feat: 添加 Anthropic 缓存 TTL 注入开关

2026-05-04 21:20:51 +08:00 · 2026-04-30 13:38:22 +08:00
parent 094e1171ef
commit 73b872998e
12 changed files with 394 additions and 54 deletions
--- a/backend/internal/handler/admin/setting_handler.go
+++ b/backend/internal/handler/admin/setting_handler.go
@@ -209,6 +209,7 @@ func (h *SettingHandler) GetSettings(c *gin.Context) {
 		EnableFingerprintUnification:           settings.EnableFingerprintUnification,
 		EnableMetadataPassthrough:              settings.EnableMetadataPassthrough,
 		EnableCCHSigning:                       settings.EnableCCHSigning,
+		EnableAnthropicCacheTTL1hInjection:     settings.EnableAnthropicCacheTTL1hInjection,
 		WebSearchEmulationEnabled:              settings.WebSearchEmulationEnabled,
 		PaymentVisibleMethodAlipaySource:       settings.PaymentVisibleMethodAlipaySource,
 		PaymentVisibleMethodWxpaySource:        settings.PaymentVisibleMethodWxpaySource,
@@ -444,6 +445,7 @@ type UpdateSettingsRequest struct {
 	EnableFingerprintUnification       *bool `json:"enable_fingerprint_unification"`
 	EnableMetadataPassthrough          *bool `json:"enable_metadata_passthrough"`
 	EnableCCHSigning                   *bool `json:"enable_cch_signing"`
+	EnableAnthropicCacheTTL1hInjection *bool `json:"enable_anthropic_cache_ttl_1h_injection"`

 	// Payment visible method routing
 	PaymentVisibleMethodAlipaySource  *string `json:"payment_visible_method_alipay_source"`
@@ -1273,6 +1275,12 @@ func (h *SettingHandler) UpdateSettings(c *gin.Context) {
 			}
 			return previousSettings.EnableCCHSigning
 		}(),
+		EnableAnthropicCacheTTL1hInjection: func() bool {
+			if req.EnableAnthropicCacheTTL1hInjection != nil {
+				return *req.EnableAnthropicCacheTTL1hInjection
+			}
+			return previousSettings.EnableAnthropicCacheTTL1hInjection
+		}(),
 		PaymentVisibleMethodAlipaySource: func() string {
 			if req.PaymentVisibleMethodAlipaySource != nil {
 				return strings.TrimSpace(*req.PaymentVisibleMethodAlipaySource)
@@ -1570,6 +1578,7 @@ func (h *SettingHandler) UpdateSettings(c *gin.Context) {
 		EnableFingerprintUnification:           updatedSettings.EnableFingerprintUnification,
 		EnableMetadataPassthrough:              updatedSettings.EnableMetadataPassthrough,
 		EnableCCHSigning:                       updatedSettings.EnableCCHSigning,
+		EnableAnthropicCacheTTL1hInjection:     updatedSettings.EnableAnthropicCacheTTL1hInjection,
 		PaymentVisibleMethodAlipaySource:       updatedSettings.PaymentVisibleMethodAlipaySource,
 		PaymentVisibleMethodWxpaySource:        updatedSettings.PaymentVisibleMethodWxpaySource,
 		PaymentVisibleMethodAlipayEnabled:      updatedSettings.PaymentVisibleMethodAlipayEnabled,
@@ -1949,6 +1958,9 @@ func diffSettings(before *service.SystemSettings, after *service.SystemSettings,
 	if before.EnableCCHSigning != after.EnableCCHSigning {
 		changed = append(changed, "enable_cch_signing")
 	}
+	if before.EnableAnthropicCacheTTL1hInjection != after.EnableAnthropicCacheTTL1hInjection {
+		changed = append(changed, "enable_anthropic_cache_ttl_1h_injection")
+	}
 	if before.PaymentVisibleMethodAlipaySource != after.PaymentVisibleMethodAlipaySource {
 		changed = append(changed, "payment_visible_method_alipay_source")
 	}
--- a/backend/internal/handler/dto/settings.go
+++ b/backend/internal/handler/dto/settings.go
@@ -145,6 +145,7 @@ type SystemSettings struct {
 	EnableFingerprintUnification       bool `json:"enable_fingerprint_unification"`
 	EnableMetadataPassthrough          bool `json:"enable_metadata_passthrough"`
 	EnableCCHSigning                   bool `json:"enable_cch_signing"`
+	EnableAnthropicCacheTTL1hInjection bool `json:"enable_anthropic_cache_ttl_1h_injection"`

