpenAIGatewayService_SelectAccountWithScheduler_SessionStickyRateLimitedAccountFallsBackToFreshCandidate(t *testing.T) {
+	ctx := context.Background()
+	groupID := int64(10101)
+	rateLimitedUntil := time.Now().Add(30 * time.Minute)
+	staleSticky := &Account{ID: 31001, Platform: PlatformOpenAI, Type: AccountTypeOAuth, Status: StatusActive, Schedulable: true, Concurrency: 1, Priority: 0}
+	staleBackup := &Account{ID: 31002, Platform: PlatformOpenAI, Type: AccountTypeOAuth, Status: StatusActive, Schedulable: true, Concurrency: 1, Priority: 5}
+	freshSticky := &Account{ID: 31001, Platform: PlatformOpenAI, Type: AccountTypeOAuth, Status: StatusActive, Schedulable: true, Concurrency: 1, Priority: 0, RateLimitResetAt: &rateLimitedUntil}
+	freshBackup := &Account{ID: 31002, Platform: PlatformOpenAI, Type: AccountTypeOAuth, Status: StatusActive, Schedulable: true, Concurrency: 1, Priority: 5}
+	cache := &stubGatewayCache{sessionBindings: map[string]int64{"openai:session_hash_rate_limited": 31001}}
+	snapshotCache := &openAISnapshotCacheStub{snapshotAccounts: []*Account{staleSticky, staleBackup}, accountsByID: map[int64]*Account{31001: freshSticky, 31002: freshBackup}}
+	snapshotService := &SchedulerSnapshotService{cache: snapshotCache}
+	svc := &OpenAIGatewayService{accountRepo: stubOpenAIAccountRepo{accounts: []Account{*freshSticky, *freshBackup}}, cache: cache, cfg: &config.Config{}, schedulerSnapshot: snapshotService, concurrencyService: NewConcurrencyService(stubConcurrencyCache{})}
+
+	selection, decision, err := svc.SelectAccountWithScheduler(ctx, &groupID, "", "session_hash_rate_limited", "gpt-5.1", nil, OpenAIUpstreamTransportAny)
+	require.NoError(t, err)
+	require.NotNil(t, selection)
+	require.NotNil(t, selection.Account)
+	require.Equal(t, int64(31002), selection.Account.ID)
+	require.Equal(t, openAIAccountScheduleLayerLoadBalance, decision.Layer)
+}
+
+func TestOpenAIGatewayService_SelectAccountForModelWithExclusions_SkipsFreshlyRateLimitedSnapshotCandidate(t *testing.T) {
+	ctx := context.Background()
+	groupID := int64(10102)
+	rateLimitedUntil := time.Now().Add(30 * time.Minute)
+	stalePrimary := &Account{ID: 32001, Platform: PlatformOpenAI, Type: AccountTypeOAuth, Status: StatusActive, Schedulable: true, Concurrency: 1, Priority: 0}
+	staleSecondary := &Account{ID: 32002, Platform: PlatformOpenAI, Type: AccountTypeOAuth, Status: StatusActive, Schedulable: true, Concurrency: 1, Priority: 5}
+	freshPrimary := &Account{ID: 32001, Platform: PlatformOpenAI, Type: AccountTypeOAuth, Status: StatusActive, Schedulable: true, Concurrency: 1, Priority: 0, RateLimitResetAt: &rateLimitedUntil}
+	freshSecondary := &Account{ID: 32002, Platform: PlatformOpenAI, Type: AccountTypeOAuth, Status: StatusActive, Schedulable: true, Concurrency: 1, Priority: 5}
+	snapshotCache := &openAISnapshotCacheStub{snapshotAccounts: []*Account{stalePrimary, staleSecondary}, accountsByID: map[int64]*Account{32001: freshPrimary, 32002: freshSecondary}}
+	snapshotService := &SchedulerSnapshotService{cache: snapshotCache}
+	svc := &OpenAIGatewayService{accountRepo: stubOpenAIAccountRepo{accounts: []Account{*freshPrimary, *freshSecondary}}, cfg: &config.Config{}, schedulerSnapshot: snapshotService}
+
+	account, err := svc.SelectAccountForModelWithExclusions(ctx, &groupID, "", "gpt-5.1", nil)
+	require.NoError(t, err)
+	require.NotNil(t, account)
+	require.Equal(t, int64(32002), account.ID)
+}
+
 func TestOpenAIGatewayService_SelectAccountWithScheduler_PreviousResponseSticky(t *testing.T) {
 	ctx := context.Background()
 	groupID := int64(9)
diff --git a/backend/internal/service/openai_gateway_service.go b/backend/internal/service/openai_gateway_service.go
index 5c8c2710..e642ff60 100644
--- a/backend/internal/service/openai_gateway_service.go
+++ b/backend/internal/service/openai_gateway_service.go
@@ -1026,7 +1026,7 @@ func (s *OpenAIGatewayService) selectAccountForModelWithExclusions(ctx context.C
 
 	// 3. 按优先级 + LRU 选择最佳账号
 	// Select by priority + LRU
-	selected := s.selectBestAccount(accounts, requestedModel, excludedIDs)
+	selected := s.selectBestAccount(ctx, accounts, requestedModel, excludedIDs)
 
 	if selected == nil {
 		if requestedModel != "" {
@@ -1099,7 +1099,7 @@ func (s *OpenAIGatewayService) tryStickySessionHit(ctx context.Context, groupID
 //
 // selectBestAccount selects the best account from candidates (priority + LRU).
 // Returns nil if no available account.
-func (s *OpenAIGatewayService) selectBestAccount(accounts []Account, requestedModel string, excludedIDs map[int64]struct{}) *Account {
+func (s *OpenAIGatewayService) selectBestAccount(ctx context.Context, accounts []Account, requestedModel string, excludedIDs map[int64]struct{}) *Account {
 	var selected *Account
 
 	for i := range accounts {
@@ -1111,27 +1111,20 @@ func (s *OpenAIGatewayService) selectBestAccount(accounts []Account, requestedMo
 			continue
 		}
 
-		// 调度器快照可能暂时过时，这里重新检查可调度性和平台
-		// Scheduler snapshots can be temporarily stale; re-check schedulability and platform
-		if !acc.IsSchedulable() || !acc.IsOpenAI() {
-			continue
-		}
-
-		// 检查模型支持
-		// Check model support
-		if requestedModel != "" && !acc.IsModelSupported(requestedModel) {
+		fresh := s.resolveFreshSchedulableOpenAIAccount(ctx, acc, requestedModel)
+		if fresh == nil {
 			continue
 		}
 
 		// 选择优先级最高且最久未使用的账号
 		// Select highest priority and least recently used
 		if selected == nil {
-			selected = acc
+			selected = fresh
 			continue
 		}
 
-		if s.isBetterAccount(acc, selected) {
-			selected = acc
+		if s.isBetterAccount(fresh, selected) {
+			selected = fresh
 		}
 	}
 
@@ -1309,13 +1302,17 @@ func (s *OpenAIGatewayService) SelectAccountWithLoadAwareness(ctx context.Contex
 		ordered := append([]*Account(nil), candidates...)
 		sortAccountsByPriorityAndLastUsed(ordered, false)
 		for _, acc := range ordered {
-			result, err := s.tryAcquireAccountSlot(ctx, acc.ID, acc.Concurrency)
+			fresh := s.resolveFreshSchedulableOpenAIAccount(ctx, acc, requestedModel)
+			if fresh == nil {
+				continue
+			}
+			result, err := s.tryAcquireAccountSlot(ctx, fresh.ID, fresh.Concurrency)
 			if err == nil && result.Acquired {
 				if sessionHash != "" {
-					_ = s.setStickySessionAccountID(ctx, groupID, sessionHash, acc.ID, openaiStickySessionTTL)
+					_ = s.setStickySessionAccountID(ctx, groupID, sessionHash, fresh.ID, openaiStickySessionTTL)
 				}
 				return &AccountSelectionResult{
-					Account:     acc,
+					Account:     fresh,
 					Acquired:    true,
 					ReleaseFunc: result.ReleaseFunc,
 				}, nil
@@ -1359,13 +1356,17 @@ func (s *OpenAIGatewayService) SelectAccountWithLoadAwareness(ctx context.Contex
 			shuffleWithinSortGroups(available)
 
 			for _, item := range available {
-				result, err := s.tryAcquireAccountSlot(ctx, item.account.ID, item.account.Concurrency)
+				fresh := s.resolveFreshSchedulableOpenAIAccount(ctx, item.account, requestedModel)
+				if fresh == nil {
+					continue
+				}
+				result, err := s.tryAcquireAccountSlot(ctx, fresh.ID, fresh.Concurrency)
 				if err == nil && result.Acquired {
 					if sessionHash != "" {
-						_ = s.setStickySessionAccountID(ctx, groupID, sessionHash, item.account.ID, openaiStickySessionTTL)
+						_ = s.setStickySessionAccountID(ctx, groupID, sessionHash, fresh.ID, openaiStickySessionTTL)
 					}
 					return &AccountSelectionResult{
-						Account:     item.account,
+						Account:     fresh,
 						Acquired:    true,
 						ReleaseFunc: result.ReleaseFunc,
 					}, nil
@@ -1377,11 +1378,15 @@ func (s *OpenAIGatewayService) SelectAccountWithLoadAwareness(ctx context.Contex
 	// ============ Layer 3: Fallback wait ============
 	sortAccountsByPriorityAndLastUsed(candidates, false)
 	for _, acc := range candidates {
+		fresh := s.resolveFreshSchedulableOpenAIAccount(ctx, acc, requestedModel)
+		if fresh == nil {
+			continue
+		}
 		return &AccountSelectionResult{
-			Account: acc,
+			Account: fresh,
 			WaitPlan: &AccountWaitPlan{
-				AccountID:      acc.ID,
-				MaxConcurrency: acc.Concurrency,
+				AccountID:      fresh.ID,
+				MaxConcurrency: fresh.Concurrency,
 				Timeout:        cfg.FallbackWaitTimeout,
 				MaxWaiting:     cfg.FallbackMaxWaiting,
 			},
@@ -1418,6 +1423,29 @@ func (s *OpenAIGatewayService) tryAcquireAccountSlot(ctx context.Context, accoun
 	return s.concurrencyService.AcquireAccountSlot(ctx, accountID, maxConcurrency)
 }
 
