初始化空白分支文件

.env

@@ -0,0 +1,23 @@
+# 如果你的服务器在中国，可能需要设置代理
+# HTTPS_PROXY=http://192.168.1.x:7890
+# HTTP_PROXY=http://192.168.1.x:7890
+
+LITELLM_LOG=INFO
+#LITELLM_LOG=DEBUG
+
+DAILY_LIMIT_LOG_LEVEL=DEBUG
+
+# Master Key 配置
+LITELLM_MASTER_KEY=sk-7v9m3CvOR2r9rDr3yhTg4WiHjWPfTeH6p5qwEIjwE0lpfOKK
+
+# 你的 NewAPI 配置
+NEWAPI_BASE_URL=http://192.168.1.4:23002/v1
+NEWAPI_API_KEY=sk-xFRYNClzkWXIzC9tJPbOI5LMk66uZfqrR6FYH9SWVw1dKKCN
+
+# 利用 LiteLLM 的默认回退机制，直接把 NewAPI 伪装成标准的 OpenAI 接口
+OPENAI_API_BASE=http://192.168.1.4:23002/v1
+OPENAI_API_KEY=sk-xFRYNClzkWXIzC9tJPbOI5LMk66uZfqrR6FYH9SWVw1dKKCN
+
+#火山平台抖音
+HUOSHAN_BASE_URL=https://ark.cn-beijing.volces.com/api/v3
+HUOSHAN_API_KEY=aacf11dd-2cdd-4d9a-a819-3e3a0310847b

config.yaml

@@ -0,0 +1,304 @@
+##yaml锚点方便配置。手动设定价格，方便ui中查看
+x-definitions: &token_pricing
+  input_cost_per_token: 1
+  output_cost_per_token: 1
+
+model_list:
+
+ #=======================火山平台，限时免费 
+ 
+  #日常使用ds v3.2  ，强制think  
+  - model_name: deepseek-v3.2
+    litellm_params:
+      <<: *token_pricing
+      model:  openai/deepseek-v3-2-251201
+      api_base: os.environ/HUOSHAN_BASE_URL
+      api_key: os.environ/HUOSHAN_API_KEY 
+      # 说明中建议设置30分钟(1800秒)以上，避免深度思考过程中断连接
+      timeout: 1800       
+      # 关键配置2：启用思考模式
+      # extra_body 中的参数会被直接合并到请求的 JSON body 中
+      extra_body:
+        thinking:
+          type: "enabled"
+          
+  - model_name: deepseek-v3.2-nothink
+    litellm_params:
+      <<: *token_pricing
+      model:  openai/deepseek-v3-2-251201
+      api_base: os.environ/HUOSHAN_BASE_URL
+      api_key: os.environ/HUOSHAN_API_KEY 
+      extra_body:
+        thinking:
+          type: "disabled" 
+ 
+        
+  - model_name: doubao-seed-1-8
+    litellm_params:
+      <<: *token_pricing
+      model: openai/doubao-seed-1-8-251228
+      api_base: os.environ/HUOSHAN_BASE_URL
+      api_key: os.environ/HUOSHAN_API_KEY 
+
+  - model_name: doubao-seed-1-8-nothink
+    litellm_params:
+      <<: *token_pricing
+      model: openai/doubao-seed-1-8-251228
+      api_base: os.environ/HUOSHAN_BASE_URL
+      api_key: os.environ/HUOSHAN_API_KEY 
+      extra_body:
+        reasoning_effort: "minimal"        
+
+  - model_name: glm-4.7
+    litellm_params:
+      <<: *token_pricing
+      model: openai/glm-4-7-251222
+      api_base: os.environ/HUOSHAN_BASE_URL
+      api_key: os.environ/HUOSHAN_API_KEY 
+
+  - model_name: glm-4.7-nothink
+    litellm_params:
+      <<: *token_pricing
+      model: openai/glm-4-7-251222
+      api_base: os.environ/HUOSHAN_BASE_URL
+      api_key: os.environ/HUOSHAN_API_KEY 
+      extra_body:
+        thinking:
+          type: "disabled" 
+
+  #=======================fanyi
+
+  - model_name: test-newapi-dbao
+    litellm_params:
+      <<: *token_pricing
+      model: openai/doubao-seed-1-8-251228
+      api_base: os.environ/NEWAPI_BASE_URL
+      api_key: os.environ/NEWAPI_API_KEY
+
+  - model_name: test-ggllmm
+    litellm_params:
+      <<: *token_pricing
+      model: openai/glm-4-7-251222
+      api_base: os.environ/HUOSHAN_BASE_URL