 	// Web Search Emulation
 	WebSearchEmulationEnabled bool `json:"web_search_emulation_enabled"`
--- a/backend/internal/service/domain_constants.go
+++ b/backend/internal/service/domain_constants.go
@@ -336,6 +336,8 @@ const (
 	SettingKeyEnableMetadataPassthrough = "enable_metadata_passthrough"
 	// SettingKeyEnableCCHSigning 是否对 billing header 中的 cch 进行 xxHash64 签名（默认 false）
 	SettingKeyEnableCCHSigning = "enable_cch_signing"
+	// SettingKeyEnableAnthropicCacheTTL1hInjection 是否对 Anthropic OAuth/SetupToken 请求体注入 1h cache_control ttl（默认 false）
+	SettingKeyEnableAnthropicCacheTTL1hInjection = "enable_anthropic_cache_ttl_1h_injection"

 	// Balance Low Notification
 	SettingKeyBalanceLowNotifyEnabled     = "balance_low_notify_enabled"      // 全局开关
--- a/backend/internal/service/gateway_body_order_test.go
+++ b/backend/internal/service/gateway_body_order_test.go
@@ -1,13 +1,91 @@
 package service

 import (
+	"context"
+	"errors"
 	"strings"
 	"testing"

+	"github.com/Wei-Shaw/sub2api/internal/config"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/claude"
 	"github.com/stretchr/testify/require"
+	"github.com/tidwall/gjson"
 )

+type gatewayTTLSettingRepo struct {
+	data map[string]string
+}
+
+func (r *gatewayTTLSettingRepo) Get(context.Context, string) (*Setting, error) {
+	return nil, ErrSettingNotFound
+}
+
+func (r *gatewayTTLSettingRepo) GetValue(_ context.Context, key string) (string, error) {
+	if r == nil {
+		return "", ErrSettingNotFound
+	}
+	v, ok := r.data[key]
+	if !ok {
+		return "", ErrSettingNotFound
+	}
+	return v, nil
+}
+
+func (r *gatewayTTLSettingRepo) Set(_ context.Context, key, value string) error {
+	if r == nil {
+		return errors.New("setting repo is nil")
+	}
+	if r.data == nil {
+		r.data = map[string]string{}
+	}
+	r.data[key] = value
+	return nil
+}
+
+func (r *gatewayTTLSettingRepo) GetMultiple(_ context.Context, keys []string) (map[string]string, error) {
+	result := make(map[string]string)
+	if r == nil {
+		return result, nil
+	}
+	for _, key := range keys {
+		if v, ok := r.data[key]; ok {
+			result[key] = v
+		}
+	}
+	return result, nil
+}
+
+func (r *gatewayTTLSettingRepo) SetMultiple(_ context.Context, settings map[string]string) error {
+	if r == nil {
+		return errors.New("setting repo is nil")
+	}
+	if r.data == nil {
+		r.data = map[string]string{}
+	}
+	for key, value := range settings {
+		r.data[key] = value
+	}
+	return nil
+}
+
+func (r *gatewayTTLSettingRepo) GetAll(context.Context) (map[string]string, error) {
+	result := make(map[string]string)
+	if r == nil {
+		return result, nil
+	}
+	for key, value := range r.data {
+		result[key] = value
+	}
+	return result, nil
+}
+
+func (r *gatewayTTLSettingRepo) Delete(_ context.Context, key string) error {
+	if r != nil {
+		delete(r.data, key)
+	}
+	return nil
+}
+
 func assertJSONTokenOrder(t *testing.T, body string, tokens ...string) {
 	t.Helper()

