feat(antigravity): 添加 URL fallback 机制 (sandbox → daily → prod)

backend/internal/pkg/antigravity/client.go

@@ -5,8 +5,11 @@ import (
 	"bytes"
 	"context"
 	"encoding/json"
+	"errors"
 	"fmt"
 	"io"
+	"log"
+	"net"
 	"net/http"
 	"net/url"
 	"strings"
@@ -22,10 +25,10 @@ func resolveHost(urlStr string) string {
 	return parsed.Host
 }
 
-// NewAPIRequest 创建 Antigravity API 请求（v1internal 端点）
-func NewAPIRequest(ctx context.Context, action, accessToken string, body []byte) (*http.Request, error) {
+// NewAPIRequestWithURL 使用指定的 base URL 创建 Antigravity API 请求（v1internal 端点）
+func NewAPIRequestWithURL(ctx context.Context, baseURL, action, accessToken string, body []byte) (*http.Request, error) {
 	// 构建 URL，流式请求添加 ?alt=sse 参数
-	apiURL := fmt.Sprintf("%s/v1internal:%s", BaseURL, action)
+	apiURL := fmt.Sprintf("%s/v1internal:%s", baseURL, action)
 	isStream := action == "streamGenerateContent"
 	if isStream {
 		apiURL += "?alt=sse"
@@ -53,11 +56,15 @@ func NewAPIRequest(ctx context.Context, action, accessToken string, body []byte)
 		req.Host = host
 	}
 
-	// 注意：requestType 已在 JSON body 的 V1InternalRequest 中设置，不需要 HTTP Header
-
 	return req, nil
 }
 
+// NewAPIRequest 使用默认 URL 创建 Antigravity API 请求（v1internal 端点）
+// 向后兼容：仅使用默认 BaseURL
+func NewAPIRequest(ctx context.Context, action, accessToken string, body []byte) (*http.Request, error) {
+	return NewAPIRequestWithURL(ctx, BaseURL, action, accessToken, body)
+}
+
 // TokenResponse Google OAuth token 响应
 type TokenResponse struct {
 	AccessToken  string `json:"access_token"`
@@ -164,6 +171,38 @@ func NewClient(proxyURL string) *Client {
 	}
 }
 
+// isConnectionError 判断是否为连接错误（网络超时、DNS 失败、连接拒绝）
+func isConnectionError(err error) bool {
+	if err == nil {
+		return false
+	}
+
+	// 检查超时错误
+	var netErr net.Error
+	if errors.As(err, &netErr) && netErr.Timeout() {
+		return true
+	}
+
+	// 检查连接错误（DNS 失败、连接拒绝）
+	var opErr *net.OpError
+	if errors.As(err, &opErr) {
+		return true
+	}
+
+	// 检查 URL 错误
+	var urlErr *url.Error
+	return errors.As(err, &urlErr)
+}
+
+// shouldFallbackToNextURL 判断是否应切换到下一个 URL
+// 仅连接错误和 HTTP 429 触发 URL 降级
+func shouldFallbackToNextURL(err error, statusCode int) bool {
+	if isConnectionError(err) {
+		return true
+	}
+	return statusCode == http.StatusTooManyRequests
+}
+
 // ExchangeCode 用 authorization code 交换 token
 func (c *Client) ExchangeCode(ctx context.Context, code, codeVerifier string) (*TokenResponse, error) {
 	params := url.Values{}
@@ -272,6 +311,7 @@ func (c *Client) GetUserInfo(ctx context.Context, accessToken string) (*UserInfo
 }
 
 // LoadCodeAssist 获取账户信息，返回解析后的结构体和原始 JSON
+// 支持 URL fallback：sandbox → daily → prod
 func (c *Client) LoadCodeAssist(ctx context.Context, accessToken string) (*LoadCodeAssistResponse, map[string]any, error) {
 	reqBody := LoadCodeAssistRequest{}
 	reqBody.Metadata.IDEType = "ANTIGRAVITY"
@@ -281,40 +321,65 @@ func (c *Client) LoadCodeAssist(ctx context.Context, accessToken string) (*LoadC
 		return nil, nil, fmt.Errorf("序列化请求失败: %w", err)
 	}
 
