From b252821c59d399bc067b60f669c242669fa3012b Mon Sep 17 00:00:00 2001
From: wucm667 <stevenwucongmin@gmail.com>
Date: Sat, 19 Sep 2026 17:22:39 +0800
Subject: [PATCH] fix(accounts): preserve refresh errors during usage queries

---
 .../internal/service/account_usage_service.go |  8 +-
 .../account_usage_service_batch_test.go       | 81 +++++++++++++++++++
 2 files changed, 84 insertions(+), 5 deletions(-)

diff --git a/backend/internal/service/account_usage_service.go b/backend/internal/service/account_usage_service.go
index f49d3aaa71..3347aa328d 100644
--- a/backend/internal/service/account_usage_service.go
+++ b/backend/internal/service/account_usage_service.go
@@ -359,11 +359,9 @@ func (s *AccountUsageService) getUsageForAccount(ctx context.Context, account *A
 	}
 
 	if account.Platform == PlatformOpenAI && account.Type == AccountTypeOAuth {
-		usage, err := s.getOpenAIUsage(ctx, account, forceProbe)
-		if err == nil {
-			s.tryClearRecoverableAccountError(ctx, account)
-		}
-		return usage, err
+		// Usage can come from a stored snapshot even when a probe fails. Neither
+		// that nor a working access token proves a rejected refresh token recovered.
+		return s.getOpenAIUsage(ctx, account, forceProbe)
 	}
 
 	if account.Platform == PlatformGemini {
diff --git a/backend/internal/service/account_usage_service_batch_test.go b/backend/internal/service/account_usage_service_batch_test.go
index 067bf0ddd1..b4eb32ed27 100644
--- a/backend/internal/service/account_usage_service_batch_test.go
+++ b/backend/internal/service/account_usage_service_batch_test.go
@@ -189,3 +189,84 @@ func TestAccountUsageService_GetUsageBatch_BestEffortByAccount(t *testing.T) {
 		t.Fatalf("expected API key account error to be preserved, got %q", errorsByAccount[7003])
 	}
 }
+
+// Model ClearError's persisted effect so this regression catches both the write
+// and mutations of the account returned by the repository.
+type usageErrorAccountRepo struct {
+	stubOpenAIAccountRepo
+	clearCalls int
+}
+
+func (r *usageErrorAccountRepo) ClearError(ctx context.Context, id int64) error {
+	r.clearCalls++
+	account, err := r.GetByID(ctx, id)
+	if err != nil {
+		return err
+	}
+	account.Status = StatusActive
+	account.ErrorMessage = ""
+	return nil
+}
+
+func TestAccountUsageService_OpenAIQueriesPreserveRefreshError(t *testing.T) {
+	for _, scenario := range []string{"cached_expired_access", "cached_valid_access_invalid_refresh", "failed_probe_missing_access", "probe_throttled"} {
+		for _, query := range []string{"usage", "usage_batch", "forced_usage_batch", "today", "today_batch"} {
+			t.Run(scenario+"/"+query, func(t *testing.T) {
+				const message = "Token refresh failed (non-retryable): refresh_token_invalidated"
+				account := Account{
+					ID: 7358, Platform: PlatformOpenAI, Type: AccountTypeOAuth,
+					Status: StatusError, ErrorMessage: message,
+					Credentials: map[string]any{"refresh_token": "invalid-refresh-token"},
+				}
+				cache := NewUsageCache()
+				if strings.HasPrefix(scenario, "cached_") {
+					account.Credentials["access_token"] = "cached-access-token"
+					expiresAt := time.Now().Add(-time.Hour)
+					if scenario == "cached_valid_access_invalid_refresh" {
+						expiresAt = time.Now().Add(time.Hour)
+					}
+					account.Credentials["expires_at"] = expiresAt.Format(time.RFC3339)
+					account.Extra = map[string]any{"codex_5h_used_percent": 18.0, "codex_7d_used_percent": 34.0}
+				}
+				if scenario == "probe_throttled" {
+					cache.openAIProbeCache.Store(account.ID, time.Now())
+				}
+				// Forced requests take the probe path, but fail locally without
+				// credentials; no external network is needed for this regression.
+				if query == "forced_usage_batch" {
+					delete(account.Credentials, "access_token")
+				}
+				repo := &usageErrorAccountRepo{stubOpenAIAccountRepo: stubOpenAIAccountRepo{accounts: []Account{account}}}
+				svc := &AccountUsageService{accountRepo: repo, usageLogRepo: &usageBatchLogRepoStub{}, cache: cache}
+				ctx := context.Background()
+				switch query {
+				case "usage":
+					usage, err := svc.GetUsage(ctx, account.ID)
+					if err != nil || usage == nil {
+						t.Fatalf("GetUsage: usage=%v err=%v", usage, err)
+					}
+				case "usage_batch", "forced_usage_batch":
+					usage, failures, err := svc.GetUsageBatch(ctx, []int64{account.ID}, query == "forced_usage_batch")
+					if err != nil || len(failures) != 0 || usage[account.ID] == nil {
+						t.Fatalf("GetUsageBatch: usage=%v failures=%v err=%v", usage, failures, err)
+					}
+				case "today":
+					if _, err := svc.GetTodayStats(ctx, account.ID); err != nil {
+						t.Fatal(err)
+					}
+				case "today_batch":
+					if _, err := svc.GetTodayStatsBatch(ctx, []int64{account.ID}); err != nil {
+						t.Fatal(err)
+					}
+				}
+				stored, err := repo.GetByID(ctx, account.ID)
+				if err != nil {
+					t.Fatal(err)
+				}
+				if repo.clearCalls != 0 || stored.Status != StatusError || stored.ErrorMessage != message {
+					t.Fatalf("query erased refresh error: clearCalls=%d status=%q error=%q", repo.clearCalls, stored.Status, stored.ErrorMessage)
+				}
+			})
+		}
+	}
+}