@@ -71,3 +149,60 @@ func TestEnforceCacheControlLimit_PreservesTopLevelFieldOrder(t *testing.T) {
 	assertJSONTokenOrder(t, resultStr, `"alpha"`, `"system"`, `"messages"`, `"omega"`)
 	require.Equal(t, 4, strings.Count(resultStr, `"cache_control"`))
 }
+
+func TestInjectAnthropicCacheControlTTL1h_OnlyUpdatesExistingEphemeralCacheControl(t *testing.T) {
+	body := []byte(`{"alpha":1,"cache_control":{"type":"ephemeral"},"system":[{"type":"text","text":"sys","cache_control":{"type":"ephemeral","ttl":"5m"}},{"type":"text","text":"plain"}],"messages":[{"role":"user","content":[{"type":"text","text":"hi","cache_control":{"type":"ephemeral"}},{"type":"text","text":"non","cache_control":{"type":"persistent","ttl":"5m"}}]}],"tools":[{"name":"a","input_schema":{},"cache_control":{"type":"ephemeral"}}],"omega":2}`)
+
+	result := injectAnthropicCacheControlTTL1h(body)
+	resultStr := string(result)
+
+	assertJSONTokenOrder(t, resultStr, `"alpha"`, `"cache_control"`, `"system"`, `"messages"`, `"tools"`, `"omega"`)
+	require.Equal(t, "1h", gjson.GetBytes(result, "cache_control.ttl").String())
+	require.Equal(t, "1h", gjson.GetBytes(result, "system.0.cache_control.ttl").String())
+	require.False(t, gjson.GetBytes(result, "system.1.cache_control").Exists())
+	require.Equal(t, "1h", gjson.GetBytes(result, "messages.0.content.0.cache_control.ttl").String())
+	require.Equal(t, "5m", gjson.GetBytes(result, "messages.0.content.1.cache_control.ttl").String())
+	require.Equal(t, "1h", gjson.GetBytes(result, "tools.0.cache_control.ttl").String())
+}
+
+func TestGatewayCacheTTLGlobalSetting_TargetResolution(t *testing.T) {
+	repo := &gatewayTTLSettingRepo{data: map[string]string{
+		SettingKeyEnableAnthropicCacheTTL1hInjection: "true",
+	}}
+	gatewayForwardingCache.Store(&cachedGatewayForwardingSettings{})
+	svc := &GatewayService{
+		settingService: NewSettingService(repo, &config.Config{}),
+	}
+	account := &Account{Platform: PlatformAnthropic, Type: AccountTypeOAuth}
+
+	target, ok := svc.resolveCacheTTLUsageOverrideTarget(context.Background(), account)
+	require.True(t, ok)
+	require.Equal(t, cacheTTLTarget5m, target)
+
+	account.Extra = map[string]any{
+		"cache_ttl_override_enabled": true,
+		"cache_ttl_override_target":  "1h",
+	}
+	target, ok = svc.resolveCacheTTLUsageOverrideTarget(context.Background(), account)
+	require.True(t, ok)
+	require.Equal(t, cacheTTLTarget1h, target)
+}
+
+func TestGatewayCacheTTLGlobalSetting_RequestInjectionScope(t *testing.T) {
+	repo := &gatewayTTLSettingRepo{data: map[string]string{
+		SettingKeyEnableAnthropicCacheTTL1hInjection: "true",
+	}}
+	gatewayForwardingCache.Store(&cachedGatewayForwardingSettings{})
+	svc := &GatewayService{
+		settingService: NewSettingService(repo, &config.Config{}),
+	}
+
+	require.True(t, svc.shouldInjectAnthropicCacheTTL1h(context.Background(), &Account{Platform: PlatformAnthropic, Type: AccountTypeOAuth}))
+	require.True(t, svc.shouldInjectAnthropicCacheTTL1h(context.Background(), &Account{Platform: PlatformAnthropic, Type: AccountTypeSetupToken}))
+	require.False(t, svc.shouldInjectAnthropicCacheTTL1h(context.Background(), &Account{Platform: PlatformAnthropic, Type: AccountTypeAPIKey}))
+	require.False(t, svc.shouldInjectAnthropicCacheTTL1h(context.Background(), &Account{Platform: PlatformOpenAI, Type: AccountTypeOAuth}))
+
+	repo.data[SettingKeyEnableAnthropicCacheTTL1hInjection] = "false"
+	gatewayForwardingCache.Store(&cachedGatewayForwardingSettings{})
+	require.False(t, svc.shouldInjectAnthropicCacheTTL1h(context.Background(), &Account{Platform: PlatformAnthropic, Type: AccountTypeOAuth}))
+}
--- a/backend/internal/service/gateway_service.go
+++ b/backend/internal/service/gateway_service.go
@@ -62,6 +62,11 @@ const (
 	claudeMimicDebugInfoKey = "claude_mimic_debug_info"
 )

+const (
+	cacheTTLTarget5m = "5m"
+	cacheTTLTarget1h = "1h"
+)
+
 // ForceCacheBillingContextKey 强制缓存计费上下文键
 // 用于粘性会话切换时，将 input_tokens 转为 cache_read_input_tokens 计费
 type forceCacheBillingKeyType struct{}
@@ -4226,6 +4231,87 @@ func enforceCacheControlLimit(body []byte) []byte {
 	return body
 }