-	url := BaseURL + "/v1internal:loadCodeAssist"
-	req, err := http.NewRequestWithContext(ctx, http.MethodPost, url, strings.NewReader(string(bodyBytes)))
-	if err != nil {
-		return nil, nil, fmt.Errorf("创建请求失败: %w", err)
+	// 获取可用的 URL 列表
+	availableURLs := DefaultURLAvailability.GetAvailableURLs()
+	if len(availableURLs) == 0 {
+		availableURLs = BaseURLs // 所有 URL 都不可用时，重试所有
 	}
-	req.Header.Set("Authorization", "Bearer "+accessToken)
-	req.Header.Set("Content-Type", "application/json")
-	req.Header.Set("User-Agent", UserAgent)
 
-	resp, err := c.httpClient.Do(req)
-	if err != nil {
-		return nil, nil, fmt.Errorf("loadCodeAssist 请求失败: %w", err)
-	}
-	defer func() { _ = resp.Body.Close() }()
+	var lastErr error
+	for urlIdx, baseURL := range availableURLs {
+		apiURL := baseURL + "/v1internal:loadCodeAssist"
+		req, err := http.NewRequestWithContext(ctx, http.MethodPost, apiURL, strings.NewReader(string(bodyBytes)))
+		if err != nil {
+			lastErr = fmt.Errorf("创建请求失败: %w", err)
+			continue
+		}
+		req.Header.Set("Authorization", "Bearer "+accessToken)
+		req.Header.Set("Content-Type", "application/json")
+		req.Header.Set("User-Agent", UserAgent)
+
+		resp, err := c.httpClient.Do(req)
+		if err != nil {
+			lastErr = fmt.Errorf("loadCodeAssist 请求失败: %w", err)
+			if shouldFallbackToNextURL(err, 0) && urlIdx < len(availableURLs)-1 {
+				DefaultURLAvailability.MarkUnavailable(baseURL)
+				log.Printf("[antigravity] loadCodeAssist URL fallback: %s -> %s", baseURL, availableURLs[urlIdx+1])
+				continue
+			}
+			return nil, nil, lastErr
+		}
 
-	respBodyBytes, err := io.ReadAll(resp.Body)
-	if err != nil {
-		return nil, nil, fmt.Errorf("读取响应失败: %w", err)
-	}
+		respBodyBytes, err := io.ReadAll(resp.Body)
+		_ = resp.Body.Close() // 立即关闭，避免循环内 defer 导致的资源泄漏
+		if err != nil {
+			return nil, nil, fmt.Errorf("读取响应失败: %w", err)
+		}
 
-	if resp.StatusCode != http.StatusOK {
-		return nil, nil, fmt.Errorf("loadCodeAssist 失败 (HTTP %d): %s", resp.StatusCode, string(respBodyBytes))
-	}
+		// 检查是否需要 URL 降级
+		if shouldFallbackToNextURL(nil, resp.StatusCode) && urlIdx < len(availableURLs)-1 {
+			DefaultURLAvailability.MarkUnavailable(baseURL)
+			log.Printf("[antigravity] loadCodeAssist URL fallback (HTTP %d): %s -> %s", resp.StatusCode, baseURL, availableURLs[urlIdx+1])
+			continue
+		}
 
-	var loadResp LoadCodeAssistResponse
-	if err := json.Unmarshal(respBodyBytes, &loadResp); err != nil {
-		return nil, nil, fmt.Errorf("响应解析失败: %w", err)
-	}
+		if resp.StatusCode != http.StatusOK {
+			return nil, nil, fmt.Errorf("loadCodeAssist 失败 (HTTP %d): %s", resp.StatusCode, string(respBodyBytes))
+		}
 
-	// 解析原始 JSON 为 map
-	var rawResp map[string]any
-	_ = json.Unmarshal(respBodyBytes, &rawResp)
+		var loadResp LoadCodeAssistResponse
+		if err := json.Unmarshal(respBodyBytes, &loadResp); err != nil {
+			return nil, nil, fmt.Errorf("响应解析失败: %w", err)
+		}
+
+		// 解析原始 JSON 为 map
+		var rawResp map[string]any
+		_ = json.Unmarshal(respBodyBytes, &rawResp)
+
+		return &loadResp, rawResp, nil
+	}
 
-	return &loadResp, rawResp, nil
+	return nil, nil, lastErr
 }
 
 // ModelQuotaInfo 模型配额信息
@@ -339,6 +404,7 @@ type FetchAvailableModelsResponse struct {
 }
 