+func (s *OpenAIGatewayService) resolveFreshSchedulableOpenAIAccount(ctx context.Context, account *Account, requestedModel string) *Account {
+	if account == nil {
+		return nil
+	}
+
+	fresh := account
+	if s.schedulerSnapshot != nil {
+		current, err := s.getSchedulableAccount(ctx, account.ID)
+		if err != nil || current == nil {
+			return nil
+		}
+		fresh = current
+	}
+
+	if !fresh.IsSchedulable() || !fresh.IsOpenAI() {
+		return nil
+	}
+	if requestedModel != "" && !fresh.IsModelSupported(requestedModel) {
+		return nil
+	}
+	return fresh
+}
+
 func (s *OpenAIGatewayService) getSchedulableAccount(ctx context.Context, accountID int64) (*Account, error) {
 	if s.schedulerSnapshot != nil {
 		return s.schedulerSnapshot.GetAccount(ctx, accountID)
diff --git a/backend/internal/service/openai_ws_account_sticky_test.go b/backend/internal/service/openai_ws_account_sticky_test.go
index 3fe08179..9a8803d3 100644
--- a/backend/internal/service/openai_ws_account_sticky_test.go
+++ b/backend/internal/service/openai_ws_account_sticky_test.go
@@ -48,6 +48,43 @@ func TestOpenAIGatewayService_SelectAccountByPreviousResponseID_Hit(t *testing.T
 	}
 }
 
+func TestOpenAIGatewayService_SelectAccountByPreviousResponseID_RateLimitedMiss(t *testing.T) {
+	ctx := context.Background()
+	groupID := int64(23)
+	rateLimitedUntil := time.Now().Add(30 * time.Minute)
+	account := Account{
+		ID:               12,
+		Platform:         PlatformOpenAI,
+		Type:             AccountTypeAPIKey,
+		Status:           StatusActive,
+		Schedulable:      true,
+		Concurrency:      1,
+		RateLimitResetAt: &rateLimitedUntil,
+		Extra: map[string]any{
+			"openai_apikey_responses_websockets_v2_enabled": true,
+		},
+	}
+	cache := &stubGatewayCache{}
+	store := NewOpenAIWSStateStore(cache)
+	cfg := newOpenAIWSV2TestConfig()
+	svc := &OpenAIGatewayService{
+		accountRepo:        stubOpenAIAccountRepo{accounts: []Account{account}},
+		cache:              cache,
+		cfg:                cfg,
+		concurrencyService: NewConcurrencyService(stubConcurrencyCache{}),
+		openaiWSStateStore: store,
+	}
+
+	require.NoError(t, store.BindResponseAccount(ctx, groupID, "resp_prev_rl", account.ID, time.Hour))
+
+	selection, err := svc.SelectAccountByPreviousResponseID(ctx, &groupID, "resp_prev_rl", "gpt-5.1", nil)
+	require.NoError(t, err)
+	require.Nil(t, selection, "限额中的账号不应继续命中 previous_response_id 粘连")
+	boundAccountID, getErr := store.GetResponseAccount(ctx, groupID, "resp_prev_rl")
+	require.NoError(t, getErr)
+	require.Zero(t, boundAccountID)
+}
+
 func TestOpenAIGatewayService_SelectAccountByPreviousResponseID_Excluded(t *testing.T) {
 	ctx := context.Background()
 	groupID := int64(23)
diff --git a/backend/internal/service/openai_ws_forwarder.go b/backend/internal/service/openai_ws_forwarder.go
index f9e93f85..36bf8ff8 100644
--- a/backend/internal/service/openai_ws_forwarder.go
+++ b/backend/internal/service/openai_ws_forwarder.go
@@ -3798,7 +3798,7 @@ func (s *OpenAIGatewayService) SelectAccountByPreviousResponseID(
 	if s.getOpenAIWSProtocolResolver().Resolve(account).Transport != OpenAIUpstreamTransportResponsesWebsocketV2 {
 		return nil, nil
 	}
-	if shouldClearStickySession(account, requestedModel) || !account.IsOpenAI() {
+	if shouldClearStickySession(account, requestedModel) || !account.IsOpenAI() || !account.IsSchedulable() {
 		_ = store.DeleteResponseAccount(ctx, derefGroupID(groupID), responseID)
 		return nil, nil
 	}
diff --git a/frontend/src/components/account/AccountStatusIndicator.vue b/frontend/src/components/account/AccountStatusIndicator.vue
index e8331c25..1dc4f287 100644
--- a/frontend/src/components/account/AccountStatusIndicator.vue
+++ b/frontend/src/components/account/AccountStatusIndicator.vue
@@ -3,7 +3,7 @@
     <!-- Rate Limit Display (429) - Two-line layout -->
     <div v-if="isRateLimited" class="flex flex-col items-center gap-1">
       <span class="badge text-xs badge-warning">{{ t('admin.accounts.status.rateLimited') }}</span>
-      <span class="text-[11px] text-gray-400 dark:text-gray-500">{{ rateLimitCountdown }}</span>
+      <span class="text-[11px] text-gray-400 dark:text-gray-500">{{ rateLimitResumeText }}</span>
     </div>
 
     <!-- Overload Display (529) - Two-line layout -->
@@ -67,9 +67,9 @@
       </span>
       <!-- Tooltip -->
       <div
-        class="pointer-events-none absolute bottom-full left-1/2 z-50 mb-2 -translate-x-1/2 whitespace-nowrap rounded bg-gray-900 px-2 py-1 text-xs text-white opacity-0 transition-opacity group-hover:opacity-100 dark:bg-gray-700"
+        class="pointer-events-none absolute bottom-full left-1/2 z-50 mb-2 w-56 -translate-x-1/2 whitespace-normal rounded bg-gray-900 px-3 py-2 text-center text-xs leading-relaxed text-white opacity-0 transition-opacity group-hover:opacity-100 dark:bg-gray-700"
       >
-        {{ t('admin.accounts.status.rateLimitedUntil', { time: formatTime(account.rate_limit_reset_at) }) }}
+        {{ t('admin.accounts.status.rateLimitedUntil', { time: formatDateTime(account.rate_limit_reset_at) }) }}
         <div
           class="absolute left-1/2 top-full -translate-x-1/2 border-4 border-transparent border-t-gray-900 dark:border-t-gray-700"
         ></div>
@@ -97,7 +97,7 @@
         </span>
         <!-- Tooltip -->
         <div
-          class="pointer-events-none absolute bottom-full left-1/2 z-50 mb-2 -translate-x-1/2 whitespace-nowrap rounded bg-gray-900 px-2 py-1 text-xs text-white opacity-0 transition-opacity group-hover:opacity-100 dark:bg-gray-700"
+          class="pointer-events-none absolute bottom-full left-1/2 z-50 mb-2 w-56 -translate-x-1/2 whitespace-normal rounded bg-gray-900 px-3 py-2 text-center text-xs leading-relaxed text-white opacity-0 transition-opacity group-hover:opacity-100 dark:bg-gray-700"
         >
           {{ t('admin.accounts.status.modelRateLimitedUntil', { model: formatScopeName(item.model), time: formatTime(item.reset_at) }) }}
           <div
@@ -117,7 +117,7 @@
       </span>
       <!-- Tooltip -->
       <div
-        class="pointer-events-none absolute bottom-full left-1/2 z-50 mb-2 -translate-x-1/2 whitespace-nowrap rounded bg-gray-900 px-2 py-1 text-xs text-white opacity-0 transition-opacity group-hover:opacity-100 dark:bg-gray-700"
+        class="pointer-events-none absolute bottom-full left-1/2 z-50 mb-2 w-56 -translate-x-1/2 whitespace-normal rounded bg-gray-900 px-3 py-2 text-center text-xs leading-relaxed text-white opacity-0 transition-opacity group-hover:opacity-100 dark:bg-gray-700"
       >
         {{ t('admin.accounts.status.overloadedUntil', { time: formatTime(account.overload_until) }) }}
         <div
@@ -132,7 +132,7 @@
 import { computed } from 'vue'
 import { useI18n } from 'vue-i18n'
 import type { Account } from '@/types'
-import { formatCountdownWithSuffix, formatTime } from '@/utils/format'
+import { formatCountdown, formatDateTime, formatCountdownWithSuffix, formatTime } from '@/utils/format'
 
 const { t } = useI18n()
 