+// injectAnthropicCacheControlTTL1h 将已有 ephemeral cache_control 块的 ttl 强制写为 1h。
+// 仅修改已经存在的 cache_control，不新增缓存断点。
+func injectAnthropicCacheControlTTL1h(body []byte) []byte {
+	return forceEphemeralCacheControlTTL(body, cacheTTLTarget1h)
+}
+
+func forceEphemeralCacheControlTTL(body []byte, ttl string) []byte {
+	if len(body) == 0 || ttl == "" {
+		return body
+	}
+	out := body
+	var paths []string
+	addPath := func(path string, value gjson.Result) {
+		cc := value.Get("cache_control")
+		if !cc.Exists() || cc.Get("type").String() != "ephemeral" {
+			return
+		}
+		if cc.Get("ttl").String() == ttl {
+			return
+		}
+		paths = append(paths, path+".cache_control.ttl")
+	}
+
+	if topCC := gjson.GetBytes(body, "cache_control"); topCC.Exists() && topCC.Get("type").String() == "ephemeral" && topCC.Get("ttl").String() != ttl {
+		paths = append(paths, "cache_control.ttl")
+	}
+
+	system := gjson.GetBytes(body, "system")
+	if system.IsArray() {
+		idx := -1
+		system.ForEach(func(_, block gjson.Result) bool {
+			idx++
+			addPath(fmt.Sprintf("system.%d", idx), block)
+			return true
+		})
+	}
+
+	messages := gjson.GetBytes(body, "messages")
+	if messages.IsArray() {
+		msgIdx := -1
+		messages.ForEach(func(_, msg gjson.Result) bool {
+			msgIdx++
+			content := msg.Get("content")
+			if !content.IsArray() {
+				return true
+			}
+			contentIdx := -1
+			content.ForEach(func(_, block gjson.Result) bool {
+				contentIdx++
+				addPath(fmt.Sprintf("messages.%d.content.%d", msgIdx, contentIdx), block)
+				return true
+			})
+			return true
+		})
+	}
+
+	tools := gjson.GetBytes(body, "tools")
+	if tools.IsArray() {
+		idx := -1
+		tools.ForEach(func(_, tool gjson.Result) bool {
+			idx++
+			addPath(fmt.Sprintf("tools.%d", idx), tool)
+			return true
+		})
+	}
+
+	for _, path := range paths {
+		if next, err := sjson.SetBytes(out, path, ttl); err == nil {
+			out = next
+		}
+	}
+	return out
+}
+
+func (s *GatewayService) shouldInjectAnthropicCacheTTL1h(ctx context.Context, account *Account) bool {
+	if account == nil || !account.IsAnthropicOAuthOrSetupToken() || s == nil || s.settingService == nil {
+		return false
+	}
+	return s.settingService.IsAnthropicCacheTTL1hInjectionEnabled(ctx)
+}
+
 // Forward 转发请求到Claude API
 func (s *GatewayService) Forward(ctx context.Context, c *gin.Context, account *Account, parsed *ParsedRequest) (*ForwardResult, error) {
 	startTime := time.Now()
@@ -4385,6 +4471,10 @@ func (s *GatewayService) Forward(ctx context.Context, c *gin.Context, account *A
 		logger.LegacyPrintf("service.gateway", "Model mapping applied: %s -> %s (account: %s, source=%s)", originalModel, mappedModel, account.Name, mappingSource)
 	}

+	if s.shouldInjectAnthropicCacheTTL1h(ctx, account) {
+		body = injectAnthropicCacheControlTTL1h(body)
+	}
+
 	// 获取凭证
 	token, tokenType, err := s.GetAccessToken(ctx, account)
 	if err != nil {
@@ -7225,9 +7315,9 @@ func (s *GatewayService) handleStreamingResponse(ctx context.Context, resp *http
 			}
 		}

-		// Cache TTL Override: 重写 SSE 事件中的 cache_creation 分类
-		if account.IsCacheTTLOverrideEnabled() {
-			overrideTarget := account.GetCacheTTLOverrideTarget()
+		// Cache TTL Override: 重写 SSE 事件中的 cache_creation 分类。
+		// 账号级设置优先；全局 1h 请求注入开启时，默认把 usage 计费归回 5m。
+		if overrideTarget, ok := s.resolveCacheTTLUsageOverrideTarget(ctx, account); ok {
 			if eventType == "message_start" {
 				if msg, ok := event["message"].(map[string]any); ok {
 					if u, ok := msg["usage"].(map[string]any); ok {
@@ -7634,6 +7724,19 @@ func rewriteCacheCreationJSON(usageObj map[string]any, target string) bool {
 	return true
 }