+      api_key: os.environ/HUOSHAN_API_KEY 
+      extra_body:
+        thinking:
+          type: "disabled" 
+
+  - model_name: test-dbao
+    litellm_params:
+      <<: *token_pricing
+      model: openai/doubao-seed-1-8-251228
+      api_base: os.environ/HUOSHAN_BASE_URL
+      api_key: os.environ/HUOSHAN_API_KEY 
+      extra_body:
+        reasoning_effort: "minimal" 
+
+  - model_name: fanyi
+    litellm_params:
+      <<: *token_pricing
+      model: openai/doubao-seed-1-8-251228
+      api_base: os.environ/HUOSHAN_BASE_URL
+      api_key: os.environ/HUOSHAN_API_KEY 
+      extra_body:
+        reasoning_effort: "minimal"    
+
+          
+  - model_name: fanyi
+    litellm_params:
+      <<: *token_pricing
+      model: openai/glm-4-7-251222
+      api_base: os.environ/HUOSHAN_BASE_URL
+      api_key: os.environ/HUOSHAN_API_KEY 
+      extra_body:
+        thinking:
+          type: "disabled" 
+
+
+
+
+          
+  #=======================Claude系列
+
+  #=======================Claude-Opus-v4.5系列
+  - model_name: Claude-Opus-v4.5
+    litellm_params:
+      <<: *token_pricing
+      model: openai/claude-opus-4-5-thinking
+      api_base: os.environ/NEWAPI_BASE_URL
+      api_key: os.environ/NEWAPI_API_KEY
+
+  - model_name: Claude-Opus-v4.5
+    litellm_params:
+      <<: *token_pricing
+      model: openai/claude-opus-4-5
+      api_base: os.environ/NEWAPI_BASE_URL
+      api_key: os.environ/NEWAPI_API_KEY
+
+  - model_name: Claude-Opus-v4.5
+    litellm_params:
+      <<: *token_pricing
+      model: openai/claude-4.5-opus
+      api_base: os.environ/NEWAPI_BASE_URL
+      api_key: os.environ/NEWAPI_API_KEY
+
+  - model_name: Claude-Opus-v4.5
+    litellm_params:
+      <<: *token_pricing
+      model: openai/claude-opus-4-5-20251101
+      api_base: os.environ/NEWAPI_BASE_URL
+      api_key: os.environ/NEWAPI_API_KEY
+
+  - model_name: Claude-Opus-v4.5
+    litellm_params:
+      <<: *token_pricing
+      model: openai/cursor2-claude-4.1-opus
+      api_base: os.environ/NEWAPI_BASE_URL
+      api_key: os.environ/NEWAPI_API_KEY
+
+  #=======================Claude-Sonnet-v4.5系列
+  - model_name: Claude-Sonnet-v4.5
+    litellm_params:
+      <<: *token_pricing
+      model: openai/claude-sonnet-4-5-20250929
+      api_base: os.environ/NEWAPI_BASE_URL
+      api_key: os.environ/NEWAPI_API_KEY
+
+  - model_name: Claude-Sonnet-v4.5
+    litellm_params:
+      <<: *token_pricing
+      model: openai/claude-sonnet-4-5
+      api_base: os.environ/NEWAPI_BASE_URL
+      api_key: os.environ/NEWAPI_API_KEY
+
+  - model_name: Claude-Sonnet-v4.5
+    litellm_params:
+      <<: *token_pricing
+      model: openai/claude-4.5-sonnet-official-reverse
+      api_base: os.environ/NEWAPI_BASE_URL
+      api_key: os.environ/NEWAPI_API_KEY
+
+  - model_name: Claude-Sonnet-v4.5
+    litellm_params:
+      <<: *token_pricing
+      model: openai/claude-4.5-sonnet
+      api_base: os.environ/NEWAPI_BASE_URL
+      api_key: os.environ/NEWAPI_API_KEY
+
+  - model_name: Claude-Sonnet-v4.5
+    litellm_params:
+      <<: *token_pricing