 // FetchAvailableModels 获取可用模型和配额信息，返回解析后的结构体和原始 JSON
+// 支持 URL fallback：sandbox → daily → prod
 func (c *Client) FetchAvailableModels(ctx context.Context, accessToken, projectID string) (*FetchAvailableModelsResponse, map[string]any, error) {
 	reqBody := FetchAvailableModelsRequest{Project: projectID}
 	bodyBytes, err := json.Marshal(reqBody)
@@ -346,38 +412,63 @@ func (c *Client) FetchAvailableModels(ctx context.Context, accessToken, projectI
 		return nil, nil, fmt.Errorf("序列化请求失败: %w", err)
 	}
 
-	apiURL := BaseURL + "/v1internal:fetchAvailableModels"
-	req, err := http.NewRequestWithContext(ctx, http.MethodPost, apiURL, strings.NewReader(string(bodyBytes)))
-	if err != nil {
-		return nil, nil, fmt.Errorf("创建请求失败: %w", err)
+	// 获取可用的 URL 列表
+	availableURLs := DefaultURLAvailability.GetAvailableURLs()
+	if len(availableURLs) == 0 {
+		availableURLs = BaseURLs // 所有 URL 都不可用时，重试所有
 	}
-	req.Header.Set("Authorization", "Bearer "+accessToken)
-	req.Header.Set("Content-Type", "application/json")
-	req.Header.Set("User-Agent", UserAgent)
 
-	resp, err := c.httpClient.Do(req)
-	if err != nil {
-		return nil, nil, fmt.Errorf("fetchAvailableModels 请求失败: %w", err)
-	}
-	defer func() { _ = resp.Body.Close() }()
+	var lastErr error
+	for urlIdx, baseURL := range availableURLs {
+		apiURL := baseURL + "/v1internal:fetchAvailableModels"
+		req, err := http.NewRequestWithContext(ctx, http.MethodPost, apiURL, strings.NewReader(string(bodyBytes)))
+		if err != nil {
+			lastErr = fmt.Errorf("创建请求失败: %w", err)
+			continue
+		}
+		req.Header.Set("Authorization", "Bearer "+accessToken)
+		req.Header.Set("Content-Type", "application/json")
+		req.Header.Set("User-Agent", UserAgent)
+
+		resp, err := c.httpClient.Do(req)
+		if err != nil {
+			lastErr = fmt.Errorf("fetchAvailableModels 请求失败: %w", err)
+			if shouldFallbackToNextURL(err, 0) && urlIdx < len(availableURLs)-1 {
+				DefaultURLAvailability.MarkUnavailable(baseURL)
+				log.Printf("[antigravity] fetchAvailableModels URL fallback: %s -> %s", baseURL, availableURLs[urlIdx+1])
+				continue
+			}
+			return nil, nil, lastErr
+		}
 
-	respBodyBytes, err := io.ReadAll(resp.Body)
-	if err != nil {
-		return nil, nil, fmt.Errorf("读取响应失败: %w", err)
-	}
+		respBodyBytes, err := io.ReadAll(resp.Body)
+		_ = resp.Body.Close() // 立即关闭，避免循环内 defer 导致的资源泄漏
+		if err != nil {
+			return nil, nil, fmt.Errorf("读取响应失败: %w", err)
+		}
 
-	if resp.StatusCode != http.StatusOK {
-		return nil, nil, fmt.Errorf("fetchAvailableModels 失败 (HTTP %d): %s", resp.StatusCode, string(respBodyBytes))
-	}
+		// 检查是否需要 URL 降级
+		if shouldFallbackToNextURL(nil, resp.StatusCode) && urlIdx < len(availableURLs)-1 {
+			DefaultURLAvailability.MarkUnavailable(baseURL)
+			log.Printf("[antigravity] fetchAvailableModels URL fallback (HTTP %d): %s -> %s", resp.StatusCode, baseURL, availableURLs[urlIdx+1])
+			continue
+		}
 
-	var modelsResp FetchAvailableModelsResponse
-	if err := json.Unmarshal(respBodyBytes, &modelsResp); err != nil {
-		return nil, nil, fmt.Errorf("响应解析失败: %w", err)
-	}
+		if resp.StatusCode != http.StatusOK {
+			return nil, nil, fmt.Errorf("fetchAvailableModels 失败 (HTTP %d): %s", resp.StatusCode, string(respBodyBytes))
+		}
 
-	// 解析原始 JSON 为 map
-	var rawResp map[string]any
-	_ = json.Unmarshal(respBodyBytes, &rawResp)
+		var modelsResp FetchAvailableModelsResponse
+		if err := json.Unmarshal(respBodyBytes, &modelsResp); err != nil {
+			return nil, nil, fmt.Errorf("响应解析失败: %w", err)
+		}
+
+		// 解析原始 JSON 为 map
+		var rawResp map[string]any
+		_ = json.Unmarshal(respBodyBytes, &rawResp)
+
+		return &modelsResp, rawResp, nil
+	}
 