+func (s *GatewayService) resolveCacheTTLUsageOverrideTarget(ctx context.Context, account *Account) (string, bool) {
+	if account == nil {
+		return "", false
+	}
+	if account.IsCacheTTLOverrideEnabled() {
+		return account.GetCacheTTLOverrideTarget(), true
+	}
+	if account.IsAnthropicOAuthOrSetupToken() && s != nil && s.settingService != nil && s.settingService.IsAnthropicCacheTTL1hInjectionEnabled(ctx) {
+		return cacheTTLTarget5m, true
+	}
+	return "", false
+}
+
 func (s *GatewayService) handleNonStreamingResponse(ctx context.Context, resp *http.Response, c *gin.Context, account *Account, originalModel, mappedModel string) (*ClaudeUsage, error) {
 	// 更新5h窗口状态
 	s.rateLimitService.UpdateSessionWindow(ctx, account, resp.Header)
@@ -7670,9 +7773,9 @@ func (s *GatewayService) handleNonStreamingResponse(ctx context.Context, resp *h
 		}
 	}

-	// Cache TTL Override: 重写 non-streaming 响应中的 cache_creation 分类
-	if account.IsCacheTTLOverrideEnabled() {
-		overrideTarget := account.GetCacheTTLOverrideTarget()
+	// Cache TTL Override: 重写 non-streaming 响应中的 cache_creation 分类。
+	// 账号级设置优先；全局 1h 请求注入开启时，默认把 usage 计费归回 5m。
+	if overrideTarget, ok := s.resolveCacheTTLUsageOverrideTarget(ctx, account); ok {
 		if applyCacheTTLOverride(&response.Usage, overrideTarget) {
 			// 同步更新 body JSON 中的嵌套 cache_creation 对象
 			if newBody, err := sjson.SetBytes(body, "usage.cache_creation.ephemeral_5m_input_tokens", response.Usage.CacheCreation5mTokens); err == nil {
@@ -8240,10 +8343,11 @@ func (s *GatewayService) recordUsageCore(ctx context.Context, input *recordUsage
 		result.Usage.InputTokens = 0
 	}

-	// Cache TTL Override: 确保计费时 token 分类与账号设置一致
+	// Cache TTL Override: 确保计费时 token 分类与账号设置一致。
+	// 账号级设置优先；全局 1h 请求注入开启时，默认把 usage 计费归回 5m。
 	cacheTTLOverridden := false
-	if account.IsCacheTTLOverrideEnabled() {
-		applyCacheTTLOverride(&result.Usage, account.GetCacheTTLOverrideTarget())
+	if overrideTarget, ok := s.resolveCacheTTLUsageOverrideTarget(ctx, account); ok {
+		applyCacheTTLOverride(&result.Usage, overrideTarget)
 		cacheTTLOverridden = (result.Usage.CacheCreation5mTokens + result.Usage.CacheCreation1hTokens) > 0
 	}

--- a/backend/internal/service/setting_service.go
+++ b/backend/internal/service/setting_service.go
@@ -85,6 +85,7 @@ type cachedGatewayForwardingSettings struct {
 	fingerprintUnification       bool
 	metadataPassthrough          bool
 	cchSigning                   bool
+	anthropicCacheTTL1hInjection bool
 	expiresAt                    int64 // unix nano
 }

@@ -1245,6 +1246,7 @@ func (s *SettingService) buildSystemSettingsUpdates(ctx context.Context, setting
 	updates[SettingKeyEnableFingerprintUnification] = strconv.FormatBool(settings.EnableFingerprintUnification)
 	updates[SettingKeyEnableMetadataPassthrough] = strconv.FormatBool(settings.EnableMetadataPassthrough)
 	updates[SettingKeyEnableCCHSigning] = strconv.FormatBool(settings.EnableCCHSigning)
+	updates[SettingKeyEnableAnthropicCacheTTL1hInjection] = strconv.FormatBool(settings.EnableAnthropicCacheTTL1hInjection)
 	updates[SettingPaymentVisibleMethodAlipaySource] = settings.PaymentVisibleMethodAlipaySource
 	updates[SettingPaymentVisibleMethodWxpaySource] = settings.PaymentVisibleMethodWxpaySource
 	updates[SettingPaymentVisibleMethodAlipayEnabled] = strconv.FormatBool(settings.PaymentVisibleMethodAlipayEnabled)
@@ -1308,6 +1310,7 @@ func (s *SettingService) refreshCachedSettings(settings *SystemSettings) {
 		fingerprintUnification:       settings.EnableFingerprintUnification,
 		metadataPassthrough:          settings.EnableMetadataPassthrough,
 		cchSigning:                   settings.EnableCCHSigning,
+		anthropicCacheTTL1hInjection: settings.EnableAnthropicCacheTTL1hInjection,
 		expiresAt:                    time.Now().Add(gatewayForwardingCacheTTL).UnixNano(),
 	})
 	openAIAdvancedSchedulerSettingSF.Forget(openAIAdvancedSchedulerSettingKey)
@@ -1415,22 +1418,30 @@ func (s *SettingService) IsBackendModeEnabled(ctx context.Context) bool {
 	return false
 }