@@ -231,7 +231,12 @@ const hasError = computed(() => {
 
 // Computed: countdown text for rate limit (429)
 const rateLimitCountdown = computed(() => {
-  return formatCountdownWithSuffix(props.account.rate_limit_reset_at)
+  return formatCountdown(props.account.rate_limit_reset_at)
+})
+
+const rateLimitResumeText = computed(() => {
+  if (!rateLimitCountdown.value) return ''
+  return t('admin.accounts.status.rateLimitedAutoResume', { time: rateLimitCountdown.value })
 })
 
 // Computed: countdown text for overload (529)
diff --git a/frontend/src/i18n/locales/en.ts b/frontend/src/i18n/locales/en.ts
index 1efff120..862fb632 100644
--- a/frontend/src/i18n/locales/en.ts
+++ b/frontend/src/i18n/locales/en.ts
@@ -1694,7 +1694,8 @@ export default {
         rateLimited: 'Rate Limited',
         overloaded: 'Overloaded',
         tempUnschedulable: 'Temp Unschedulable',
-        rateLimitedUntil: 'Rate limited until {time}',
+        rateLimitedUntil: 'Rate limited and removed from scheduling. Auto resumes at {time}',
+        rateLimitedAutoResume: 'Auto resumes in {time}',
         modelRateLimitedUntil: '{model} rate limited until {time}',
         overloadedUntil: 'Overloaded until {time}',
         viewTempUnschedDetails: 'View temp unschedulable details'
diff --git a/frontend/src/i18n/locales/zh.ts b/frontend/src/i18n/locales/zh.ts
index b2c38928..2cc5e8b1 100644
--- a/frontend/src/i18n/locales/zh.ts
+++ b/frontend/src/i18n/locales/zh.ts
@@ -1853,7 +1853,8 @@ export default {
         rateLimited: '限流中',
         overloaded: '过载中',
         tempUnschedulable: '临时不可调度',
-        rateLimitedUntil: '限流中，重置时间：{time}',
+        rateLimitedUntil: '限流中，当前不参与调度，预计 {time} 自动恢复',
+        rateLimitedAutoResume: '{time} 自动恢复',
         modelRateLimitedUntil: '{model} 限流至 {time}',
         overloadedUntil: '负载过重，重置时间：{time}',
         viewTempUnschedDetails: '查看临时不可调度详情'

From 0c1dcad42988c8850cad40aa9bf8b15791f809d4 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E7=A5=9E=E4=B9=90?= <6682635@qq.com>
Date: Sat, 7 Mar 2026 21:09:34 +0800
Subject: [PATCH 3/6] =?UTF-8?q?test:=20=E4=BF=AE=E5=A4=8D=20PR842=20?=
 =?UTF-8?q?=E7=9A=84=20CI=20=E5=A4=B1=E8=B4=A5?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 backend/internal/server/api_contract_test.go              | 1 +
 backend/internal/service/ratelimit_service_openai_test.go | 2 ++
 2 files changed, 3 insertions(+)

diff --git a/backend/internal/server/api_contract_test.go b/backend/internal/server/api_contract_test.go
index aafbbe21..a0b4542b 100644
--- a/backend/internal/server/api_contract_test.go
+++ b/backend/internal/server/api_contract_test.go
@@ -210,6 +210,7 @@ func TestAPIContracts(t *testing.T) {
 							"sora_video_price_per_request": null,
 							"sora_video_price_per_request_hd": null,
 							"claude_code_only": false,
+						"allow_messages_dispatch": false,
 						"fallback_group_id": null,
 						"fallback_group_id_on_invalid_request": null,
 						"created_at": "2025-01-02T03:04:05Z",
diff --git a/backend/internal/service/ratelimit_service_openai_test.go b/backend/internal/service/ratelimit_service_openai_test.go
index 94b9a170..89c754c8 100644
--- a/backend/internal/service/ratelimit_service_openai_test.go
+++ b/backend/internal/service/ratelimit_service_openai_test.go
@@ -1,3 +1,5 @@
+//go:build unit
+
 package service
 
 import (

From 45d57018eb1bc2a37a3aa8d2531d57ec0eb023a4 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E7=A5=9E=E4=B9=90?= <6682635@qq.com>
Date: Sat, 7 Mar 2026 23:59:39 +0800
Subject: [PATCH 4/6] =?UTF-8?q?fix:=20=E4=BF=AE=E5=A4=8D=20OpenAI=20WS=20?=
 =?UTF-8?q?=E9=99=90=E6=B5=81=E7=8A=B6=E6=80=81=E4=B8=8E=E8=B0=83=E5=BA=A6?=
 =?UTF-8?q?=E5=90=8C=E6=AD=A5?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 .../service/openai_gateway_service.go         |  38 +-
 .../internal/service/openai_ws_forwarder.go   |  48 ++-
 .../openai_ws_ratelimit_signal_test.go        | 392 ++++++++++++++++++
 3 files changed, 471 insertions(+), 7 deletions(-)
 create mode 100644 backend/internal/service/openai_ws_ratelimit_signal_test.go

diff --git a/backend/internal/service/openai_gateway_service.go b/backend/internal/service/openai_gateway_service.go
index e642ff60..40a4e377 100644
--- a/backend/internal/service/openai_gateway_service.go
+++ b/backend/internal/service/openai_gateway_service.go
@@ -3899,6 +3899,30 @@ func buildCodexUsageExtraUpdates(snapshot *OpenAICodexUsageSnapshot, fallbackNow
 	return updates
 }
 
+func codexUsagePercentExhausted(value *float64) bool {
+	return value != nil && *value >= 100-1e-9
+}
+
+func codexRateLimitResetAtFromSnapshot(snapshot *OpenAICodexUsageSnapshot, fallbackNow time.Time) *time.Time {
+	if snapshot == nil {
+		return nil
+	}
+	normalized := snapshot.Normalize()
+	if normalized == nil {
+		return nil
+	}
+	baseTime := codexSnapshotBaseTime(snapshot, fallbackNow)
+	if codexUsagePercentExhausted(normalized.Used7dPercent) && normalized.Reset7dSeconds != nil {
+		resetAt := baseTime.Add(time.Duration(*normalized.Reset7dSeconds) * time.Second)
+		return &resetAt
+	}
+	if codexUsagePercentExhausted(normalized.Used5hPercent) && normalized.Reset5hSeconds != nil {
+		resetAt := baseTime.Add(time.Duration(*normalized.Reset5hSeconds) * time.Second)
+		return &resetAt
+	}
+	return nil
+}
+
 // updateCodexUsageSnapshot saves the Codex usage snapshot to account's Extra field
 func (s *OpenAIGatewayService) updateCodexUsageSnapshot(ctx context.Context, accountID int64, snapshot *OpenAICodexUsageSnapshot) {
 	if snapshot == nil {
@@ -3908,16 +3932,22 @@ func (s *OpenAIGatewayService) updateCodexUsageSnapshot(ctx context.Context, acc
 		return
 	}
 
-	updates := buildCodexUsageExtraUpdates(snapshot, time.Now())
-	if len(updates) == 0 {
+	now := time.Now()
+	updates := buildCodexUsageExtraUpdates(snapshot, now)
+	resetAt := codexRateLimitResetAtFromSnapshot(snapshot, now)
+	if len(updates) == 0 && resetAt == nil {
 		return
 	}
 
-	// Update account's Extra field asynchronously
 	go func() {
 		updateCtx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
 		defer cancel()
-		_ = s.accountRepo.UpdateExtra(updateCtx, accountID, updates)
+		if len(updates) > 0 {
+			_ = s.accountRepo.UpdateExtra(updateCtx, accountID, updates)
+		}
+		if resetAt != nil {
+			_ = s.accountRepo.SetRateLimited(updateCtx, accountID, *resetAt)
+		}
 	}()
 }
 
diff --git a/backend/internal/service/openai_ws_forwarder.go b/backend/internal/service/openai_ws_forwarder.go
index 36bf8ff8..f2f8edd9 100644
--- a/backend/internal/service/openai_ws_forwarder.go
+++ b/backend/internal/service/openai_ws_forwarder.go
@@ -1853,6 +1853,10 @@ func (s *OpenAIGatewayService) forwardOpenAIWSV2(
 			wsPath,
 			account.ProxyID != nil && account.Proxy != nil,
 		)
+		var dialErr *openAIWSDialError
+		if errors.As(err, &dialErr) && dialErr != nil && dialErr.StatusCode == http.StatusTooManyRequests {
+			s.persistOpenAIWSRateLimitSignal(ctx, account, dialErr.ResponseHeaders, nil, "rate_limit_exceeded", "rate_limit_error", strings.TrimSpace(err.Error()))
+		}
 		return nil, wrapOpenAIWSFallback(classifyOpenAIWSAcquireError(err), err)
 	}
 	defer lease.Release()
@@ -2136,6 +2140,7 @@ func (s *OpenAIGatewayService) forwardOpenAIWSV2(
 