+      model: openai/claude-4.5-sonnet-docs
+      api_base: os.environ/NEWAPI_BASE_URL
+      api_key: os.environ/NEWAPI_API_KEY
+
+  - model_name: Claude-Sonnet-v4.5
+    litellm_params:
+      <<: *token_pricing
+      model: openai/claude-4.5-sonnet-2cc
+      api_base: os.environ/NEWAPI_BASE_URL
+      api_key: os.environ/NEWAPI_API_KEY
+
+  - model_name: Claude-Sonnet-v4.5
+    litellm_params:
+      <<: *token_pricing
+      model: openai/claude-sonnet-4-5-thinking
+      api_base: os.environ/NEWAPI_BASE_URL
+      api_key: os.environ/NEWAPI_API_KEY
+
+  - model_name: Claude-Sonnet-v4.5
+    litellm_params:
+      <<: *token_pricing
+      model: openai/cursor2-claude-4.5-sonnet
+      api_base: os.environ/NEWAPI_BASE_URL
+      api_key: os.environ/NEWAPI_API_KEY
+  
+  
+  
+  
+  #=======================gpt-5系列
+  - model_name: gpt-5
+    litellm_params:
+      <<: *token_pricing
+      model: openai/cursor2-gpt-5
+      api_base: os.environ/NEWAPI_BASE_URL
+      api_key: os.environ/NEWAPI_API_KEY
+
+  - model_name: gpt-5
+    litellm_params:
+      <<: *token_pricing
+      model: openai/cursor2-gpt-5-codex
+      api_base: os.environ/NEWAPI_BASE_URL
+      api_key: os.environ/NEWAPI_API_KEY
+
+  - model_name: gpt-5
+    litellm_params:
+      <<: *token_pricing
+      model: openai/gpt-5.1
+      api_base: os.environ/NEWAPI_BASE_URL
+      api_key: os.environ/NEWAPI_API_KEY
+
+  - model_name: gpt-5
+    litellm_params:
+      <<: *token_pricing
+      model: openai/gpt-5.1-codex
+      api_base: os.environ/NEWAPI_BASE_URL
+      api_key: os.environ/NEWAPI_API_KEY
+
+  - model_name: gpt-5
+    litellm_params:
+      <<: *token_pricing
+      model: openai/gpt-5.1-codex-2cx
+      api_base: os.environ/NEWAPI_BASE_URL
+      api_key: os.environ/NEWAPI_API_KEY
+
+  - model_name: gpt-5
+    litellm_params:
+      <<: *token_pricing
+      model: openai/gpt-5.1-codex-max-2cx
+      api_base: os.environ/NEWAPI_BASE_URL
+      api_key: os.environ/NEWAPI_API_KEY
+
+  - model_name: gpt-5
+    litellm_params:
+      <<: *token_pricing
+      model: openai/gpt-5.2
+      api_base: os.environ/NEWAPI_BASE_URL
+      api_key: os.environ/NEWAPI_API_KEY
+
+  - model_name: gpt-5
+    litellm_params:
+      <<: *token_pricing
+      model: openai/gpt-5.2-2cx
+      api_base: os.environ/NEWAPI_BASE_URL
+      api_key: os.environ/NEWAPI_API_KEY
+
+
+
+
+litellm_settings:
+  num_retries: 2 # retry call 3 times on each model_name (e.g. zephyr-beta)
+  request_timeout: 120 # raise Timeout error if call takes longer than 10s. Sets litellm.request_timeout 
+  #fallbacks: [{"zephyr-beta": ["gpt-4o"]}] # fallback to gpt-4o if call fails num_retries 
+  #context_window_fallbacks: [{"zephyr-beta": ["gpt-3.5-turbo-16k"]}, {"gpt-4o": ["gpt-3.5-turbo-16k"]}] # fallback to gpt-3.5-turbo-16k if context window error
+  cache: False          # 开启缓存
+#  cache_params:
+#    type: "redis"      # 指定使用 Redis
+#    supported_call_types: ["completion", "embedding", "acompletion"] # 支持的类型  
+  allowed_fails: 3 # cooldown model if it fails > 1 call in a minute.