-// GetGatewayForwardingSettings returns cached gateway forwarding settings.
-// Uses in-process atomic.Value cache with 60s TTL, zero-lock hot path.
-// Returns (fingerprintUnification, metadataPassthrough, cchSigning).
-func (s *SettingService) GetGatewayForwardingSettings(ctx context.Context) (fingerprintUnification, metadataPassthrough, cchSigning bool) {
+type gatewayForwardingSettingsResult struct {
+	fp, mp, cch, cacheTTL1h bool
+}
+
+func (s *SettingService) getGatewayForwardingSettingsCached(ctx context.Context) gatewayForwardingSettingsResult {
 	if cached, ok := gatewayForwardingCache.Load().(*cachedGatewayForwardingSettings); ok && cached != nil {
 		if time.Now().UnixNano() < cached.expiresAt {
-			return cached.fingerprintUnification, cached.metadataPassthrough, cached.cchSigning
+			return gatewayForwardingSettingsResult{
+				fp:         cached.fingerprintUnification,
+				mp:         cached.metadataPassthrough,
+				cch:        cached.cchSigning,
+				cacheTTL1h: cached.anthropicCacheTTL1hInjection,
 			}
 		}
-	type gwfResult struct {
-		fp, mp, cch bool
 	}
 	val, _, _ := gatewayForwardingSF.Do("gateway_forwarding", func() (any, error) {
 		if cached, ok := gatewayForwardingCache.Load().(*cachedGatewayForwardingSettings); ok && cached != nil {
 			if time.Now().UnixNano() < cached.expiresAt {
-				return gwfResult{cached.fingerprintUnification, cached.metadataPassthrough, cached.cchSigning}, nil
+				return gatewayForwardingSettingsResult{
+					fp:         cached.fingerprintUnification,
+					mp:         cached.metadataPassthrough,
+					cch:        cached.cchSigning,
+					cacheTTL1h: cached.anthropicCacheTTL1hInjection,
+				}, nil
 			}
 		}
 		dbCtx, cancel := context.WithTimeout(context.WithoutCancel(ctx), gatewayForwardingDBTimeout)
@@ -1439,6 +1450,7 @@ func (s *SettingService) GetGatewayForwardingSettings(ctx context.Context) (fing
 			SettingKeyEnableFingerprintUnification,
 			SettingKeyEnableMetadataPassthrough,
 			SettingKeyEnableCCHSigning,
+			SettingKeyEnableAnthropicCacheTTL1hInjection,
 		})
 		if err != nil {
 			slog.Warn("failed to get gateway forwarding settings", "error", err)
@@ -1446,9 +1458,10 @@ func (s *SettingService) GetGatewayForwardingSettings(ctx context.Context) (fing
 				fingerprintUnification:       true,
 				metadataPassthrough:          false,
 				cchSigning:                   false,
+				anthropicCacheTTL1hInjection: false,
 				expiresAt:                    time.Now().Add(gatewayForwardingErrorTTL).UnixNano(),
 			})
-			return gwfResult{true, false, false}, nil
+			return gatewayForwardingSettingsResult{fp: true}, nil
 		}
 		fp := true
 		if v, ok := values[SettingKeyEnableFingerprintUnification]; ok && v != "" {
@@ -1456,18 +1469,33 @@ func (s *SettingService) GetGatewayForwardingSettings(ctx context.Context) (fing
 		}
 		mp := values[SettingKeyEnableMetadataPassthrough] == "true"
 		cch := values[SettingKeyEnableCCHSigning] == "true"
+		cacheTTL1h := values[SettingKeyEnableAnthropicCacheTTL1hInjection] == "true"
 		gatewayForwardingCache.Store(&cachedGatewayForwardingSettings{
 			fingerprintUnification:       fp,
 			metadataPassthrough:          mp,
 			cchSigning:                   cch,
+			anthropicCacheTTL1hInjection: cacheTTL1h,
 			expiresAt:                    time.Now().Add(gatewayForwardingCacheTTL).UnixNano(),
 		})
-		return gwfResult{fp, mp, cch}, nil
+		return gatewayForwardingSettingsResult{fp: fp, mp: mp, cch: cch, cacheTTL1h: cacheTTL1h}, nil
 	})
-	if r, ok := val.(gwfResult); ok {
-		return r.fp, r.mp, r.cch
+	if r, ok := val.(gatewayForwardingSettingsResult); ok {
+		return r
 	}
-	return true, false, false // fail-open defaults
+	return gatewayForwardingSettingsResult{fp: true}
+}
+
+// GetGatewayForwardingSettings returns cached gateway forwarding settings.
+// Uses in-process atomic.Value cache with 60s TTL, zero-lock hot path.
+// Returns (fingerprintUnification, metadataPassthrough, cchSigning).
+func (s *SettingService) GetGatewayForwardingSettings(ctx context.Context) (fingerprintUnification, metadataPassthrough, cchSigning bool) {
+	result := s.getGatewayForwardingSettingsCached(ctx)
+	return result.fp, result.mp, result.cch
+}
+
+// IsAnthropicCacheTTL1hInjectionEnabled 检查是否对 Anthropic OAuth/SetupToken 请求体注入 1h cache_control ttl。
+func (s *SettingService) IsAnthropicCacheTTL1hInjectionEnabled(ctx context.Context) bool {
+	return s.getGatewayForwardingSettingsCached(ctx).cacheTTL1h
 }