 		if eventType == "error" {
 			errCodeRaw, errTypeRaw, errMsgRaw := parseOpenAIWSErrorEventFields(message)
+			s.persistOpenAIWSRateLimitSignal(ctx, account, lease.HandshakeHeaders(), message, errCodeRaw, errTypeRaw, errMsgRaw)
 			errMsg := strings.TrimSpace(errMsgRaw)
 			if errMsg == "" {
 				errMsg = "Upstream websocket error"
@@ -2639,6 +2644,10 @@ func (s *OpenAIGatewayService) ProxyResponsesWebSocketFromClient(
 				wsPath,
 				account.ProxyID != nil && account.Proxy != nil,
 			)
+			var dialErr *openAIWSDialError
+			if errors.As(acquireErr, &dialErr) && dialErr != nil && dialErr.StatusCode == http.StatusTooManyRequests {
+				s.persistOpenAIWSRateLimitSignal(ctx, account, dialErr.ResponseHeaders, nil, "rate_limit_exceeded", "rate_limit_error", strings.TrimSpace(acquireErr.Error()))
+			}
 			if errors.Is(acquireErr, errOpenAIWSPreferredConnUnavailable) {
 				return nil, NewOpenAIWSClientCloseError(
 					coderws.StatusPolicyViolation,
@@ -2777,6 +2786,7 @@ func (s *OpenAIGatewayService) ProxyResponsesWebSocketFromClient(
 			}
 			if eventType == "error" {
 				errCodeRaw, errTypeRaw, errMsgRaw := parseOpenAIWSErrorEventFields(upstreamMessage)
+				s.persistOpenAIWSRateLimitSignal(ctx, account, lease.HandshakeHeaders(), upstreamMessage, errCodeRaw, errTypeRaw, errMsgRaw)
 				fallbackReason, _ := classifyOpenAIWSErrorEventFromRaw(errCodeRaw, errTypeRaw, errMsgRaw)
 				errCode, errType, errMessage := summarizeOpenAIWSErrorEventFieldsFromRaw(errCodeRaw, errTypeRaw, errMsgRaw)
 				recoverablePrevNotFound := fallbackReason == openAIWSIngressStagePreviousResponseNotFound &&
@@ -3604,6 +3614,7 @@ func (s *OpenAIGatewayService) performOpenAIWSGeneratePrewarm(
 
 		if eventType == "error" {
 			errCodeRaw, errTypeRaw, errMsgRaw := parseOpenAIWSErrorEventFields(message)
+			s.persistOpenAIWSRateLimitSignal(ctx, account, lease.HandshakeHeaders(), message, errCodeRaw, errTypeRaw, errMsgRaw)
 			errMsg := strings.TrimSpace(errMsgRaw)
 			if errMsg == "" {
 				errMsg = "OpenAI websocket prewarm error"
@@ -3867,6 +3878,36 @@ func classifyOpenAIWSAcquireError(err error) string {
 	return "acquire_conn"
 }
 
