为 Grok 普通 429 增加有限同号重试

backend/internal/service/grok_upstream_failure.go

@@ -402,6 +402,13 @@ func grokRetryableOnSameAccount(account *Account, statusCode int, responseBody [
 		if statusCode == http.StatusTooManyRequests {
 			return true
 		}
+	case GrokFailureRateLimit:
+		// A transient 429 does not identify a bad credential. Give every Grok
+		// account a bounded same-account retry window before failover; the
+		// failover loop still caps attempts and the client receives 429 after it.
+		if statusCode == http.StatusTooManyRequests {
+			return true
+		}
 	}
 	return account.IsPoolMode() && account.IsPoolModeRetryableStatus(statusCode)
 }
@@ -411,7 +418,7 @@ func grokSameAccountRetryMetadata(account *Account, statusCode int, responseBody
 		return false, 0, time.Time{}
 	}
 	decision := classifyGrokUpstreamFailure(statusCode, responseBody, "")
-	if decision.Class != GrokFailureModelCapacity {
+	if decision.Class != GrokFailureModelCapacity && decision.Class != GrokFailureRateLimit {
 		return true, 0, time.Time{}
 	}
 	return true, 500 * time.Millisecond, time.Now().Add(30 * time.Second)

backend/internal/service/grok_upstream_failure_test.go

@@ -81,7 +81,7 @@ func TestGrokRetryableOnSameAccount_CapacityAndRateLimit(t *testing.T) {
 	account := &Account{ID: 9105, Platform: PlatformGrok, Type: AccountTypeOAuth}
 	require.True(t, grokRetryableOnSameAccount(account, http.StatusTooManyRequests,
 		[]byte(`{"error":{"message":"The model is currently at capacity due to high demand"}}`)))
-	require.False(t, grokRetryableOnSameAccount(account, http.StatusTooManyRequests,
+	require.True(t, grokRetryableOnSameAccount(account, http.StatusTooManyRequests,
 		[]byte(`{"error":{"message":"rate limit exceeded"}}`)))
 	require.False(t, grokRetryableOnSameAccount(account, http.StatusPaymentRequired,
 		[]byte(`{"error":{"message":"You have run out of credits or need a Grok subscription"}}`)))
@@ -118,9 +118,9 @@ func TestGrokSameAccountRetryMetadata_CapacityDeadline(t *testing.T) {
 
 	retryable, delay, deadline = grokSameAccountRetryMetadata(account, http.StatusTooManyRequests,
 		[]byte(`{"error":{"message":"rate limit exceeded"}}`))
-	require.False(t, retryable)
-	require.Zero(t, delay)
-	require.True(t, deadline.IsZero())
+	require.True(t, retryable)
+	require.Equal(t, 500*time.Millisecond, delay)
+	require.WithinDuration(t, time.Now().Add(30*time.Second), deadline, 2*time.Second)
 }
 
 func TestClassifyGrokUpstreamFailure_ValidationNoCool(t *testing.T) {