-	return &modelsResp, rawResp, nil
+	return nil, nil, lastErr
 }

backend/internal/pkg/antigravity/oauth.go

@@ -32,17 +32,79 @@ const (
 		"https://www.googleapis.com/auth/cclog " +
 		"https://www.googleapis.com/auth/experimentsandconfigs"
 
-	// API 端点
-	// 优先使用 sandbox daily URL，配额更宽松
-	BaseURL = "https://daily-cloudcode-pa.sandbox.googleapis.com"
-
 	// User-Agent（模拟官方客户端）
 	UserAgent = "antigravity/1.104.0 darwin/arm64"
 
 	// Session 过期时间
 	SessionTTL = 30 * time.Minute
+
+	// URL 可用性 TTL（不可用 URL 的恢复时间）
+	URLAvailabilityTTL = 5 * time.Minute
 )
 
+// BaseURLs 定义 Antigravity API 端点，按优先级排序
+// fallback 顺序: sandbox → daily → prod
+var BaseURLs = []string{
+	"https://daily-cloudcode-pa.sandbox.googleapis.com", // sandbox
+	"https://daily-cloudcode-pa.googleapis.com",         // daily
+	"https://cloudcode-pa.googleapis.com",               // prod
+}
+
+// BaseURL 默认 URL（保持向后兼容）
+var BaseURL = BaseURLs[0]
+
+// URLAvailability 管理 URL 可用性状态（带 TTL 自动恢复）
+type URLAvailability struct {
+	mu          sync.RWMutex
+	unavailable map[string]time.Time // URL -> 恢复时间
+	ttl         time.Duration
+}
+
+// DefaultURLAvailability 全局 URL 可用性管理器
+var DefaultURLAvailability = NewURLAvailability(URLAvailabilityTTL)
+
+// NewURLAvailability 创建 URL 可用性管理器
+func NewURLAvailability(ttl time.Duration) *URLAvailability {
+	return &URLAvailability{
+		unavailable: make(map[string]time.Time),
+		ttl:         ttl,
+	}
+}
+
+// MarkUnavailable 标记 URL 临时不可用
+func (u *URLAvailability) MarkUnavailable(url string) {
+	u.mu.Lock()
+	defer u.mu.Unlock()
+	u.unavailable[url] = time.Now().Add(u.ttl)
+}
+
+// IsAvailable 检查 URL 是否可用
+func (u *URLAvailability) IsAvailable(url string) bool {
+	u.mu.RLock()
+	defer u.mu.RUnlock()
+	expiry, exists := u.unavailable[url]
+	if !exists {
+		return true
+	}
+	return time.Now().After(expiry)
+}
+
+// GetAvailableURLs 返回可用的 URL 列表（保持优先级顺序）
+func (u *URLAvailability) GetAvailableURLs() []string {
+	u.mu.RLock()
+	defer u.mu.RUnlock()
+
+	now := time.Now()
+	result := make([]string, 0, len(BaseURLs))
+	for _, url := range BaseURLs {
+		expiry, exists := u.unavailable[url]
+		if !exists || now.After(expiry) {
+			result = append(result, url)
+		}
+	}
+	return result
+}
+
 // OAuthSession 保存 OAuth 授权流程的临时状态
 type OAuthSession struct {
 	State        string    `json:"state"`

backend/internal/service/antigravity_gateway_service.go

@@ -10,6 +10,7 @@ import (
 	"io"
 	"log"
 	mathrand "math/rand"
+	"net"
 	"net/http"
 	"strings"
 	"sync/atomic"
@@ -27,6 +28,32 @@ const (
 	antigravityRetryMaxDelay    = 16 * time.Second
 )
 
+// isAntigravityConnectionError 判断是否为连接错误（网络超时、DNS 失败、连接拒绝）
+func isAntigravityConnectionError(err error) bool {
+	if err == nil {
+		return false
+	}
+
+	// 检查超时错误
+	var netErr net.Error
+	if errors.As(err, &netErr) && netErr.Timeout() {
+		return true
+	}
+
+	// 检查连接错误（DNS 失败、连接拒绝）
+	var opErr *net.OpError
+	return errors.As(err, &opErr)
+}
+
+// shouldAntigravityFallbackToNextURL 判断是否应切换到下一个 URL
+// 仅连接错误和 HTTP 429 触发 URL 降级
+func shouldAntigravityFallbackToNextURL(err error, statusCode int) bool {
+	if isAntigravityConnectionError(err) {
+		return true
+	}
+	return statusCode == http.StatusTooManyRequests
+}
+
 // getSessionID 从 gin.Context 获取 session_id（用于日志追踪）
 func getSessionID(c *gin.Context) string {
 	if c == nil {
@@ -181,45 +208,70 @@ func (s *AntigravityGatewayService) TestConnection(ctx context.Context, account
 		return nil, fmt.Errorf("构建请求失败: %w", err)
 	}
 