+func isOpenAIWSRateLimitError(codeRaw, errTypeRaw, msgRaw string) bool {
+	code := strings.ToLower(strings.TrimSpace(codeRaw))
+	errType := strings.ToLower(strings.TrimSpace(errTypeRaw))
+	msg := strings.ToLower(strings.TrimSpace(msgRaw))
+
+	if strings.Contains(errType, "rate_limit") || strings.Contains(errType, "usage_limit") {
+		return true
+	}
+	if strings.Contains(code, "rate_limit") || strings.Contains(code, "usage_limit") || strings.Contains(code, "insufficient_quota") {
+		return true
+	}
+	if strings.Contains(msg, "usage limit") && strings.Contains(msg, "reached") {
+		return true
+	}
+	if strings.Contains(msg, "rate limit") && (strings.Contains(msg, "reached") || strings.Contains(msg, "exceeded")) {
+		return true
+	}
+	return false
+}
+
+func (s *OpenAIGatewayService) persistOpenAIWSRateLimitSignal(ctx context.Context, account *Account, headers http.Header, responseBody []byte, codeRaw, errTypeRaw, msgRaw string) {
+	if s == nil || s.rateLimitService == nil || account == nil || account.Platform != PlatformOpenAI {
+		return
+	}
+	if !isOpenAIWSRateLimitError(codeRaw, errTypeRaw, msgRaw) {
+		return
+	}
+	s.rateLimitService.HandleUpstreamError(ctx, account, http.StatusTooManyRequests, headers, responseBody)
+}
+
 func classifyOpenAIWSErrorEventFromRaw(codeRaw, errTypeRaw, msgRaw string) (string, bool) {
 	code := strings.ToLower(strings.TrimSpace(codeRaw))
 	errType := strings.ToLower(strings.TrimSpace(errTypeRaw))
@@ -3882,6 +3923,9 @@ func classifyOpenAIWSErrorEventFromRaw(codeRaw, errTypeRaw, msgRaw string) (stri
 	case "previous_response_not_found":
 		return "previous_response_not_found", true
 	}
+	if isOpenAIWSRateLimitError(codeRaw, errTypeRaw, msgRaw) {
+		return "upstream_rate_limited", false
+	}
 	if strings.Contains(msg, "upgrade required") || strings.Contains(msg, "status 426") {
 		return "upgrade_required", true
 	}
@@ -3927,9 +3971,7 @@ func openAIWSErrorHTTPStatusFromRaw(codeRaw, errTypeRaw string) int {
 	case strings.Contains(errType, "permission"),
 		strings.Contains(code, "forbidden"):
 		return http.StatusForbidden
-	case strings.Contains(errType, "rate_limit"),
-		strings.Contains(code, "rate_limit"),
-		strings.Contains(code, "insufficient_quota"):
+	case isOpenAIWSRateLimitError(codeRaw, errTypeRaw, ""):
 		return http.StatusTooManyRequests
 	default:
 		return http.StatusBadGateway
diff --git a/backend/internal/service/openai_ws_ratelimit_signal_test.go b/backend/internal/service/openai_ws_ratelimit_signal_test.go
new file mode 100644
index 00000000..a6b6e874
--- /dev/null
+++ b/backend/internal/service/openai_ws_ratelimit_signal_test.go
@@ -0,0 +1,392 @@
+package service
+
+import (
+	"context"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"strconv"
+	"strings"
+	"testing"
+	"time"
+
+	coderws "github.com/coder/websocket"
+	"github.com/gin-gonic/gin"
+	"github.com/gorilla/websocket"
+	"github.com/stretchr/testify/require"
+)
+
+type openAIWSRateLimitSignalRepo struct {
+	stubOpenAIAccountRepo
+	rateLimitCalls []time.Time
+	updateExtra    []map[string]any
+}
+
+type openAICodexSnapshotAsyncRepo struct {
+	stubOpenAIAccountRepo
+	updateExtraCh chan map[string]any
+	rateLimitCh   chan time.Time
+}
+
+func (r *openAIWSRateLimitSignalRepo) SetRateLimited(_ context.Context, _ int64, resetAt time.Time) error {
+	r.rateLimitCalls = append(r.rateLimitCalls, resetAt)
+	return nil
+}
+
+func (r *openAIWSRateLimitSignalRepo) UpdateExtra(_ context.Context, _ int64, updates map[string]any) error {
+	copied := make(map[string]any, len(updates))
+	for k, v := range updates {
+		copied[k] = v
+	}
+	r.updateExtra = append(r.updateExtra, copied)
+	return nil
+}
+
+func (r *openAICodexSnapshotAsyncRepo) SetRateLimited(_ context.Context, _ int64, resetAt time.Time) error {
+	if r.rateLimitCh != nil {
+		r.rateLimitCh <- resetAt
+	}
+	return nil
+}
+
+func (r *openAICodexSnapshotAsyncRepo) UpdateExtra(_ context.Context, _ int64, updates map[string]any) error {
+	if r.updateExtraCh != nil {
+		copied := make(map[string]any, len(updates))
+		for k, v := range updates {
+			copied[k] = v
+		}
+		r.updateExtraCh <- copied
+	}
+	return nil
+}
+
+func TestOpenAIGatewayService_Forward_WSv2ErrorEventUsageLimitPersistsRateLimit(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	resetAt := time.Now().Add(2 * time.Hour).Unix()
+	upgrader := websocket.Upgrader{CheckOrigin: func(r *http.Request) bool { return true }}
+	wsServer := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		conn, err := upgrader.Upgrade(w, r, nil)
+		if err != nil {
+			t.Errorf("upgrade websocket failed: %v", err)
+			return
+		}
+		defer func() { _ = conn.Close() }()
+
+		var req map[string]any
+		if err := conn.ReadJSON(&req); err != nil {
+			t.Errorf("read ws request failed: %v", err)
+			return
+		}
+		_ = conn.WriteJSON(map[string]any{
+			"type": "error",
+			"error": map[string]any{
+				"code":      "rate_limit_exceeded",
+				"type":      "usage_limit_reached",
+				"message":   "The usage limit has been reached",
+				"resets_at": resetAt,
+			},
+		})
+	}))
+	defer wsServer.Close()
+
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/openai/v1/responses", nil)
+	c.Request.Header.Set("User-Agent", "unit-test-agent/1.0")
+
+	upstream := &httpUpstreamRecorder{
+		resp: &http.Response{
+			StatusCode: http.StatusOK,
+			Header:     http.Header{"Content-Type": []string{"application/json"}},
+			Body:       io.NopCloser(strings.NewReader(`{"id":"resp_http_should_not_run"}`)),
+		},
+	}
+
+	cfg := newOpenAIWSV2TestConfig()
+	cfg.Security.URLAllowlist.Enabled = false
+	cfg.Security.URLAllowlist.AllowInsecureHTTP = true
+
+	account := Account{
+		ID:          501,
+		Name:        "openai-ws-rate-limit-event",
+		Platform:    PlatformOpenAI,
+		Type:        AccountTypeAPIKey,
+		Status:      StatusActive,
+		Schedulable: true,
+		Concurrency: 1,
+		Credentials: map[string]any{
+			"api_key":  "sk-test",
+			"base_url": wsServer.URL,
+		},
+		Extra: map[string]any{
+			"responses_websockets_v2_enabled": true,
+		},
+	}
+	repo := &openAIWSRateLimitSignalRepo{stubOpenAIAccountRepo: stubOpenAIAccountRepo{accounts: []Account{account}}}
+	rateSvc := &RateLimitService{accountRepo: repo}
+	svc := &OpenAIGatewayService{
+		accountRepo:      repo,
+		rateLimitService: rateSvc,
+		httpUpstream:     upstream,
+		cache:            &stubGatewayCache{},
+		cfg:              cfg,
+		openaiWSResolver: NewOpenAIWSProtocolResolver(cfg),
+		toolCorrector:    NewCodexToolCorrector(),
+	}
+
+	body := []byte(`{"model":"gpt-5.1","stream":false,"input":[{"type":"input_text","text":"hello"}]}`)
+	result, err := svc.Forward(context.Background(), c, &account, body)
+	require.Error(t, err)
+	require.Nil(t, result)
+	require.Equal(t, http.StatusTooManyRequests, rec.Code)
+	require.Nil(t, upstream.lastReq, "WS 限流 error event 不应回退到同账号 HTTP")
+	require.Len(t, repo.rateLimitCalls, 1)
+	require.WithinDuration(t, time.Unix(resetAt, 0), repo.rateLimitCalls[0], 2*time.Second)
+}
+
+func TestOpenAIGatewayService_Forward_WSv2Handshake429PersistsRateLimit(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("x-codex-primary-used-percent", "100")
+		w.Header().Set("x-codex-primary-reset-after-seconds", "7200")
+		w.Header().Set("x-codex-primary-window-minutes", "10080")
+		w.Header().Set("x-codex-secondary-used-percent", "3")
+		w.Header().Set("x-codex-secondary-reset-after-seconds", "1800")
+		w.Header().Set("x-codex-secondary-window-minutes", "300")
+		w.WriteHeader(http.StatusTooManyRequests)
+		_, _ = w.Write([]byte(`{"error":{"type":"rate_limit_exceeded","message":"rate limited"}}`))
+	}))
+	defer server.Close()
+
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/openai/v1/responses", nil)
+	c.Request.Header.Set("User-Agent", "unit-test-agent/1.0")
+
+	upstream := &httpUpstreamRecorder{
+		resp: &http.Response{
+			StatusCode: http.StatusOK,
+			Header:     http.Header{"Content-Type": []string{"application/json"}},
+			Body:       io.NopCloser(strings.NewReader(`{"id":"resp_http_should_not_run"}`)),
+		},
+	}
+
+	cfg := newOpenAIWSV2TestConfig()
+	cfg.Security.URLAllowlist.Enabled = false
+	cfg.Security.URLAllowlist.AllowInsecureHTTP = true
+
+	account := Account{
+		ID:          502,
+		Name:        "openai-ws-rate-limit-handshake",
+		Platform:    PlatformOpenAI,
+		Type:        AccountTypeAPIKey,
+		Status:      StatusActive,
+		Schedulable: true,
+		Concurrency: 1,
+		Credentials: map[string]any{
+			"api_key":  "sk-test",
+			"base_url": server.URL,
+		},
+		Extra: map[string]any{
+			"responses_websockets_v2_enabled": true,
+		},
+	}
+	repo := &openAIWSRateLimitSignalRepo{stubOpenAIAccountRepo: stubOpenAIAccountRepo{accounts: []Account{account}}}
+	rateSvc := &RateLimitService{accountRepo: repo}
+	svc := &OpenAIGatewayService{
+		accountRepo:      repo,
+		rateLimitService: rateSvc,
+		httpUpstream:     upstream,
+		cache:            &stubGatewayCache{},
+		cfg:              cfg,
+		openaiWSResolver: NewOpenAIWSProtocolResolver(cfg),
+		toolCorrector:    NewCodexToolCorrector(),
+	}
+
+	body := []byte(`{"model":"gpt-5.1","stream":false,"input":[{"type":"input_text","text":"hello"}]}`)
+	result, err := svc.Forward(context.Background(), c, &account, body)
+	require.Error(t, err)
+	require.Nil(t, result)
+	require.Equal(t, http.StatusTooManyRequests, rec.Code)
+	require.Nil(t, upstream.lastReq, "WS 握手 429 不应回退到同账号 HTTP")
+	require.Len(t, repo.rateLimitCalls, 1)
+	require.NotEmpty(t, repo.updateExtra, "握手 429 的 x-codex 头应立即落库")
+	require.Contains(t, repo.updateExtra[0], "codex_usage_updated_at")
+}
+
+func TestOpenAIGatewayService_ProxyResponsesWebSocketFromClient_ErrorEventUsageLimitPersistsRateLimit(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	cfg := newOpenAIWSV2TestConfig()
+	cfg.Security.URLAllowlist.Enabled = false
+	cfg.Security.URLAllowlist.AllowInsecureHTTP = true
+	cfg.Gateway.OpenAIWS.MaxConnsPerAccount = 1
+	cfg.Gateway.OpenAIWS.MinIdlePerAccount = 0
+	cfg.Gateway.OpenAIWS.MaxIdlePerAccount = 1
+	cfg.Gateway.OpenAIWS.QueueLimitPerConn = 8
+	cfg.Gateway.OpenAIWS.DialTimeoutSeconds = 3
+	cfg.Gateway.OpenAIWS.ReadTimeoutSeconds = 3
+	cfg.Gateway.OpenAIWS.WriteTimeoutSeconds = 3
+
+	resetAt := time.Now().Add(90 * time.Minute).Unix()
+	captureConn := &openAIWSCaptureConn{
+		events: [][]byte{
+			[]byte(`{"type":"error","error":{"code":"rate_limit_exceeded","type":"usage_limit_reached","message":"The usage limit has been reached","resets_at":PLACEHOLDER}}`),
+		},
+	}
+	captureConn.events[0] = []byte(strings.ReplaceAll(string(captureConn.events[0]), "PLACEHOLDER", strconv.FormatInt(resetAt, 10)))
+	captureDialer := &openAIWSCaptureDialer{conn: captureConn}
+	pool := newOpenAIWSConnPool(cfg)
+	pool.setClientDialerForTest(captureDialer)
+
+	account := Account{
+		ID:          503,
+		Name:        "openai-ingress-rate-limit",
+		Platform:    PlatformOpenAI,
+		Type:        AccountTypeAPIKey,
+		Status:      StatusActive,
+		Schedulable: true,
+		Concurrency: 1,
+		Credentials: map[string]any{
+			"api_key": "sk-test",
+		},
+		Extra: map[string]any{
+			"responses_websockets_v2_enabled": true,
+		},
+	}
+	repo := &openAIWSRateLimitSignalRepo{stubOpenAIAccountRepo: stubOpenAIAccountRepo{accounts: []Account{account}}}
+	rateSvc := &RateLimitService{accountRepo: repo}
+	svc := &OpenAIGatewayService{
+		accountRepo:      repo,
+		rateLimitService: rateSvc,
+		httpUpstream:     &httpUpstreamRecorder{},
+		cache:            &stubGatewayCache{},
+		cfg:              cfg,
+		openaiWSResolver: NewOpenAIWSProtocolResolver(cfg),
+		toolCorrector:    NewCodexToolCorrector(),
+		openaiWSPool:     pool,
+	}
+
+	serverErrCh := make(chan error, 1)
+	wsServer := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		conn, err := coderws.Accept(w, r, &coderws.AcceptOptions{CompressionMode: coderws.CompressionContextTakeover})
+		if err != nil {
+			serverErrCh <- err
+			return
+		}
+		defer func() { _ = conn.CloseNow() }()
+
+		rec := httptest.NewRecorder()
+		ginCtx, _ := gin.CreateTestContext(rec)
+		req := r.Clone(r.Context())
+		req.Header = req.Header.Clone()
+		req.Header.Set("User-Agent", "unit-test-agent/1.0")
+		ginCtx.Request = req
+
+		readCtx, cancel := context.WithTimeout(r.Context(), 3*time.Second)
+		msgType, firstMessage, readErr := conn.Read(readCtx)
+		cancel()
+		if readErr != nil {
+			serverErrCh <- readErr
+			return
+		}
+		if msgType != coderws.MessageText && msgType != coderws.MessageBinary {
+			serverErrCh <- io.ErrUnexpectedEOF
+			return
+		}
+
+		serverErrCh <- svc.ProxyResponsesWebSocketFromClient(r.Context(), ginCtx, conn, &account, "sk-test", firstMessage, nil)
+	}))
+	defer wsServer.Close()
+
+	dialCtx, cancelDial := context.WithTimeout(context.Background(), 3*time.Second)
+	clientConn, _, err := coderws.Dial(dialCtx, "ws"+strings.TrimPrefix(wsServer.URL, "http"), nil)
+	cancelDial()
+	require.NoError(t, err)
+	defer func() { _ = clientConn.CloseNow() }()
+
+	writeCtx, cancelWrite := context.WithTimeout(context.Background(), 3*time.Second)
+	err = clientConn.Write(writeCtx, coderws.MessageText, []byte(`{"type":"response.create","model":"gpt-5.1","stream":false}`))
+	cancelWrite()
+	require.NoError(t, err)
+
+	select {
+	case serverErr := <-serverErrCh:
+		require.Error(t, serverErr)
+		require.Len(t, repo.rateLimitCalls, 1)
+		require.WithinDuration(t, time.Unix(resetAt, 0), repo.rateLimitCalls[0], 2*time.Second)
+	case <-time.After(5 * time.Second):
+		t.Fatal("等待 ingress websocket 结束超时")
+	}
+}
+
+func TestOpenAIGatewayService_UpdateCodexUsageSnapshot_ExhaustedSnapshotSetsRateLimit(t *testing.T) {
+	repo := &openAICodexSnapshotAsyncRepo{
+		updateExtraCh: make(chan map[string]any, 1),
+		rateLimitCh:   make(chan time.Time, 1),
+	}
+	svc := &OpenAIGatewayService{accountRepo: repo}
+	snapshot := &OpenAICodexUsageSnapshot{
+		PrimaryUsedPercent:         ptrFloat64WS(100),
+		PrimaryResetAfterSeconds:   ptrIntWS(3600),
+		PrimaryWindowMinutes:       ptrIntWS(10080),
+		SecondaryUsedPercent:       ptrFloat64WS(12),
+		SecondaryResetAfterSeconds: ptrIntWS(1200),
+		SecondaryWindowMinutes:     ptrIntWS(300),
+	}
+	before := time.Now()
+	svc.updateCodexUsageSnapshot(context.Background(), 601, snapshot)
+
+	select {
+	case updates := <-repo.updateExtraCh:
+		require.Equal(t, 100.0, updates["codex_7d_used_percent"])
+	case <-time.After(2 * time.Second):
+		t.Fatal("等待 codex 快照落库超时")
+	}
+
+	select {
+	case resetAt := <-repo.rateLimitCh:
+		require.WithinDuration(t, before.Add(time.Hour), resetAt, 2*time.Second)
+	case <-time.After(2 * time.Second):
+		t.Fatal("等待 codex 100% 自动切换限流超时")
+	}
+}
+
+func TestOpenAIGatewayService_UpdateCodexUsageSnapshot_NonExhaustedSnapshotDoesNotSetRateLimit(t *testing.T) {
+	repo := &openAICodexSnapshotAsyncRepo{
+		updateExtraCh: make(chan map[string]any, 1),
+		rateLimitCh:   make(chan time.Time, 1),
+	}
+	svc := &OpenAIGatewayService{accountRepo: repo}
+	snapshot := &OpenAICodexUsageSnapshot{
+		PrimaryUsedPercent:         ptrFloat64WS(94),
+		PrimaryResetAfterSeconds:   ptrIntWS(3600),
+		PrimaryWindowMinutes:       ptrIntWS(10080),
+		SecondaryUsedPercent:       ptrFloat64WS(22),
+		SecondaryResetAfterSeconds: ptrIntWS(1200),
+		SecondaryWindowMinutes:     ptrIntWS(300),
+	}
+	svc.updateCodexUsageSnapshot(context.Background(), 602, snapshot)
+
+	select {
+	case <-repo.updateExtraCh:
+	case <-time.After(2 * time.Second):
+		t.Fatal("等待 codex 快照落库超时")
+	}
+
+	select {
+	case resetAt := <-repo.rateLimitCh:
+		t.Fatalf("unexpected rate limit reset at: %v", resetAt)
+	case <-time.After(200 * time.Millisecond):
+	}
+}
+
+func ptrFloat64WS(v float64) *float64 { return &v }
+func ptrIntWS(v int) *int             { return &v }
+
+func TestOpenAIWSErrorHTTPStatusFromRaw_UsageLimitReachedIs429(t *testing.T) {
+	require.Equal(t, http.StatusTooManyRequests, openAIWSErrorHTTPStatusFromRaw("", "usage_limit_reached"))
+	require.Equal(t, http.StatusTooManyRequests, openAIWSErrorHTTPStatusFromRaw("rate_limit_exceeded", ""))
+}

