refactor: remove stream smoothing config from system settings and improve base image caching

- Remove stream_smoothing configuration from SystemConfigService (moved to handler default)
- Remove stream smoothing UI controls from admin settings page
- Add AdminClearSingleAffinityAdapter for targeted cache invalidation
- Add clearSingleAffinity API endpoint to clear specific affinity cache entries
- Include global_model_id in affinity list response for UI deletion support
- Improve CI/CD workflow with hash-based base image change detection
- Add hash label to base image for reliable cache invalidation detection
- Use remote image inspection to determine if base image rebuild is needed
- Include Dockerfile.base in hash calculation for proper dependency tracking

.github/workflows/docker-publish.yml

@@ -15,6 +15,8 @@ env:
   REGISTRY: ghcr.io
   BASE_IMAGE_NAME: fawney19/aether-base
   APP_IMAGE_NAME: fawney19/aether
+  # Files that affect base image - used for hash calculation
+  BASE_FILES: "Dockerfile.base pyproject.toml frontend/package.json frontend/package-lock.json"
 
 jobs:
   check-base-changes:
@@ -23,8 +25,13 @@ jobs:
       base_changed: ${{ steps.check.outputs.base_changed }}
     steps:
       - uses: actions/checkout@v4
+
+      - name: Log in to Container Registry
+        uses: docker/login-action@v3
         with:
-          fetch-depth: 2
+          registry: ${{ env.REGISTRY }}
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
 
       - name: Check if base image needs rebuild
         id: check
@@ -34,10 +41,26 @@ jobs:
             exit 0
           fi
 
-          # Check if base-related files changed
+          # Calculate current hash of base-related files
-          if git diff --name-only HEAD~1 HEAD | grep -qE '^(Dockerfile\.base|pyproject\.toml|frontend/package.*\.json)$'; then
+          CURRENT_HASH=$(cat ${{ env.BASE_FILES }} 2>/dev/null | sha256sum | cut -d' ' -f1)
+          echo "Current base files hash: $CURRENT_HASH"
+
+          # Try to get hash label from remote image config
+          # Pull the image config and extract labels
+          REMOTE_HASH=""
+          if docker pull ${{ env.REGISTRY }}/${{ env.BASE_IMAGE_NAME }}:latest 2>/dev/null; then
+            REMOTE_HASH=$(docker inspect ${{ env.REGISTRY }}/${{ env.BASE_IMAGE_NAME }}:latest --format '{{ index .Config.Labels "org.opencontainers.image.base.hash" }}' 2>/dev/null) || true
+          fi
+
+          if [ -z "$REMOTE_HASH" ] || [ "$REMOTE_HASH" == "<no value>" ]; then
+            # No remote image or no hash label, need to rebuild
+            echo "No remote base image or hash label found, need rebuild"
+            echo "base_changed=true" >> $GITHUB_OUTPUT
+          elif [ "$CURRENT_HASH" != "$REMOTE_HASH" ]; then
+            echo "Hash mismatch: remote=$REMOTE_HASH, current=$CURRENT_HASH"
             echo "base_changed=true" >> $GITHUB_OUTPUT
           else
+            echo "Hash matches, no rebuild needed"
             echo "base_changed=false" >> $GITHUB_OUTPUT
           fi
 
@@ -61,6 +84,12 @@ jobs:
           username: ${{ github.actor }}
           password: ${{ secrets.GITHUB_TOKEN }}
 
+      - name: Calculate base files hash
+        id: hash
+        run: |
+          HASH=$(cat ${{ env.BASE_FILES }} 2>/dev/null | sha256sum | cut -d' ' -f1)
+          echo "hash=$HASH" >> $GITHUB_OUTPUT
+
       - name: Extract metadata for base image
         id: meta
         uses: docker/metadata-action@v5
@@ -69,6 +98,8 @@ jobs:
           tags: |
             type=raw,value=latest
             type=sha,prefix=
+          labels: |
+            org.opencontainers.image.base.hash=${{ steps.hash.outputs.hash }}
 
       - name: Build and push base image
         uses: docker/build-push-action@v5

frontend/src/api/cache.ts

@@ -66,6 +66,7 @@ export interface UserAffinity {
   key_name: string | null
   key_prefix: string | null  // Provider Key 脱敏显示（前4...后4）
   rate_multiplier: number
+  global_model_id: string | null  // 原始的 global_model_id（用于删除）
   model_name: string | null  // 模型名称（如 claude-haiku-4-5-20250514）
   model_display_name: string | null  // 模型显示名称（如 Claude Haiku 4.5）
   api_format: string | null  // API 格式 (claude/openai)
@@ -119,6 +120,18 @@ export const cacheApi = {
     await api.delete(`/api/admin/monitoring/cache/users/${userIdentifier}`)
   },
 