 // IsEmailVerifyEnabled 检查是否开启邮件验证
@@ -1881,6 +1909,7 @@ func (s *SettingService) InitializeDefaultSettings(ctx context.Context) error {

 		// 分组隔离（默认不允许未分组 Key 调度）
 		SettingKeyAllowUngroupedKeyScheduling:        "false",
+		SettingKeyEnableAnthropicCacheTTL1hInjection: "false",
 		SettingPaymentVisibleMethodAlipaySource:      "",
 		SettingPaymentVisibleMethodWxpaySource:       "",
 		SettingPaymentVisibleMethodAlipayEnabled:     "false",
@@ -2228,6 +2257,7 @@ func (s *SettingService) parseSettings(settings map[string]string) *SystemSettin
 	}
 	result.EnableMetadataPassthrough = settings[SettingKeyEnableMetadataPassthrough] == "true"
 	result.EnableCCHSigning = settings[SettingKeyEnableCCHSigning] == "true"
+	result.EnableAnthropicCacheTTL1hInjection = settings[SettingKeyEnableAnthropicCacheTTL1hInjection] == "true"

 	// Web search emulation: quick enabled check from the JSON config
 	if raw := settings[SettingKeyWebSearchEmulationConfig]; raw != "" {
--- a/backend/internal/service/settings_view.go
+++ b/backend/internal/service/settings_view.go
@@ -152,6 +152,7 @@ type SystemSettings struct {
 	EnableFingerprintUnification       bool // 是否统一 OAuth 账号的指纹头（默认 true）
 	EnableMetadataPassthrough          bool // 是否透传客户端原始 metadata（默认 false）
 	EnableCCHSigning                   bool // 是否对 billing header cch 进行签名（默认 false）
+	EnableAnthropicCacheTTL1hInjection bool // 是否对 Anthropic OAuth/SetupToken 请求体注入 1h cache_control ttl（默认 false）

 	// Web Search Emulation
 	WebSearchEmulationEnabled bool // 是否启用 web search 模拟
--- a/frontend/src/api/admin/settings.ts
+++ b/frontend/src/api/admin/settings.ts
@@ -439,6 +439,7 @@ export interface SystemSettings {
  enable_fingerprint_unification: boolean;
  enable_metadata_passthrough: boolean;
  enable_cch_signing: boolean;
+  enable_anthropic_cache_ttl_1h_injection: boolean;
  web_search_emulation_enabled?: boolean;