From 1307d604e7d4ab7c9922166678b618a4dffe5ef3 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E7=A5=9E=E4=B9=90?= <6682635@qq.com>
Date: Sun, 8 Mar 2026 00:14:15 +0800
Subject: [PATCH 5/6] =?UTF-8?q?fix:=20=E8=A1=A5=E9=BD=90=E6=97=A7=E8=B4=A6?=
 =?UTF-8?q?=E5=8F=B7=E7=9A=84=20OpenAI=20=E9=99=90=E6=B5=81=E8=A1=A5?=
 =?UTF-8?q?=E5=81=BF?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 .../internal/service/account_usage_service.go |  1 +
 backend/internal/service/admin_service.go     |  4 +
 .../service/openai_gateway_service.go         | 53 +++++++++++-
 .../openai_ws_ratelimit_signal_test.go        | 85 +++++++++++++++++++
 4 files changed, 141 insertions(+), 2 deletions(-)

diff --git a/backend/internal/service/account_usage_service.go b/backend/internal/service/account_usage_service.go
index 9bb3aa0b..7c001118 100644
--- a/backend/internal/service/account_usage_service.go
+++ b/backend/internal/service/account_usage_service.go
@@ -359,6 +359,7 @@ func (s *AccountUsageService) getOpenAIUsage(ctx context.Context, account *Accou
 	if account == nil {
 		return usage, nil
 	}
+	syncOpenAICodexRateLimitFromExtra(ctx, s.accountRepo, account, now)
 
 	if progress := buildCodexUsageProgressFromExtra(account.Extra, "5h", now); progress != nil {
 		usage.FiveHour = progress
diff --git a/backend/internal/service/admin_service.go b/backend/internal/service/admin_service.go
index 680268e0..ab00932c 100644
--- a/backend/internal/service/admin_service.go
+++ b/backend/internal/service/admin_service.go
@@ -1349,6 +1349,10 @@ func (s *adminServiceImpl) ListAccounts(ctx context.Context, page, pageSize int,
 	if err != nil {
 		return nil, 0, err
 	}
+	now := time.Now()
+	for i := range accounts {
+		syncOpenAICodexRateLimitFromExtra(ctx, s.accountRepo, &accounts[i], now)
+	}
 	return accounts, result.Total, nil
 }
 
diff --git a/backend/internal/service/openai_gateway_service.go b/backend/internal/service/openai_gateway_service.go
index 40a4e377..709ee808 100644
--- a/backend/internal/service/openai_gateway_service.go
+++ b/backend/internal/service/openai_gateway_service.go
@@ -1447,10 +1447,20 @@ func (s *OpenAIGatewayService) resolveFreshSchedulableOpenAIAccount(ctx context.
 }
 
 func (s *OpenAIGatewayService) getSchedulableAccount(ctx context.Context, accountID int64) (*Account, error) {
+	var (
+		account *Account
+		err     error
+	)
 	if s.schedulerSnapshot != nil {
-		return s.schedulerSnapshot.GetAccount(ctx, accountID)
+		account, err = s.schedulerSnapshot.GetAccount(ctx, accountID)
+	} else {
+		account, err = s.accountRepo.GetByID(ctx, accountID)
 	}
-	return s.accountRepo.GetByID(ctx, accountID)
+	if err != nil || account == nil {
+		return account, err
+	}
+	syncOpenAICodexRateLimitFromExtra(ctx, s.accountRepo, account, time.Now())
+	return account, nil
 }
 
 func (s *OpenAIGatewayService) schedulingConfig() config.GatewaySchedulingConfig {
@@ -3923,6 +3933,45 @@ func codexRateLimitResetAtFromSnapshot(snapshot *OpenAICodexUsageSnapshot, fallb
 	return nil
 }
 