-	// 构建 HTTP 请求（总是使用流式 endpoint，与官方客户端一致）
-	req, err := antigravity.NewAPIRequest(ctx, "streamGenerateContent", accessToken, requestBody)
-	if err != nil {
-		return nil, err
-	}
-
-	// 调试日志：Test 请求信息
-	log.Printf("[antigravity-Test] account=%s request_size=%d url=%s", account.Name, len(requestBody), req.URL.String())
-
 	// 代理 URL
 	proxyURL := ""
 	if account.ProxyID != nil && account.Proxy != nil {
 		proxyURL = account.Proxy.URL()
 	}
 
-	// 发送请求
-	resp, err := s.httpUpstream.Do(req, proxyURL, account.ID, account.Concurrency)
-	if err != nil {
-		return nil, fmt.Errorf("请求失败: %w", err)
+	// URL fallback 循环
+	availableURLs := antigravity.DefaultURLAvailability.GetAvailableURLs()
+	if len(availableURLs) == 0 {
+		availableURLs = antigravity.BaseURLs // 所有 URL 都不可用时，重试所有
 	}
-	defer func() { _ = resp.Body.Close() }()
 
-	// 读取响应
-	respBody, err := io.ReadAll(io.LimitReader(resp.Body, 2<<20))
-	if err != nil {
-		return nil, fmt.Errorf("读取响应失败: %w", err)
-	}
+	var lastErr error
+	for urlIdx, baseURL := range availableURLs {
+		// 构建 HTTP 请求（总是使用流式 endpoint，与官方客户端一致）
+		req, err := antigravity.NewAPIRequestWithURL(ctx, baseURL, "streamGenerateContent", accessToken, requestBody)
+		if err != nil {
+			lastErr = err
+			continue
+		}
 
-	if resp.StatusCode >= 400 {
-		return nil, fmt.Errorf("API 返回 %d: %s", resp.StatusCode, string(respBody))
-	}
+		// 调试日志：Test 请求信息
+		log.Printf("[antigravity-Test] account=%s request_size=%d url=%s", account.Name, len(requestBody), req.URL.String())
 
-	// 解析流式响应，提取文本
-	text := extractTextFromSSEResponse(respBody)
+		// 发送请求
+		resp, err := s.httpUpstream.Do(req, proxyURL, account.ID, account.Concurrency)
+		if err != nil {
+			lastErr = fmt.Errorf("请求失败: %w", err)
+			if shouldAntigravityFallbackToNextURL(err, 0) && urlIdx < len(availableURLs)-1 {
+				antigravity.DefaultURLAvailability.MarkUnavailable(baseURL)
+				log.Printf("[antigravity-Test] URL fallback: %s -> %s", baseURL, availableURLs[urlIdx+1])
+				continue
+			}
+			return nil, lastErr
+		}
 
-	return &TestConnectionResult{
-		Text:        text,
-		MappedModel: mappedModel,
-	}, nil
+		// 读取响应
+		respBody, err := io.ReadAll(io.LimitReader(resp.Body, 2<<20))
+		_ = resp.Body.Close() // 立即关闭，避免循环内 defer 导致的资源泄漏
+		if err != nil {
+			return nil, fmt.Errorf("读取响应失败: %w", err)
+		}
+
+		// 检查是否需要 URL 降级
+		if shouldAntigravityFallbackToNextURL(nil, resp.StatusCode) && urlIdx < len(availableURLs)-1 {
+			antigravity.DefaultURLAvailability.MarkUnavailable(baseURL)
+			log.Printf("[antigravity-Test] URL fallback (HTTP %d): %s -> %s", resp.StatusCode, baseURL, availableURLs[urlIdx+1])
+			continue
+		}
+
+		if resp.StatusCode >= 400 {
+			return nil, fmt.Errorf("API 返回 %d: %s", resp.StatusCode, string(respBody))
+		}
+
+		// 解析流式响应，提取文本
+		text := extractTextFromSSEResponse(respBody)
+
+		return &TestConnectionResult{
+			Text:        text,
+			MappedModel: mappedModel,
+		}, nil
+	}
+
+	return nil, lastErr
 }
 