+  # 关键配置：注册自定义回调，实现单上游模型每日token限制的关键
+  callbacks: ["daily_limit.proxy_handler_instance"]
+
+router_settings: # router_settings are optional
+  routing_strategy: simple-shuffle # Literal["simple-shuffle", "least-busy", "usage-based-routing","latency-based-routing"], default="simple-shuffle"
+  #model_group_alias: {"gpt-4": "gpt-4o"} # all requests with `gpt-4` will be routed to models with `gpt-4o`
+  num_retries: 3
+  timeout: 300                                  # 30 seconds
+  #cooldown_time: 60
+  
+general_settings:
+  store_model_in_db: true
+  store_prompts_in_spend_logs: true  # <--- 这是最关键的一行
+  
+
+

daily_limit.py

@@ -0,0 +1,348 @@
+import os
+import time
+import hashlib
+import litellm
+from litellm.integrations.custom_logger import CustomLogger
+from litellm.exceptions import ServiceUnavailableError, ContextWindowExceededError
+import redis
+import traceback
+
+# ==========================================
+# 🛠️ 调试配置
+# ==========================================
+# 日志级别配置
+# 可选值: "DEBUG", "INFO", "WARNING", "ERROR"
+# 修改此变量以控制日志详细程度，或使用环境变量 DAILY_LIMIT_LOG_LEVEL
+LOG_LEVEL = os.getenv("DAILY_LIMIT_LOG_LEVEL", "INFO").upper()
+
+
+def get_log_level_value(level_name):
+    levels = {"DEBUG": 10, "INFO": 20, "WARNING": 30, "ERROR": 40}
+    return levels.get(level_name, 20)
+
+
+CURRENT_LEVEL_VAL = get_log_level_value(LOG_LEVEL)
+DEBUG_MODE = CURRENT_LEVEL_VAL <= 10
+
+
+def log(msg, level="INFO"):
+    msg_val = get_log_level_value(level)
+    if msg_val >= CURRENT_LEVEL_VAL:
+        icon = {"DEBUG": "🔍", "INFO": "ℹ️", "WARNING": "⚠️", "ERROR": "❌"}.get(level, "📝")
+        print(f"{icon} [DailyLimit] {msg}", flush=True)
+
+
+def dprint(msg):
+    log(msg, level="INFO")
+
+
+dprint("Loading daily_limit.py module v19 (Specific Quotas & Docker Track)...")
+
+# ==========================================
+# Redis 与 限额配置
+# ==========================================
+REDIS_HOST = os.getenv("REDIS_HOST", "litellm_redis")
+REDIS_PORT = os.getenv("REDIS_PORT", "6379")
+REDIS_PASSWORD = os.getenv("REDIS_PASSWORD", None)
+
+MODEL_QUOTAS = {
+    "deepseek-v3-2-251201:38430740": 1500000,
+    "doubao-seed-1-8-251228:38430740": 14000,  # 测试限额
+    "glm-4-7-251222:38430740": 1500000,
+    "doubao-seed-1-8-251228:9fbb42b8": 1500000,  # 测试重名
+}
+
+
+try:
+    redis_host = os.getenv("REDIS_HOST", "redis")
+    redis_port = int(os.getenv("REDIS_PORT", 6379))
+    redis_client = redis.Redis(host=redis_host, port=redis_port, db=0, decode_responses=True)
+    log(f"✅ Redis connected successfully to {redis_host}:{redis_port}", level="INFO")
+except Exception as e:
+    log(f"❌ Redis connection failed: {str(e)}", level="ERROR")
+    redis_client = None
+
+
+def clean_model_name(name):
+    if not name:
+        return ""
+    return name.replace("openai/", "").replace("azure/", "").replace("anthropic/", "")
+
+
+def generate_api_base_hash(api_base):
+    if not api_base:
+        return "default"
+        
+    # === 新增标准化代码 ===
+    # 1. 去除首尾空格 (strip)
+    # 2. 去除末尾的斜杠 (rstrip('/'))
+    # 这样 '.../v3/' 和 '.../v3' 会生成相同的 Hash
+    normalized_base = api_base.strip().rstrip('/')
+    # ====================        
+        
+    h = hashlib.md5(api_base.encode()).hexdigest()[:8]
+    if DEBUG_MODE:
+        # Debugging hash mismatch
+        print(f"🔍 [Hash Debug] Base='{api_base}' -> Hash='{h}'", flush=True)
+    return h
+
+
+# ==========================================
+# ⏰ 时间逻辑：北京时间 9:00 (UTC 01:00) 刷新
+# ==========================================
+def get_redis_key(clean_name, api_base=None):
+    now = time.time()
+    # UTC 01:00 视为新的一天，所以减去 1小时 (3600秒)
+    adjusted_time = now - 3600
+    date_str = time.strftime("%Y-%m-%d", time.gmtime(adjusted_time))
+
+    suffix = ""
+    if api_base:
+        suffix = f":{generate_api_base_hash(api_base)}"
+
+    return f"daily_usage:{date_str}:{clean_name}{suffix}"
+
+
+# ==========================================
+# 🎯 限额查找逻辑 (支持精确匹配)
+# ==========================================
+def get_daily_limit(clean_model, api_base):
+    # 1. 尝试精确匹配 (Model + Hash)
+    # 用户可以在 MODEL_QUOTAS 中配置 "model_name:hash": limit
+    current_hash = generate_api_base_hash(api_base)
+    specific_key = f"{clean_model}:{current_hash}"
+
+    limit = MODEL_QUOTAS.get(specific_key)
+    if limit:
+        return limit
+
+    # 2. 尝试精确模型名匹配
+    limit = MODEL_QUOTAS.get(clean_model)
+    if limit:
+        return limit
+
+    # 3. 尝试模糊模型名匹配 (向下兼容)