+func codexRateLimitResetAtFromExtra(extra map[string]any, now time.Time) *time.Time {
+	if len(extra) == 0 {
+		return nil
+	}
+	if progress := buildCodexUsageProgressFromExtra(extra, "7d", now); progress != nil && codexUsagePercentExhausted(&progress.Utilization) && progress.ResetsAt != nil && now.Before(*progress.ResetsAt) {
+		resetAt := progress.ResetsAt.UTC()
+		return &resetAt
+	}
+	if progress := buildCodexUsageProgressFromExtra(extra, "5h", now); progress != nil && codexUsagePercentExhausted(&progress.Utilization) && progress.ResetsAt != nil && now.Before(*progress.ResetsAt) {
+		resetAt := progress.ResetsAt.UTC()
+		return &resetAt
+	}
+	return nil
+}
+
+func applyOpenAICodexRateLimitFromExtra(account *Account, now time.Time) (*time.Time, bool) {
+	if account == nil || !account.IsOpenAI() {
+		return nil, false
+	}
+	resetAt := codexRateLimitResetAtFromExtra(account.Extra, now)
+	if resetAt == nil {
+		return nil, false
+	}
+	if account.RateLimitResetAt != nil && now.Before(*account.RateLimitResetAt) && !account.RateLimitResetAt.Before(*resetAt) {
+		return account.RateLimitResetAt, false
+	}
+	account.RateLimitResetAt = resetAt
+	return resetAt, true
+}
+
+func syncOpenAICodexRateLimitFromExtra(ctx context.Context, repo AccountRepository, account *Account, now time.Time) *time.Time {
+	resetAt, changed := applyOpenAICodexRateLimitFromExtra(account, now)
+	if !changed || resetAt == nil || repo == nil || account == nil || account.ID <= 0 {
+		return resetAt
+	}
+	_ = repo.SetRateLimited(ctx, account.ID, *resetAt)
+	return resetAt
+}
+
 // updateCodexUsageSnapshot saves the Codex usage snapshot to account's Extra field
 func (s *OpenAIGatewayService) updateCodexUsageSnapshot(ctx context.Context, accountID int64, snapshot *OpenAICodexUsageSnapshot) {
 	if snapshot == nil {
diff --git a/backend/internal/service/openai_ws_ratelimit_signal_test.go b/backend/internal/service/openai_ws_ratelimit_signal_test.go
index a6b6e874..28cb8e00 100644
--- a/backend/internal/service/openai_ws_ratelimit_signal_test.go
+++ b/backend/internal/service/openai_ws_ratelimit_signal_test.go
@@ -10,6 +10,7 @@ import (
 	"testing"
 	"time"
 
+	"github.com/Wei-Shaw/sub2api/internal/pkg/pagination"
 	coderws "github.com/coder/websocket"
 	"github.com/gin-gonic/gin"
 	"github.com/gorilla/websocket"
@@ -28,6 +29,11 @@ type openAICodexSnapshotAsyncRepo struct {
 	rateLimitCh   chan time.Time
 }
 
+type openAICodexExtraListRepo struct {
+	stubOpenAIAccountRepo
+	rateLimitCh chan time.Time
+}
+
 func (r *openAIWSRateLimitSignalRepo) SetRateLimited(_ context.Context, _ int64, resetAt time.Time) error {
 	r.rateLimitCalls = append(r.rateLimitCalls, resetAt)
 	return nil
@@ -60,6 +66,22 @@ func (r *openAICodexSnapshotAsyncRepo) UpdateExtra(_ context.Context, _ int64, u
 	return nil
 }
 
+func (r *openAICodexExtraListRepo) SetRateLimited(_ context.Context, _ int64, resetAt time.Time) error {
+	if r.rateLimitCh != nil {
+		r.rateLimitCh <- resetAt
+	}
+	return nil
+}
+
+func (r *openAICodexExtraListRepo) ListWithFilters(_ context.Context, params pagination.PaginationParams, platform, accountType, status, search string, groupID int64) ([]Account, *pagination.PaginationResult, error) {
+	_ = platform
+	_ = accountType
+	_ = status
+	_ = search
+	_ = groupID
+	return r.accounts, &pagination.PaginationResult{Total: int64(len(r.accounts)), Page: params.Page, PageSize: params.PageSize}, nil
+}
+
 func TestOpenAIGatewayService_Forward_WSv2ErrorEventUsageLimitPersistsRateLimit(t *testing.T) {
 	gin.SetMode(gin.TestMode)
 
@@ -386,6 +408,69 @@ func TestOpenAIGatewayService_UpdateCodexUsageSnapshot_NonExhaustedSnapshotDoesN
 func ptrFloat64WS(v float64) *float64 { return &v }
 func ptrIntWS(v int) *int             { return &v }
 
+func TestOpenAIGatewayService_GetSchedulableAccount_ExhaustedCodexExtraSetsRateLimit(t *testing.T) {
+	resetAt := time.Now().Add(6 * 24 * time.Hour)
+	account := Account{
+		ID:          701,
+		Platform:    PlatformOpenAI,
+		Type:        AccountTypeOAuth,
+		Status:      StatusActive,
+		Schedulable: true,
+		Concurrency: 1,
+		Extra: map[string]any{
+			"codex_7d_used_percent": 100.0,
+			"codex_7d_reset_at":     resetAt.UTC().Format(time.RFC3339),
+		},
+	}
+	repo := &openAICodexExtraListRepo{stubOpenAIAccountRepo: stubOpenAIAccountRepo{accounts: []Account{account}}, rateLimitCh: make(chan time.Time, 1)}
+	svc := &OpenAIGatewayService{accountRepo: repo}
+
+	fresh, err := svc.getSchedulableAccount(context.Background(), account.ID)
+	require.NoError(t, err)
+	require.NotNil(t, fresh)
+	require.NotNil(t, fresh.RateLimitResetAt)
+	require.WithinDuration(t, resetAt.UTC(), *fresh.RateLimitResetAt, time.Second)
+	select {
+	case persisted := <-repo.rateLimitCh:
+		require.WithinDuration(t, resetAt.UTC(), persisted, time.Second)
+	case <-time.After(2 * time.Second):
+		t.Fatal("等待旧快照补写限流状态超时")
+	}
+}
+
+func TestAdminService_ListAccounts_ExhaustedCodexExtraReturnsRateLimitedAccount(t *testing.T) {
+	resetAt := time.Now().Add(4 * 24 * time.Hour)
+	repo := &openAICodexExtraListRepo{
+		stubOpenAIAccountRepo: stubOpenAIAccountRepo{accounts: []Account{{
+			ID:          702,
+			Platform:    PlatformOpenAI,
+			Type:        AccountTypeOAuth,
+			Status:      StatusActive,
+			Schedulable: true,
+			Concurrency: 1,
+			Extra: map[string]any{
+				"codex_7d_used_percent": 100.0,
+				"codex_7d_reset_at":     resetAt.UTC().Format(time.RFC3339),
+			},
+		}}},
+		rateLimitCh: make(chan time.Time, 1),
+	}
+	svc := &adminServiceImpl{accountRepo: repo}
+
+	accounts, total, err := svc.ListAccounts(context.Background(), 1, 20, PlatformOpenAI, AccountTypeOAuth, "", "", 0)
+	require.NoError(t, err)
+	require.Equal(t, int64(1), total)
+	require.Len(t, accounts, 1)
+	require.NotNil(t, accounts[0].RateLimitResetAt)
+	require.WithinDuration(t, resetAt.UTC(), *accounts[0].RateLimitResetAt, time.Second)
+	select {
+	case persisted := <-repo.rateLimitCh:
+		require.WithinDuration(t, resetAt.UTC(), persisted, time.Second)
+	case <-time.After(2 * time.Second):
+		t.Fatal("等待列表补写限流状态超时")
+	}
+}
+
 func TestOpenAIWSErrorHTTPStatusFromRaw_UsageLimitReachedIs429(t *testing.T) {
 	require.Equal(t, http.StatusTooManyRequests, openAIWSErrorHTTPStatusFromRaw("", "usage_limit_reached"))
 	require.Equal(t, http.StatusTooManyRequests, openAIWSErrorHTTPStatusFromRaw("rate_limit_exceeded", ""))

From 9301dae63e08304aa37ca9e43e9ff836a0486d56 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E7=A5=9E=E4=B9=90?= <6682635@qq.com>
Date: Sun, 8 Mar 2026 04:37:03 +0800
Subject: [PATCH 6/6] =?UTF-8?q?fix:=20=E4=BF=AE=E5=A4=8D=20OpenAI=20WS=20?=
 =?UTF-8?q?=E7=94=A8=E9=87=8F=E5=88=B7=E6=96=B0=E9=81=97=E6=BC=8F=E5=9C=BA?=
 =?UTF-8?q?=E6=99=AF?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 .../internal/service/account_test_service.go  | 17 +++--
 .../account_test_service_openai_test.go       | 36 +++++++++++
 .../components/account/AccountUsageCell.vue   | 16 ++++-
 .../__tests__/AccountUsageCell.spec.ts        | 62 +++++++++++++++++++
 4 files changed, 124 insertions(+), 7 deletions(-)

diff --git a/backend/internal/service/account_test_service.go b/backend/internal/service/account_test_service.go
index 5b22c645..b44f29fd 100644
--- a/backend/internal/service/account_test_service.go
+++ b/backend/internal/service/account_test_service.go
@@ -406,13 +406,22 @@ func (s *AccountTestService) testOpenAIAccountConnection(c *gin.Context, account
 	}
 	defer func() { _ = resp.Body.Close() }()
 