 // buildGeminiTestRequest 构建 Gemini 格式测试请求
@@ -484,62 +536,86 @@ func (s *AntigravityGatewayService) Forward(ctx context.Context, c *gin.Context,
 	// 如果客户端请求非流式，在响应处理阶段会收集完整流式响应后转换返回
 	action := "streamGenerateContent"
 
+	// URL fallback 循环
+	availableURLs := antigravity.DefaultURLAvailability.GetAvailableURLs()
+	if len(availableURLs) == 0 {
+		availableURLs = antigravity.BaseURLs // 所有 URL 都不可用时，重试所有
+	}
+
 	// 重试循环
 	var resp *http.Response
-	for attempt := 1; attempt <= antigravityMaxRetries; attempt++ {
-		// 检查 context 是否已取消（客户端断开连接）
-		select {
-		case <-ctx.Done():
-			log.Printf("%s status=context_canceled error=%v", prefix, ctx.Err())
-			return nil, ctx.Err()
-		default:
-		}
+urlFallbackLoop:
+	for urlIdx, baseURL := range availableURLs {
+		for attempt := 1; attempt <= antigravityMaxRetries; attempt++ {
+			// 检查 context 是否已取消（客户端断开连接）
+			select {
+			case <-ctx.Done():
+				log.Printf("%s status=context_canceled error=%v", prefix, ctx.Err())
+				return nil, ctx.Err()
+			default:
+			}
 
-		upstreamReq, err := antigravity.NewAPIRequest(ctx, action, accessToken, geminiBody)
-		if err != nil {
-			return nil, err
-		}
+			upstreamReq, err := antigravity.NewAPIRequestWithURL(ctx, baseURL, action, accessToken, geminiBody)
+			if err != nil {
+				return nil, err
+			}
 
-		resp, err = s.httpUpstream.Do(upstreamReq, proxyURL, account.ID, account.Concurrency)
-		if err != nil {
-			if attempt < antigravityMaxRetries {
-				log.Printf("%s status=request_failed retry=%d/%d error=%v", prefix, attempt, antigravityMaxRetries, err)
-				if !sleepAntigravityBackoffWithContext(ctx, attempt) {
-					log.Printf("%s status=context_canceled_during_backoff", prefix)
-					return nil, ctx.Err()
+			resp, err = s.httpUpstream.Do(upstreamReq, proxyURL, account.ID, account.Concurrency)
+			if err != nil {
+				// 检查是否应触发 URL 降级
+				if shouldAntigravityFallbackToNextURL(err, 0) && urlIdx < len(availableURLs)-1 {
+					antigravity.DefaultURLAvailability.MarkUnavailable(baseURL)
+					log.Printf("%s URL fallback (connection error): %s -> %s", prefix, baseURL, availableURLs[urlIdx+1])
+					continue urlFallbackLoop
 				}
-				continue
+				if attempt < antigravityMaxRetries {
+					log.Printf("%s status=request_failed retry=%d/%d error=%v", prefix, attempt, antigravityMaxRetries, err)
+					if !sleepAntigravityBackoffWithContext(ctx, attempt) {
+						log.Printf("%s status=context_canceled_during_backoff", prefix)
+						return nil, ctx.Err()
+					}
+					continue
+				}
+				log.Printf("%s status=request_failed retries_exhausted error=%v", prefix, err)
+				return nil, s.writeClaudeError(c, http.StatusBadGateway, "upstream_error", "Upstream request failed after retries")
 			}
-			log.Printf("%s status=request_failed retries_exhausted error=%v", prefix, err)
-			return nil, s.writeClaudeError(c, http.StatusBadGateway, "upstream_error", "Upstream request failed after retries")
-		}
 
-		if resp.StatusCode >= 400 && s.shouldRetryUpstreamError(resp.StatusCode) {
-			respBody, _ := io.ReadAll(io.LimitReader(resp.Body, 2<<20))
-			_ = resp.Body.Close()
+			// 检查是否应触发 URL 降级（仅 429）
+			if resp.StatusCode == http.StatusTooManyRequests && urlIdx < len(availableURLs)-1 {
+				respBody, _ := io.ReadAll(io.LimitReader(resp.Body, 2<<20))
+				_ = resp.Body.Close()
+				antigravity.DefaultURLAvailability.MarkUnavailable(baseURL)
+				log.Printf("%s URL fallback (HTTP 429): %s -> %s body=%s", prefix, baseURL, availableURLs[urlIdx+1], truncateForLog(respBody, 200))
+				continue urlFallbackLoop
+			}
+
+			if resp.StatusCode >= 400 && s.shouldRetryUpstreamError(resp.StatusCode) {
+				respBody, _ := io.ReadAll(io.LimitReader(resp.Body, 2<<20))
+				_ = resp.Body.Close()
 
