oadmin/sub2api
1
0
Fork 0
Code Issues Pull Requests 1 Actions 3 Packages Projects Releases Wiki Activity

feat: MODEL_CAPACITY_EXHAUSTED 使用固定1s间隔重试60次，不切换账号

Browse Source 
MODEL_CAPACITY_EXHAUSTED (503) 表示模型容量不足，所有账号共享同一容量池，
切换账号无意义。改为固定1s间隔重试最多60次，重试耗尽后直接返回上游错误。

- 新增 antigravityModelCapacityRetryMaxAttempts=60 和 antigravityModelCapacityRetryWait=1s
- shouldTriggerAntigravitySmartRetry 新增 isModelCapacityExhausted 返回值
- handleSmartRetry 对 MODEL_CAPACITY_EXHAUSTED 使用独立重试策略
- handleModelRateLimit 对 MODEL_CAPACITY_EXHAUSTED 仅标记 Handled，不设限流
- 重试耗尽后不设置模型限流、不清除粘性会话、不切换账号
This commit is contained in:
Edric Li
2026-02-10 02:03:06 +08:00
parent d6c2921f2b
commit 6114f69cca
3 changed files with 207 additions and 83 deletions
Download Patch File Download Diff File Expand all files Collapse all files

76
backend/internal/service/antigravity_rate_limit_test.go
View File

@@ -188,13 +188,14 @@ func TestHandleUpstreamError_429_NonModelRateLimit(t *testing.T) {
require.Equal(t, "claude-sonnet-4-5", repo.modelRateLimitCalls[0].modelKey)
}
// TestHandleUpstreamError_503_ModelRateLimit 测试 503 模型限流场景
func TestHandleUpstreamError_503_ModelRateLimit(t *testing.T) {
// TestHandleUpstreamError_503_ModelCapacityExhausted 测试 503 模型容量不足场景
// MODEL_CAPACITY_EXHAUSTED 时应等待重试，不切换账号
func TestHandleUpstreamError_503_ModelCapacityExhausted(t *testing.T) {
repo := &stubAntigravityAccountRepo{}
svc := &AntigravityGatewayService{accountRepo: repo}
account := &Account{ID: 3, Name: "acc-3", Platform: PlatformAntigravity}
// 503 + MODEL_CAPACITY_EXHAUSTED → 模型限流
// 503 + MODEL_CAPACITY_EXHAUSTED → 等待重试，不切换账号
body := []byte(`{
"error": {
"status": "UNAVAILABLE",
@@ -207,13 +208,13 @@ func TestHandleUpstreamError_503_ModelRateLimit(t *testing.T) {
result := svc.handleUpstreamError(context.Background(), "[test]", account, http.StatusServiceUnavailable, http.Header{}, body, "gemini-3-pro-high", 0, "", false)
// 应该触发模型限流
// MODEL_CAPACITY_EXHAUSTED 应该标记为已处理，不切换账号，不设置模型限流
// 实际重试由 handleSmartRetry 处理
require.NotNil(t, result)
require.True(t, result.Handled)
require.NotNil(t, result.SwitchError)
require.Equal(t, "gemini-3-pro-high", result.SwitchError.RateLimitedModel)
require.Len(t, repo.modelRateLimitCalls, 1)
require.Equal(t, "gemini-3-pro-high", repo.modelRateLimitCalls[0].modelKey)
require.False(t, result.ShouldRetry, "MODEL_CAPACITY_EXHAUSTED should not trigger retry from handleModelRateLimit path")
require.Nil(t, result.SwitchError, "MODEL_CAPACITY_EXHAUSTED should not trigger account switch")
require.Empty(t, repo.modelRateLimitCalls, "MODEL_CAPACITY_EXHAUSTED should not set model rate limit")
}
// TestHandleUpstreamError_503_NonModelRateLimit 测试 503 非模型限流场景（不处理）
@@ -301,11 +302,12 @@ func TestParseGeminiRateLimitResetTime_QuotaResetDelay_RoundsUp(t *testing.T) {
func TestParseAntigravitySmartRetryInfo(t *testing.T) {
tests := []struct {
name string
body string
expectedDelay time.Duration
expectedModel string
expectedNil bool
name string
body string
expectedDelay time.Duration
expectedModel string
expectedNil bool
expectedIsModelCapacityExhausted bool
}{
{
name: "valid complete response with RATE_LIMIT_EXCEEDED",
@@ -368,8 +370,9 @@ func TestParseAntigravitySmartRetryInfo(t *testing.T) {
"message": "No capacity available for model gemini-3-pro-high on the server"