+	if isOAuth && s.accountRepo != nil {
+		if updates, err := extractOpenAICodexProbeUpdates(resp); err == nil && len(updates) > 0 {
+			_ = s.accountRepo.UpdateExtra(ctx, account.ID, updates)
+			mergeAccountExtra(account, updates)
+		}
+		if snapshot := ParseCodexRateLimitHeaders(resp.Header); snapshot != nil {
+			if resetAt := codexRateLimitResetAtFromSnapshot(snapshot, time.Now()); resetAt != nil {
+				_ = s.accountRepo.SetRateLimited(ctx, account.ID, *resetAt)
+				account.RateLimitResetAt = resetAt
+			}
+		}
+	}
+
 	if resp.StatusCode != http.StatusOK {
 		body, _ := io.ReadAll(resp.Body)
 		if isOAuth && s.accountRepo != nil {
-			if updates, err := extractOpenAICodexProbeUpdates(resp); err == nil && len(updates) > 0 {
-				_ = s.accountRepo.UpdateExtra(ctx, account.ID, updates)
-				mergeAccountExtra(account, updates)
-			}
 			if resetAt := (&RateLimitService{}).calculateOpenAI429ResetTime(resp.Header); resetAt != nil {
 				_ = s.accountRepo.SetRateLimited(ctx, account.ID, *resetAt)
 				account.RateLimitResetAt = resetAt
diff --git a/backend/internal/service/account_test_service_openai_test.go b/backend/internal/service/account_test_service_openai_test.go
index 61a755a7..efa6f7da 100644
--- a/backend/internal/service/account_test_service_openai_test.go
+++ b/backend/internal/service/account_test_service_openai_test.go
@@ -4,7 +4,9 @@ package service
 
 import (
 	"context"
+	"io"
 	"net/http"
+	"strings"
 	"testing"
 	"time"
 
@@ -30,6 +32,40 @@ func (r *openAIAccountTestRepo) SetRateLimited(_ context.Context, id int64, rese
 	return nil
 }
 
+func TestAccountTestService_OpenAISuccessPersistsSnapshotFromHeaders(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	ctx, recorder := newSoraTestContext()
+
+	resp := newJSONResponse(http.StatusOK, "")
+	resp.Body = io.NopCloser(strings.NewReader(`data: {"type":"response.completed"}
+
+`))
+	resp.Header.Set("x-codex-primary-used-percent", "88")
+	resp.Header.Set("x-codex-primary-reset-after-seconds", "604800")
+	resp.Header.Set("x-codex-primary-window-minutes", "10080")
+	resp.Header.Set("x-codex-secondary-used-percent", "42")
+	resp.Header.Set("x-codex-secondary-reset-after-seconds", "18000")
+	resp.Header.Set("x-codex-secondary-window-minutes", "300")
+
+	repo := &openAIAccountTestRepo{}
+	upstream := &queuedHTTPUpstream{responses: []*http.Response{resp}}
+	svc := &AccountTestService{accountRepo: repo, httpUpstream: upstream}
+	account := &Account{
+		ID:          89,
+		Platform:    PlatformOpenAI,
+		Type:        AccountTypeOAuth,
+		Concurrency: 1,
+		Credentials: map[string]any{"access_token": "test-token"},
+	}
+
+	err := svc.testOpenAIAccountConnection(ctx, account, "gpt-5.4")
+	require.NoError(t, err)
+	require.NotEmpty(t, repo.updatedExtra)
+	require.Equal(t, 42.0, repo.updatedExtra["codex_5h_used_percent"])
+	require.Equal(t, 88.0, repo.updatedExtra["codex_7d_used_percent"])
+	require.Contains(t, recorder.Body.String(), "test_complete")
+}
+
 func TestAccountTestService_OpenAI429PersistsSnapshotAndRateLimit(t *testing.T) {
 	gin.SetMode(gin.TestMode)
 	ctx, _ := newSoraTestContext()
diff --git a/frontend/src/components/account/AccountUsageCell.vue b/frontend/src/components/account/AccountUsageCell.vue
index 44c8e209..0026920d 100644
--- a/frontend/src/components/account/AccountUsageCell.vue
+++ b/frontend/src/components/account/AccountUsageCell.vue
@@ -412,14 +412,24 @@ const isActiveOpenAIRateLimited = computed(() => {
 })
 
 const preferFetchedOpenAIUsage = computed(() => {
-  return isActiveOpenAIRateLimited.value && hasOpenAIUsageFallback.value
+  return (isActiveOpenAIRateLimited.value || isOpenAICodexSnapshotStale.value) && hasOpenAIUsageFallback.value
 })
 
 const openAIUsageRefreshKey = computed(() => buildOpenAIUsageRefreshKey(props.account))
 
+const isOpenAICodexSnapshotStale = computed(() => {
+  if (props.account.platform !== 'openai' || props.account.type !== 'oauth') return false
+  const extra = props.account.extra as Record<string, unknown> | undefined
+  const updatedAtRaw = extra?.codex_usage_updated_at
+  if (!updatedAtRaw) return true
+  const updatedAt = Date.parse(String(updatedAtRaw))
+  if (Number.isNaN(updatedAt)) return true
+  return Date.now() - updatedAt >= 10 * 60 * 1000
+})
+
 const shouldAutoLoadUsageOnMount = computed(() => {
   if (props.account.platform === 'openai' && props.account.type === 'oauth') {
-    return isActiveOpenAIRateLimited.value || !hasCodexUsage.value
+    return isActiveOpenAIRateLimited.value || !hasCodexUsage.value || isOpenAICodexSnapshotStale.value
   }
   return shouldFetchUsage.value
 })
@@ -807,7 +817,7 @@ onMounted(() => {
 watch(openAIUsageRefreshKey, (nextKey, prevKey) => {
   if (!prevKey || nextKey === prevKey) return
   if (props.account.platform !== 'openai' || props.account.type !== 'oauth') return
-  if (!isActiveOpenAIRateLimited.value && hasCodexUsage.value) return
+  if (!isActiveOpenAIRateLimited.value && hasCodexUsage.value && !isOpenAICodexSnapshotStale.value) return
 
   loadUsage().catch((e) => {
     console.error('Failed to refresh OpenAI usage:', e)
diff --git a/frontend/src/components/account/__tests__/AccountUsageCell.spec.ts b/frontend/src/components/account/__tests__/AccountUsageCell.spec.ts
index 400eafbb..2681f0cb 100644
--- a/frontend/src/components/account/__tests__/AccountUsageCell.spec.ts
+++ b/frontend/src/components/account/__tests__/AccountUsageCell.spec.ts
@@ -69,6 +69,67 @@ describe('AccountUsageCell', () => {
   })
 
 
+  it('OpenAI OAuth 快照已过期时首屏会重新请求 usage', async () => {
+    getUsage.mockResolvedValue({
+      five_hour: {
+        utilization: 15,
+        resets_at: '2026-03-08T12:00:00Z',
+        remaining_seconds: 3600,
+        window_stats: {
+          requests: 3,
+          tokens: 300,
+          cost: 0.03,
+          standard_cost: 0.03,
+          user_cost: 0.03
+        }
+      },
+      seven_day: {
+        utilization: 77,
+        resets_at: '2026-03-13T12:00:00Z',
+        remaining_seconds: 3600,
+        window_stats: {
+          requests: 3,
+          tokens: 300,
+          cost: 0.03,
+          standard_cost: 0.03,
+          user_cost: 0.03
+        }
+      }
+    })
+
+    const wrapper = mount(AccountUsageCell, {
+      props: {
+        account: {
+          id: 2000,
+          platform: 'openai',
+          type: 'oauth',
+          extra: {
+            codex_usage_updated_at: '2026-03-07T00:00:00Z',
+            codex_5h_used_percent: 12,
+            codex_5h_reset_at: '2026-03-08T12:00:00Z',
+            codex_7d_used_percent: 34,
+            codex_7d_reset_at: '2026-03-13T12:00:00Z'
+          }
+        } as any
+      },
+      global: {
+        stubs: {
+          UsageProgressBar: {
+            props: ['label', 'utilization', 'resetsAt', 'windowStats', 'color'],
+            template: '<div class="usage-bar">{{ label }}|{{ utilization }}|{{ windowStats?.tokens }}</div>'
+          },
+          AccountQuotaInfo: true
+        }
+      }
+    })
+
+    await flushPromises()
+
+    expect(getUsage).toHaveBeenCalledWith(2000)
+    expect(wrapper.text()).toContain('5h|15|300')
+    expect(wrapper.text()).toContain('7d|77|300')
+  })
+
   it('OpenAI OAuth 有现成快照且未限额时不会首屏请求 usage', async () => {
     const wrapper = mount(AccountUsageCell, {
       props: {
@@ -77,6 +138,7 @@ describe('AccountUsageCell', () => {
           platform: 'openai',
           type: 'oauth',
           extra: {
+            codex_usage_updated_at: '2099-03-07T10:00:00Z',
             codex_5h_used_percent: 12,
             codex_5h_reset_at: '2099-03-07T12:00:00Z',
             codex_7d_used_percent: 34,