+  /**
+   * 清除单条缓存亲和性
+   *
+   * @param affinityKey API Key ID
+   * @param endpointId Endpoint ID
+   * @param modelId GlobalModel ID
+   * @param apiFormat API 格式 (claude/openai)
+   */
+  async clearSingleAffinity(affinityKey: string, endpointId: string, modelId: string, apiFormat: string): Promise<void> {
+    await api.delete(`/api/admin/monitoring/cache/affinity/${affinityKey}/${endpointId}/${modelId}/${apiFormat}`)
+  },
+
   /**
    * 清除所有缓存
    */

frontend/src/views/admin/CacheMonitoring.vue

@@ -142,32 +142,37 @@ async function resetAffinitySearch() {
   await fetchAffinityList()
 }
 
-async function clearUserCache(identifier: string, displayName?: string) {
+async function clearSingleAffinity(item: UserAffinity) {
-  const target = identifier?.trim()
+  const affinityKey = item.affinity_key?.trim()
-  if (!target) {
+  const endpointId = item.endpoint_id?.trim()
-    showError('无法识别标识符')
+  const modelId = item.global_model_id?.trim()
+  const apiFormat = item.api_format?.trim()
+
+  if (!affinityKey || !endpointId || !modelId || !apiFormat) {
+    showError('缓存记录信息不完整，无法删除')
     return
   }
 
-  const label = displayName || target
+  const label = item.user_api_key_name || affinityKey
+  const modelLabel = item.model_display_name || item.model_name || modelId
   const confirmed = await showConfirm({
     title: '确认清除',
-    message: `确定要清除 ${label} 的缓存吗？`,
+    message: `确定要清除 ${label} 在模型 ${modelLabel} 上的缓存亲和性吗？`,
     confirmText: '确认清除',
     variant: 'destructive'
   })
 
   if (!confirmed) return
 
-  clearingRowAffinityKey.value = target
+  clearingRowAffinityKey.value = affinityKey
   try {
-    await cacheApi.clearUserCache(target)
+    await cacheApi.clearSingleAffinity(affinityKey, endpointId, modelId, apiFormat)
     showSuccess('清除成功')
     await fetchCacheStats()
     await fetchAffinityList(tableKeyword.value.trim() || undefined)
   } catch (error) {
     showError('清除失败')
-    log.error('清除用户缓存失败', error)
+    log.error('清除单条缓存失败', error)
   } finally {
     clearingRowAffinityKey.value = null
   }
@@ -618,7 +623,7 @@ onBeforeUnmount(() => {
                 class="h-7 w-7 text-muted-foreground/70 hover:text-destructive"
                 :disabled="clearingRowAffinityKey === item.affinity_key"
                 title="清除缓存"
-                @click="clearUserCache(item.affinity_key, item.user_api_key_name || item.affinity_key)"
+                @click="clearSingleAffinity(item)"
               >
                 <Trash2 class="h-3.5 w-3.5" />
               </Button>
@@ -668,7 +673,7 @@ onBeforeUnmount(() => {
               variant="ghost"
               class="h-7 w-7 text-muted-foreground/70 hover:text-destructive shrink-0"
               :disabled="clearingRowAffinityKey === item.affinity_key"
-              @click="clearUserCache(item.affinity_key, item.user_api_key_name || item.affinity_key)"
+              @click="clearSingleAffinity(item)"
             >
               <Trash2 class="h-3.5 w-3.5" />
             </Button>