+    for quota_model, limit_val in MODEL_QUOTAS.items():
+        # 排除包含冒号的特定规则，防止误匹配
+        if ":" not in quota_model and quota_model in clean_model:
+            return limit_val
+
+    return None
+
+
+# ==========================================
+# 👻 核心逻辑：软更名踢出 & 自动恢复
+# ==========================================
+def soft_evict_from_router(target_upstream_model, target_api_base=None):
+    """
+    不删除节点，而是修改节点的 model_name，让 Router 找不到它。
+    同时备份原始信息以便恢复。
+    """
+    try:
+        from litellm.proxy.proxy_server import llm_router
+
+        if not llm_router or not llm_router.model_list:
+            return
+
+        evicted = False
+        target_upstream_clean = clean_model_name(target_upstream_model)
+
+        # Iterate over the actual list in memory
+        for i, model_entry in enumerate(llm_router.model_list):
+            params = model_entry.get("litellm_params", {})
+            upstream_model = params.get("model", "")
+            current_api_base = params.get("api_base", "")
+
+            # 匹配上游模型名称
+            if target_upstream_clean not in clean_model_name(upstream_model):
+                continue
+
+            # Strict api_base matching if provided
+            if target_api_base and current_api_base != target_api_base:
+                continue
+
+            current_alias = model_entry.get("model_name")
+
+            # 如果它还没有被标记为 blocked
+            if current_alias and "blocked" not in current_alias:
+                log(
+                    f"👻 [ROUTER HIDE] Renaming alias '{current_alias}' -> '{current_alias}-blocked' for {target_upstream_model}",
+                    level="INFO",
+                )
+
+                # Backup original state
+                model_entry["_daily_limit_backup"] = {"model_name": current_alias, "api_base": current_api_base}
+
+                # 1. 修改别名 - Modify IN PLACE
+                llm_router.model_list[i]["model_name"] = f"{current_alias}-blocked"
+
+                # 2. 破坏连接
+                llm_router.model_list[i]["litellm_params"]["api_base"] = "http://0.0.0.0"
+
+                evicted = True
+
+        if evicted:
+            # Re-initialize router to reflect changes immediately if possible
+            # But changing model_list in place is usually enough for next lookup
+            dprint("✅ Node effectively removed from routing pool.")
+
+    except Exception as e:
+        print(f"❌ Router eviction failed: {e}", flush=True)
+
+
+last_restore_check = 0
+RESTORE_CHECK_INTERVAL = 60  # Check every 60 seconds
+
+
+def restore_blocked_models():
+    """
+    检查是否有 blocked 的模型配额已重置，如果是则恢复。
+    """
+    global last_restore_check
+    now = time.time()
+    if now - last_restore_check < RESTORE_CHECK_INTERVAL:
+        return
+
+    last_restore_check = now
+
+    try:
+        from litellm.proxy.proxy_server import llm_router
+
+        if not llm_router or not llm_router.model_list:
+            return
+
+        restored_count = 0
+        for i, model_entry in enumerate(llm_router.model_list):
+            backup = model_entry.get("_daily_limit_backup")
+            if not backup:
+                continue
+
+            # 获取必要信息检查配额
+            params = model_entry.get("litellm_params", {})
+            upstream_model = params.get("model", "")
+            clean_model = clean_model_name(upstream_model)
+            original_api_base = backup.get("api_base")
+
+            # Check quota
+            daily_limit = get_daily_limit(clean_model, original_api_base)
+
+            should_restore = False
+            if not daily_limit:
+                should_restore = True
+            else:
+                key = get_redis_key(clean_model, original_api_base)
+                current_usage = redis_client.get(key)
+                current_val = float(current_usage) if current_usage else 0.0
+
+                if current_val < float(daily_limit):
+                    should_restore = True
+
+            if should_restore:
+                dprint(f"♻️ [RESTORE] Restoring {backup['model_name']} (Quota OK)")
+
+                # Restore IN PLACE
+                llm_router.model_list[i]["model_name"] = backup["model_name"]
+                llm_router.model_list[i]["litellm_params"]["api_base"] = original_api_base
+                del llm_router.model_list[i]["_daily_limit_backup"]
+                restored_count += 1
+
+        if restored_count > 0:
+            dprint(f"✅ Restored {restored_count} models to routing pool.")