-			if attempt < antigravityMaxRetries {
-				log.Printf("%s status=%d retry=%d/%d body=%s", prefix, resp.StatusCode, attempt, antigravityMaxRetries, truncateForLog(respBody, 500))
-				if !sleepAntigravityBackoffWithContext(ctx, attempt) {
-					log.Printf("%s status=context_canceled_during_backoff", prefix)
-					return nil, ctx.Err()
+				if attempt < antigravityMaxRetries {
+					log.Printf("%s status=%d retry=%d/%d body=%s", prefix, resp.StatusCode, attempt, antigravityMaxRetries, truncateForLog(respBody, 500))
+					if !sleepAntigravityBackoffWithContext(ctx, attempt) {
+						log.Printf("%s status=context_canceled_during_backoff", prefix)
+						return nil, ctx.Err()
+					}
+					continue
 				}
-				continue
-			}
-			// 所有重试都失败，标记限流状态
-			if resp.StatusCode == 429 {
-				s.handleUpstreamError(ctx, prefix, account, resp.StatusCode, resp.Header, respBody)
-			}
-			// 最后一次尝试也失败
-			resp = &http.Response{
-				StatusCode: resp.StatusCode,
-				Header:     resp.Header.Clone(),
-				Body:       io.NopCloser(bytes.NewReader(respBody)),
+				// 所有重试都失败，标记限流状态
+				if resp.StatusCode == 429 {
+					s.handleUpstreamError(ctx, prefix, account, resp.StatusCode, resp.Header, respBody)
+				}
+				// 最后一次尝试也失败
+				resp = &http.Response{
+					StatusCode: resp.StatusCode,
+					Header:     resp.Header.Clone(),
+					Body:       io.NopCloser(bytes.NewReader(respBody)),
+				}
+				break urlFallbackLoop
 			}
-			break
-		}
 
-		break
+			break urlFallbackLoop
+		}
 	}
 	defer func() { _ = resp.Body.Close() }()
 
@@ -1003,61 +1079,85 @@ func (s *AntigravityGatewayService) ForwardGemini(ctx context.Context, c *gin.Co
 	// 如果客户端请求非流式，在响应处理阶段会收集完整流式响应后返回
 	upstreamAction := "streamGenerateContent"
 
+	// URL fallback 循环
+	availableURLs := antigravity.DefaultURLAvailability.GetAvailableURLs()
+	if len(availableURLs) == 0 {
+		availableURLs = antigravity.BaseURLs // 所有 URL 都不可用时，重试所有
+	}
+
 	// 重试循环
 	var resp *http.Response
-	for attempt := 1; attempt <= antigravityMaxRetries; attempt++ {
-		// 检查 context 是否已取消（客户端断开连接）
-		select {
-		case <-ctx.Done():
-			log.Printf("%s status=context_canceled error=%v", prefix, ctx.Err())
-			return nil, ctx.Err()
-		default:
-		}
+urlFallbackLoop:
+	for urlIdx, baseURL := range availableURLs {
+		for attempt := 1; attempt <= antigravityMaxRetries; attempt++ {
+			// 检查 context 是否已取消（客户端断开连接）
+			select {
+			case <-ctx.Done():
+				log.Printf("%s status=context_canceled error=%v", prefix, ctx.Err())
+				return nil, ctx.Err()
+			default:
+			}
 
-		upstreamReq, err := antigravity.NewAPIRequest(ctx, upstreamAction, accessToken, wrappedBody)
-		if err != nil {
-			return nil, err
-		}
+			upstreamReq, err := antigravity.NewAPIRequestWithURL(ctx, baseURL, upstreamAction, accessToken, wrappedBody)
+			if err != nil {
+				return nil, err
+			}
 