frontend/src/views/admin/SystemSettings.vue

@@ -465,77 +465,6 @@
         </div>
       </CardSection>
 
-      <!-- 流式输出配置 -->
-      <CardSection
-        title="流式输出"
-        description="配置流式响应的输出效果"
-      >
-        <div class="grid grid-cols-1 md:grid-cols-2 gap-6">
-          <div class="md:col-span-2">
-            <div class="flex items-center space-x-2">
-              <Checkbox
-                id="stream-smoothing-enabled"
-                v-model:checked="systemConfig.stream_smoothing_enabled"
-              />
-              <div>
-                <Label
-                  for="stream-smoothing-enabled"
-                  class="cursor-pointer"
-                >
-                  启用平滑输出
-                </Label>
-                <p class="text-xs text-muted-foreground">
-                  将上游返回的大块内容拆分成小块，模拟打字效果
-                </p>
-              </div>
-            </div>
-          </div>
-
-          <div>
-            <Label
-              for="stream-smoothing-chunk-size"
-              class="block text-sm font-medium"
-            >
-              每块字符数
-            </Label>
-            <Input
-              id="stream-smoothing-chunk-size"
-              v-model.number="systemConfig.stream_smoothing_chunk_size"
-              type="number"
-              min="1"
-              max="100"
-              placeholder="20"
-              class="mt-1"
-              :disabled="!systemConfig.stream_smoothing_enabled"
-            />
-            <p class="mt-1 text-xs text-muted-foreground">
-              每次输出的字符数量（1-100）
-            </p>
-          </div>
-
-          <div>
-            <Label
-              for="stream-smoothing-delay-ms"
-              class="block text-sm font-medium"
-            >
-              输出间隔 (毫秒)
-            </Label>
-            <Input
-              id="stream-smoothing-delay-ms"
-              v-model.number="systemConfig.stream_smoothing_delay_ms"
-              type="number"
-              min="1"
-              max="100"
-              placeholder="8"
-              class="mt-1"
-              :disabled="!systemConfig.stream_smoothing_enabled"
-            />
-            <p class="mt-1 text-xs text-muted-foreground">
-              每块之间的延迟毫秒数（1-100）
-            </p>
-          </div>
-        </div>
-      </CardSection>
     </div>
 
     <!-- 导入配置对话框 -->
@@ -884,10 +813,6 @@ interface SystemConfig {
   log_retention_days: number
   cleanup_batch_size: number
   audit_log_retention_days: number
-  // 流式输出
-  stream_smoothing_enabled: boolean
-  stream_smoothing_chunk_size: number
-  stream_smoothing_delay_ms: number
 }
 
 const loading = ref(false)
@@ -937,10 +862,6 @@ const systemConfig = ref<SystemConfig>({
   log_retention_days: 365,
   cleanup_batch_size: 1000,
   audit_log_retention_days: 30,
-  // 流式输出
-  stream_smoothing_enabled: false,
-  stream_smoothing_chunk_size: 20,
-  stream_smoothing_delay_ms: 8,
 })
 
 // 计算属性：KB 和 字节 之间的转换
@@ -997,10 +918,6 @@ async function loadSystemConfig() {
       'log_retention_days',
       'cleanup_batch_size',
       'audit_log_retention_days',
-      // 流式输出
-      'stream_smoothing_enabled',
-      'stream_smoothing_chunk_size',
-      'stream_smoothing_delay_ms',
     ]
 
     for (const key of configs) {
@@ -1108,22 +1025,6 @@ async function saveSystemConfig() {
         value: systemConfig.value.audit_log_retention_days,
         description: '审计日志保留天数'
       },
-      // 流式输出
-      {
-        key: 'stream_smoothing_enabled',
-        value: systemConfig.value.stream_smoothing_enabled,
-        description: '是否启用流式平滑输出'
-      },
-      {
-        key: 'stream_smoothing_chunk_size',
-        value: systemConfig.value.stream_smoothing_chunk_size,
-        description: '流式平滑输出每个小块的字符数'
-      },
-      {
-        key: 'stream_smoothing_delay_ms',
-        value: systemConfig.value.stream_smoothing_delay_ms,
-        description: '流式平滑输出每个小块之间的延迟毫秒数'
-      },
     ]
 
     const promises = configItems.map(item =>

src/api/admin/monitoring/cache.py

@@ -186,6 +186,30 @@ async def clear_user_cache(
     return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
 
 
+@router.delete("/affinity/{affinity_key}/{endpoint_id}/{model_id}/{api_format}")
+async def clear_single_affinity(
+    affinity_key: str,
+    endpoint_id: str,
+    model_id: str,
+    api_format: str,
+    request: Request,
+    db: Session = Depends(get_db),
+) -> Any:
+    """
+    Clear a single cache affinity entry
+
+    Parameters:
+    - affinity_key: API Key ID
+    - endpoint_id: Endpoint ID
+    - model_id: Model ID (GlobalModel ID)
+    - api_format: API format (claude/openai)
+    """
+    adapter = AdminClearSingleAffinityAdapter(
+        affinity_key=affinity_key, endpoint_id=endpoint_id, model_id=model_id, api_format=api_format
+    )
+    return await pipeline.run(adapter=adapter, http_request=request, db=db, mode=adapter.mode)
+
+
 @router.delete("")
 async def clear_all_cache(
     request: Request,
@@ -655,6 +679,7 @@ class AdminListAffinitiesAdapter(AdminApiAdapter):
                 "key_name": key.name if key else None,
                 "key_prefix": provider_key_masked,
                 "rate_multiplier": key.rate_multiplier if key else 1.0,
+                "global_model_id": affinity.get("model_name"),  # 原始的 global_model_id
                 "model_name": (
                     global_model_map.get(affinity.get("model_name")).name
                     if affinity.get("model_name") and global_model_map.get(affinity.get("model_name"))
@@ -817,6 +842,65 @@ class AdminClearUserCacheAdapter(AdminApiAdapter):
             raise HTTPException(status_code=500, detail=f"清除失败: {exc}")
 