+
+    except Exception as e:
+        log(f"❌ Restore check failed: {e}", level="ERROR")
+
+
+# ==========================================
+# Input Callback
+# ==========================================
+def quota_interceptor(kwargs, user_api_key_dict=None):
+    """
+    Standard LiteLLM input callback signature:
+    def input_callback(kwargs, user_api_key_dict):
+    """
+    try:
+        restore_blocked_models()
+        # check_and_hide(kwargs, source="input_callback")  # Removed to avoid rigid blocking
+    except (ServiceUnavailableError, ContextWindowExceededError):
+        raise
+    except Exception:
+        # Don't let other errors block the request
+        pass
+
+
+# ==========================================
+# Handler & Logging
+# ==========================================
+class DailyLimitHandler(CustomLogger):
+    def __init__(self):
+        # Register the callback properly
+        if quota_interceptor not in litellm.input_callback:
+            litellm.input_callback.append(quota_interceptor)
+            dprint("✅ quota_interceptor registered")
+        if self not in litellm.callbacks:
+            litellm.callbacks.append(self)
+            dprint("✅ DailyLimitHandler registered")
+
+    # 统计逻辑
+    async def async_log_success_event(self, kwargs, response_obj, start_time, end_time):
+        self._record_usage(kwargs, response_obj)
+
+    def log_success_event(self, kwargs, response_obj, start_time, end_time):
+        self._record_usage(kwargs, response_obj)
+
+    def _record_usage(self, kwargs, response_obj):
+        if not redis_client:
+            return
+        try:
+            litellm_params = kwargs.get("litellm_params") or {}
+            raw_model = litellm_params.get("model") or kwargs.get("model")
+            clean_model = clean_model_name(raw_model)
+            api_base = litellm_params.get("api_base")
+
+            usage = None
+            if hasattr(response_obj, "usage"):
+                usage = response_obj.usage
+            elif isinstance(response_obj, dict):
+                usage = response_obj.get("usage")
+            if hasattr(usage, "model_dump"):
+                usage = usage.model_dump()
+            elif hasattr(usage, "dict"):
+                usage = usage.dict()
+
+            if not usage:
+                return
+
+            total_tokens = usage.get("total_tokens", 0)
+            if clean_model and total_tokens > 0:
+                key = get_redis_key(clean_model, api_base)
+                current_total = redis_client.incrby(key, int(total_tokens))
+                redis_client.expire(key, 172800)
+
+                # Get Limit for display
+                daily_limit = get_daily_limit(clean_model, api_base)
+                display_limit = f"/{int(daily_limit):,}" if daily_limit else ""
+
+                # Show usage if DEBUG level is enabled
+                key_display = key.split(":", 2)[-1]
+                log(
+                    f"📊 [Usage] {key_display}: +{total_tokens} (Today: {current_total:,}{display_limit})",
+                    level="DEBUG",
+                )
+
+                # Enforcement Logic
+                if daily_limit and current_total >= float(daily_limit):
+                    log(f"🚫 [BLOCK] {clean_model} Limit: {current_total}/{daily_limit}", level="WARNING")
+                    soft_evict_from_router(raw_model, api_base)
+            else:
+                if DEBUG_MODE:
+                    if not clean_model:
+                        log("⚠️ [Usage Skip] No clean model name found", level="DEBUG")
+                    if total_tokens <= 0:
+                        log(f"⚠️ [Usage Skip] Total tokens is {total_tokens}", level="DEBUG")
+
+        except Exception as e:
+            log(f"❌ _record_usage failed: {e}\n{traceback.format_exc()}", level="ERROR")
+
+
+proxy_handler_instance = DailyLimitHandler()

docker-compose.yml

@@ -0,0 +1,77 @@
+services:
+  litellm:
+    image: ghcr.io/berriai/litellm:main-latest
+    container_name: litellm
+    restart: unless-stopped
+    volumes:
+      - ./config.yaml:/app/config.yaml
+      # 【关键】挂载自定义脚本到容器根目录
+      - ./daily_limit.py:/app/daily_limit.py      
+    command:
+      - "--config=/app/config.yaml"
+    ports:
+      - "23003:4000"
+    environment:
+      # ⚠️ 必须设置这个环境变量，否则 Python 找不到你的脚本
+      PYTHONPATH: "/app"    
+      DATABASE_URL: "postgresql://llmproxy:dbpassword9090@db:5432/litellm"
+      STORE_MODEL_IN_DB: "True"
+      # --- Modified: 添加 Redis 配置 ---
+      REDIS_HOST: "redis"       # 对应下方 redis 服务的名字
+      REDIS_PORT: "6379"
+      # -------------------------------
+    env_file:
+      - .env
+    depends_on:
+      - db
+      - redis  # --- New: 依赖 redis 启动 ---
+    healthcheck:
+      test:
+        - CMD-SHELL
+        - python3 -c "import urllib.request; urllib.request.urlopen('http://localhost:4000/health/liveliness')"
+      interval: 300s
+      timeout: 10s
+      retries: 3
+      start_period: 40s
+
+  # --- New: 添加 Redis 服务 ---
+  redis:
+    image: redis:alpine       # 使用 alpine 版本，体积非常小
+    container_name: litellm_redis
+    restart: unless-stopped
+#     --maxmemory 512mb : 限制 Redis 最多使用 512MB 内存
+#     --maxmemory-policy allkeys-lru : 内存满了之后，自动删除最久未使用的缓存（无论是否过期）
+#     --save "" : 如果你不需要持久化缓存到硬盘（重启后清空），可以把 save 设为空，减少磁盘IO
+#    command: redis-server --save 60 1 --maxmemory 128mb --maxmemory-policy allkeys-lru --loglevel warning
+    command: redis-server --maxmemory 32mb --maxmemory-policy allkeys-lru --save 300 1 --loglevel warning
+    # 配置详解：
+    # --maxmemory 32mb: 严格限制内存使用 32MB
+    # --maxmemory-policy allkeys-lru: 内存满时挤掉旧数据（安全兜底）
+    # --save 300 1: 每300秒(5分钟)如果至少有1个key变化，就保存快照到磁盘 (降低IO)    
+    volumes: 
+    # 映射数据目录，确保重启容器后，今天的计数还在
+      - redis_data:/data
+  # ---------------------------
+
+  db:
+    image: postgres:16
+    restart: unless-stopped
+    container_name: litellm_db
+    environment:
+      POSTGRES_DB: litellm
+      POSTGRES_USER: llmproxy
+      POSTGRES_PASSWORD: dbpassword9090
+    volumes:
+      - postgres_data:/var/lib/postgresql/data
+    healthcheck:
+      test: ["CMD-SHELL", "pg_isready -d litellm -U llmproxy"]
+      interval: 5s
+      timeout: 5s
+      retries: 10
+
+volumes:
+  postgres_data:
+    driver: local
+    name: litellm_postgres_data
+  redis_data:  # 如果开启了 redis 持久化，这里也要打开
+    driver: local

generate_keys.py

@@ -0,0 +1,153 @@
+import os
+import yaml
+import hashlib
+import re
+from typing import Dict, Any
+
+# ==========================================
+# 🛠️ 配置
+# ==========================================
+CONFIG_FILE = "config.yaml"
+ENV_FILE = ".env"
+
+
+def load_env(env_path: str) -> Dict[str, str]:
+    """读取 .env 文件并返回字典"""
+    env_vars = {}
+    if not os.path.exists(env_path):
+        print(f"⚠️ Warning: {env_path} not found. Env var substitution might fail.")