-		resp, err = s.httpUpstream.Do(upstreamReq, proxyURL, account.ID, account.Concurrency)
-		if err != nil {
-			if attempt < antigravityMaxRetries {
-				log.Printf("%s status=request_failed retry=%d/%d error=%v", prefix, attempt, antigravityMaxRetries, err)
-				if !sleepAntigravityBackoffWithContext(ctx, attempt) {
-					log.Printf("%s status=context_canceled_during_backoff", prefix)
-					return nil, ctx.Err()
+			resp, err = s.httpUpstream.Do(upstreamReq, proxyURL, account.ID, account.Concurrency)
+			if err != nil {
+				// 检查是否应触发 URL 降级
+				if shouldAntigravityFallbackToNextURL(err, 0) && urlIdx < len(availableURLs)-1 {
+					antigravity.DefaultURLAvailability.MarkUnavailable(baseURL)
+					log.Printf("%s URL fallback (connection error): %s -> %s", prefix, baseURL, availableURLs[urlIdx+1])
+					continue urlFallbackLoop
 				}
-				continue
+				if attempt < antigravityMaxRetries {
+					log.Printf("%s status=request_failed retry=%d/%d error=%v", prefix, attempt, antigravityMaxRetries, err)
+					if !sleepAntigravityBackoffWithContext(ctx, attempt) {
+						log.Printf("%s status=context_canceled_during_backoff", prefix)
+						return nil, ctx.Err()
+					}
+					continue
+				}
+				log.Printf("%s status=request_failed retries_exhausted error=%v", prefix, err)
+				return nil, s.writeGoogleError(c, http.StatusBadGateway, "Upstream request failed after retries")
 			}
-			log.Printf("%s status=request_failed retries_exhausted error=%v", prefix, err)
-			return nil, s.writeGoogleError(c, http.StatusBadGateway, "Upstream request failed after retries")
-		}
 
-		if resp.StatusCode >= 400 && s.shouldRetryUpstreamError(resp.StatusCode) {
-			respBody, _ := io.ReadAll(io.LimitReader(resp.Body, 2<<20))
-			_ = resp.Body.Close()
+			// 检查是否应触发 URL 降级（仅 429）
+			if resp.StatusCode == http.StatusTooManyRequests && urlIdx < len(availableURLs)-1 {
+				respBody, _ := io.ReadAll(io.LimitReader(resp.Body, 2<<20))
+				_ = resp.Body.Close()
+				antigravity.DefaultURLAvailability.MarkUnavailable(baseURL)
+				log.Printf("%s URL fallback (HTTP 429): %s -> %s body=%s", prefix, baseURL, availableURLs[urlIdx+1], truncateForLog(respBody, 200))
+				continue urlFallbackLoop
+			}
+
+			if resp.StatusCode >= 400 && s.shouldRetryUpstreamError(resp.StatusCode) {
+				respBody, _ := io.ReadAll(io.LimitReader(resp.Body, 2<<20))
+				_ = resp.Body.Close()
 
-			if attempt < antigravityMaxRetries {
-				log.Printf("%s status=%d retry=%d/%d", prefix, resp.StatusCode, attempt, antigravityMaxRetries)
-				if !sleepAntigravityBackoffWithContext(ctx, attempt) {
-					log.Printf("%s status=context_canceled_during_backoff", prefix)
-					return nil, ctx.Err()
+				if attempt < antigravityMaxRetries {
+					log.Printf("%s status=%d retry=%d/%d", prefix, resp.StatusCode, attempt, antigravityMaxRetries)
+					if !sleepAntigravityBackoffWithContext(ctx, attempt) {
+						log.Printf("%s status=context_canceled_during_backoff", prefix)
+						return nil, ctx.Err()
+					}
+					continue
 				}
-				continue
-			}
-			// 所有重试都失败，标记限流状态
-			if resp.StatusCode == 429 {
-				s.handleUpstreamError(ctx, prefix, account, resp.StatusCode, resp.Header, respBody)
-			}
-			resp = &http.Response{
-				StatusCode: resp.StatusCode,
-				Header:     resp.Header.Clone(),
-				Body:       io.NopCloser(bytes.NewReader(respBody)),
+				// 所有重试都失败，标记限流状态
+				if resp.StatusCode == 429 {
+					s.handleUpstreamError(ctx, prefix, account, resp.StatusCode, resp.Header, respBody)
+				}
+				resp = &http.Response{
+					StatusCode: resp.StatusCode,
+					Header:     resp.Header.Clone(),
+					Body:       io.NopCloser(bytes.NewReader(respBody)),
+				}
+				break urlFallbackLoop
 			}
-			break
-		}
 
-		break
+			break urlFallbackLoop
+		}
 	}
 	defer func() {
 		if resp != nil && resp.Body != nil {