 
+@dataclass
+class AdminClearSingleAffinityAdapter(AdminApiAdapter):
+    affinity_key: str
+    endpoint_id: str
+    model_id: str
+    api_format: str
+
+    async def handle(self, context: ApiRequestContext) -> Dict[str, Any]:  # type: ignore[override]
+        db = context.db
+        try:
+            redis_client = get_redis_client_sync()
+            affinity_mgr = await get_affinity_manager(redis_client)
+
+            # 直接获取指定的亲和性记录（无需遍历全部）
+            existing_affinity = await affinity_mgr.get_affinity(
+                self.affinity_key, self.api_format, self.model_id
+            )
+
+            if not existing_affinity:
+                raise HTTPException(status_code=404, detail="未找到指定的缓存亲和性记录")
+
+            # 验证 endpoint_id 是否匹配
+            if existing_affinity.endpoint_id != self.endpoint_id:
+                raise HTTPException(status_code=404, detail="未找到指定的缓存亲和性记录")
+
+            # 失效单条记录
+            await affinity_mgr.invalidate_affinity(
+                self.affinity_key, self.api_format, self.model_id, endpoint_id=self.endpoint_id
+            )
+
+            # 获取用于日志的信息
+            api_key = db.query(ApiKey).filter(ApiKey.id == self.affinity_key).first()
+            api_key_name = api_key.name if api_key else None
+
+            logger.info(
+                f"已清除单条缓存亲和性: affinity_key={self.affinity_key[:8]}..., "
+                f"endpoint_id={self.endpoint_id[:8]}..., model_id={self.model_id[:8]}..."
+            )
+
+            context.add_audit_metadata(
+                action="cache_clear_single",
+                affinity_key=self.affinity_key,
+                endpoint_id=self.endpoint_id,
+                model_id=self.model_id,
+            )
+            return {
+                "status": "ok",
+                "message": f"已清除缓存亲和性: {api_key_name or self.affinity_key[:8]}",
+                "affinity_key": self.affinity_key,
+                "endpoint_id": self.endpoint_id,
+                "model_id": self.model_id,
+            }
+        except HTTPException:
+            raise
+        except Exception as exc:
+            logger.exception(f"清除单条缓存亲和性失败: {exc}")
+            raise HTTPException(status_code=500, detail=f"清除失败: {exc}")
+
+
 class AdminClearAllCacheAdapter(AdminApiAdapter):
     async def handle(self, context: ApiRequestContext) -> Dict[str, Any]:  # type: ignore[override]
         try:

src/config/settings.py

@@ -143,7 +143,6 @@ class Config:
         # STREAM_STATS_DELAY: 统计记录延迟（秒），等待流完全关闭
         self.stream_prefetch_lines = int(os.getenv("STREAM_PREFETCH_LINES", "5"))
         self.stream_stats_delay = float(os.getenv("STREAM_STATS_DELAY", "0.1"))
-        # 注：流式平滑输出配置已移至数据库系统设置（stream_smoothing_*）
 
         # 验证连接池配置
         self._validate_pool_config()

src/services/system/config.py

@@ -78,19 +78,6 @@ class SystemConfigService:
             "value": False,
             "description": "是否自动删除过期的API Key（True=物理删除，False=仅禁用），仅管理员可配置",
         },
-        # 流式平滑输出配置
-        "stream_smoothing_enabled": {
-            "value": False,
-            "description": "是否启用流式平滑输出，自动根据文本长度调整输出速度",
-        },
-        "stream_smoothing_chunk_size": {
-            "value": 20,
-            "description": "流式平滑输出每个小块的字符数",
-        },
-        "stream_smoothing_delay_ms": {
-            "value": 8,
-            "description": "流式平滑输出每个小块之间的延迟毫秒数",
-        },
     }
 
     @classmethod