}
}`,
expectedDelay: 39 * time.Second,
expectedModel: "gemini-3-pro-high",
expectedDelay: 39 * time.Second,
expectedModel: "gemini-3-pro-high",
expectedIsModelCapacityExhausted: true,
},
{
name: "503 UNAVAILABLE without MODEL_CAPACITY_EXHAUSTED - should return nil",
@@ -480,6 +483,9 @@ func TestParseAntigravitySmartRetryInfo(t *testing.T) {
if result.ModelName != tt.expectedModel {
t.Errorf("ModelName = %q, want %q", result.ModelName, tt.expectedModel)
}
if result.IsModelCapacityExhausted != tt.expectedIsModelCapacityExhausted {
t.Errorf("IsModelCapacityExhausted = %v, want %v", result.IsModelCapacityExhausted, tt.expectedIsModelCapacityExhausted)
}
})
}
}
@@ -491,13 +497,14 @@ func TestShouldTriggerAntigravitySmartRetry(t *testing.T) {
apiKeyAccount := &Account{Type: AccountTypeAPIKey}
tests := []struct {
name string
account *Account
body string
expectedShouldRetry bool
expectedShouldRateLimit bool
minWait time.Duration
modelName string
name string
account *Account
body string
expectedShouldRetry bool
expectedShouldRateLimit bool
expectedIsModelCapacityExhausted bool
minWait time.Duration
modelName string
}{
{
name: "OAuth account with short delay (< 7s) - smart retry",
@@ -611,13 +618,14 @@ func TestShouldTriggerAntigravitySmartRetry(t *testing.T) {
]
}
}`,
expectedShouldRetry: false,
expectedShouldRateLimit: true,
minWait: 39 * time.Second,
modelName: "gemini-3-pro-high",
expectedShouldRetry: true,
expectedShouldRateLimit: false,
expectedIsModelCapacityExhausted: true,
minWait: 1 * time.Second,
modelName: "gemini-3-pro-high",
},
{
name: "503 UNAVAILABLE with MODEL_CAPACITY_EXHAUSTED - no retryDelay - use default rate limit",
name: "503 UNAVAILABLE with MODEL_CAPACITY_EXHAUSTED - no retryDelay - use fixed wait",
account: oauthAccount,
body: `{
"error": {
@@ -629,10 +637,11 @@ func TestShouldTriggerAntigravitySmartRetry(t *testing.T) {
"message": "No capacity available for model gemini-2.5-flash on the server"
}
}`,
expectedShouldRetry: false,
expectedShouldRateLimit: true,
minWait: 30 * time.Second,
modelName: "gemini-2.5-flash",
expectedShouldRetry: true,
expectedShouldRateLimit: false,
expectedIsModelCapacityExhausted: true,
minWait: 1 * time.Second,
modelName: "gemini-2.5-flash",
},
{
name: "429 RESOURCE_EXHAUSTED with RATE_LIMIT_EXCEEDED - no retryDelay - use default rate limit",
@@ -656,13 +665,16 @@ func TestShouldTriggerAntigravitySmartRetry(t *testing.T) {
for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) {
shouldRetry, shouldRateLimit, wait, model := shouldTriggerAntigravitySmartRetry(tt.account, []byte(tt.body))
shouldRetry, shouldRateLimit, wait, model, isModelCapacityExhausted := shouldTriggerAntigravitySmartRetry(tt.account, []byte(tt.body))
if shouldRetry != tt.expectedShouldRetry {
t.Errorf("shouldRetry = %v, want %v", shouldRetry, tt.expectedShouldRetry)
}
if shouldRateLimit != tt.expectedShouldRateLimit {
t.Errorf("shouldRateLimit = %v, want %v", shouldRateLimit, tt.expectedShouldRateLimit)
}
if isModelCapacityExhausted != tt.expectedIsModelCapacityExhausted {
t.Errorf("isModelCapacityExhausted = %v, want %v", isModelCapacityExhausted, tt.expectedIsModelCapacityExhausted)
}
if shouldRetry {
if wait < tt.minWait {
t.Errorf("wait = %v, want >= %v", wait, tt.minWait)