  // Payment configuration
@@ -609,6 +610,7 @@ export interface UpdateSettingsRequest {
  enable_fingerprint_unification?: boolean;
  enable_metadata_passthrough?: boolean;
  enable_cch_signing?: boolean;
+  enable_anthropic_cache_ttl_1h_injection?: boolean;
  // Payment configuration
  payment_enabled?: boolean;
  payment_min_amount?: number;
--- a/frontend/src/i18n/locales/en.ts
+++ b/frontend/src/i18n/locales/en.ts
@@ -5019,6 +5019,8 @@ export default {
        metadataPassthroughHint: 'Pass through client\'s original metadata.user_id without rewriting. May improve upstream cache hit rates.',
        cchSigning: 'CCH Signing',
        cchSigningHint: 'Sign the billing header in forwarded requests with CCH hash. When disabled, the placeholder is preserved.',
+        anthropicCacheTTL1hInjection: 'Anthropic Cache TTL Injection',
+        anthropicCacheTTL1hInjectionHint: 'When enabled, existing ephemeral cache_control blocks in Anthropic OAuth/Setup Token request bodies are forced to 1h; response usage is billed back as 5m by default, with account-level TTL billing override taking priority.',
      },
      webSearchEmulation: {
        title: 'Web Search Emulation',
--- a/frontend/src/i18n/locales/zh.ts
+++ b/frontend/src/i18n/locales/zh.ts
@@ -5178,6 +5178,8 @@ export default {
        metadataPassthroughHint: '透传客户端原始 metadata.user_id，不进行重写。可能提高上游缓存命中率。',
        cchSigning: 'CCH 签名',
        cchSigningHint: '对转发请求的 billing header 进行 CCH 哈希签名。关闭时保留原始占位符。',
+        anthropicCacheTTL1hInjection: 'Anthropic 缓存 TTL 注入',
+        anthropicCacheTTL1hInjectionHint: '开启后，对 Anthropic OAuth/Setup Token 请求体中已有的 ephemeral 缓存块强制写入 1h；响应 usage 默认按 5m 回写计费，账号级 TTL 计费设置优先。',
      },
      webSearchEmulation: {
        title: 'Web Search 模拟',
--- a/frontend/src/views/admin/SettingsView.vue
+++ b/frontend/src/views/admin/SettingsView.vue
@@ -3057,6 +3057,31 @@
                </div>
                <Toggle v-model="form.enable_cch_signing" />
              </div>
+
+              <!-- Anthropic Cache TTL 1h Injection -->
+              <div class="flex items-center justify-between">
+                <div>
+                  <label
+                    class="text-sm font-medium text-gray-700 dark:text-gray-300"
+                  >
+                    {{
+                      t(
+                        "admin.settings.gatewayForwarding.anthropicCacheTTL1hInjection",
+                      )
+                    }}
+                  </label>
+                  <p class="mt-0.5 text-xs text-gray-500 dark:text-gray-400">
+                    {{
+                      t(
+                        "admin.settings.gatewayForwarding.anthropicCacheTTL1hInjectionHint",
+                      )
+                    }}
+                  </p>
+                </div>
+                <Toggle
+                  v-model="form.enable_anthropic_cache_ttl_1h_injection"
+                />
+              </div>
            </div>
          </div>
          <!-- Web Search Emulation -->
@@ -5810,6 +5835,7 @@ const form = reactive<SettingsForm>({
  enable_fingerprint_unification: true,
  enable_metadata_passthrough: false,
  enable_cch_signing: false,
+  enable_anthropic_cache_ttl_1h_injection: false,
  // Balance & quota notification
  balance_low_notify_enabled: false,
  balance_low_notify_threshold: 0,
@@ -6718,6 +6744,8 @@ async function saveSettings() {
      enable_fingerprint_unification: form.enable_fingerprint_unification,
      enable_metadata_passthrough: form.enable_metadata_passthrough,
      enable_cch_signing: form.enable_cch_signing,
+      enable_anthropic_cache_ttl_1h_injection:
+        form.enable_anthropic_cache_ttl_1h_injection,
      // Payment configuration
      payment_enabled: form.payment_enabled,
      payment_min_amount: Number(form.payment_min_amount) || 0,
--- a/frontend/src/views/admin/tests/SettingsView.spec.ts
+++ b/frontend/src/views/admin/tests/SettingsView.spec.ts
@@ -362,6 +362,7 @@ const baseSettingsResponse = {
  enable_fingerprint_unification: true,
  enable_metadata_passthrough: false,
  enable_cch_signing: false,
+  enable_anthropic_cache_ttl_1h_injection: false,
  payment_enabled: true,
  payment_min_amount: 1,
  payment_max_amount: 10000,
@@ -567,6 +568,26 @@ describe("admin SettingsView payment visible method controls", () => {
    expect(payload).not.toHaveProperty("payment_visible_method_wxpay_enabled");
  });

+  it("submits Anthropic cache TTL injection gateway setting", async () => {
+    getSettings.mockResolvedValueOnce({
+      ...baseSettingsResponse,
+      enable_anthropic_cache_ttl_1h_injection: true,
+    });
+
+    const wrapper = mountView();
+
+    await flushPromises();
+    await wrapper.find("form").trigger("submit.prevent");
+    await flushPromises();
+
+    expect(updateSettings).toHaveBeenCalledTimes(1);
+    expect(updateSettings).toHaveBeenCalledWith(
+      expect.objectContaining({
+        enable_anthropic_cache_ttl_1h_injection: true,
+      }),
+    );
+  });
+
  it("updates provider enablement immediately and reloads providers", async () => {
    const provider = {
      id: 7,