+        return env_vars
+
+    with open(env_path, "r", encoding="utf-8") as f:
+        for line in f:
+            line = line.strip()
+            if not line or line.startswith("#"):
+                continue
+            try:
+                key, value = line.split("=", 1)
+                env_vars[key.strip()] = value.strip().strip('"').strip("'")
+            except ValueError:
+                continue
+    return env_vars
+
+
+def resolve_env_var(value: str, env_vars: Dict[str, str]) -> str:
+    """
+    处理环境变量替换，支持以下两种格式：
+    1. ${VAR_NAME} - 标准格式
+    2. os.environ/VAR_NAME - LiteLLM 特有格式
+    """
+    if not isinstance(value, str):
+        return value
+
+    # 1. 处理 os.environ/VAR_NAME 格式
+    if value.startswith("os.environ/"):
+        var_name = value.replace("os.environ/", "")
+        return env_vars.get(var_name, os.getenv(var_name, value))
+
+    # 2. 处理 ${VAR_NAME} 或 $VAR_NAME 格式
+    pattern = re.compile(r"\$\{?([A-Z0-9_]+)\}?")
+
+    def replace_match(match):
+        var_name = match.group(1)
+        # 优先从 .env 读取，其次从系统环境变量读取
+        return env_vars.get(var_name, os.getenv(var_name, match.group(0)))
+
+    return pattern.sub(replace_match, value)
+
+
+def clean_model_name(name):
+    if not name:
+        return ""
+    return name.replace("openai/", "").replace("azure/", "").replace("anthropic/", "")
+
+
+def generate_api_base_hash(api_base):
+    if not api_base:
+        return "default"
+    # Logic matches daily_limit.py v18
+    
+    # === 新增标准化代码 ===
+    # 1. 去除首尾空格 (strip)
+    # 2. 去除末尾的斜杠 (rstrip('/'))
+    # 这样 '.../v3/' 和 '.../v3' 会生成相同的 Hash
+    normalized_base = api_base.strip().rstrip('/')
+    # ====================
+
+    
+    return hashlib.md5(api_base.encode()).hexdigest()[:8]
+
+
+def main():
+    print("🔍 Generating Daily Limit Keys from config.yaml...\n")
+
+    # 1. Load Env
+    env_vars = load_env(ENV_FILE)
+
+    # 2. Load Config
+    if not os.path.exists(CONFIG_FILE):
+        print(f"❌ Error: {CONFIG_FILE} not found!")
+        return
+
+    try:
+        with open(CONFIG_FILE, "r", encoding="utf-8") as f:
+            config = yaml.safe_load(f)
+    except Exception as e:
+        print(f"❌ Error parsing yaml: {e}")
+        return
+
+    model_list = config.get("model_list", [])
+    if not model_list:
+        print("⚠️ No models found in config.yaml")
+        return
+
+    results = []
+    seen_keys = set()
+
+    # Table Header
+    # Model Alias | Upstream Model | API Base (Env/URL) | Generated Key
+    header = f"{'Model Name (Alias)':<25} | {'Upstream Model':<30} | {'API Base (Env/URL)':<40} | {'Generated Key'}"
+    print(header)
+    print("-" * len(header))
+
+    for model in model_list:
+        litellm_params = model.get("litellm_params", {})
+
+        # Get raw values
+        model_alias = model.get("model_name", "unknown")
+        upstream_model = litellm_params.get("model", "")
+        api_base_raw = litellm_params.get("api_base", "")
+
+        # Resolve env vars for hash generation
+        api_base_resolved = resolve_env_var(api_base_raw, env_vars)
+
+        # Generate components
+        clean_upstream = clean_model_name(upstream_model)
+        base_hash = generate_api_base_hash(api_base_resolved)
+
+        # Generate final key
+        final_key = f"{clean_upstream}:{base_hash}"
+
+        # Deduplicate
+        if final_key in seen_keys:
+            continue
+        seen_keys.add(final_key)
+
+        # Truncate api_base_raw if it's too long for display
+        display_api_base = api_base_raw
+        if len(display_api_base) > 37:
+            display_api_base = display_api_base[:34] + "..."
+
+        print(f"{model_alias:<25} | {clean_upstream:<30} | {display_api_base:<40} | {final_key}")
+        results.append(f'"{final_key}": 1500000')
+
+    print("\n📋 [Copy & Paste into daily_limit.py MODEL_QUOTAS]:")
+    print("-" * 50)
+    print("MODEL_QUOTAS = {")
+    print(",\n".join(f"    {item}" for item in results))
+    print("}")
+    print("-" * 50)
+
+
+if __name__ == "__main__":
+    main()