chore: revert changes to daily_limit.py as requested

daily_limit.py

@@ -29,9 +29,7 @@ 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(
+        icon = {"DEBUG": "🔍", "INFO": "ℹ️", "WARNING": "⚠️", "ERROR": "❌"}.get(level, "📝")
-            level, "📝"
-        )
         print(f"{icon} [DailyLimit] {msg}", flush=True)
 
 
@@ -59,9 +57,7 @@ MODEL_QUOTAS = {
 try:
     redis_host = os.getenv("REDIS_HOST", "redis")
     redis_port = int(os.getenv("REDIS_PORT", 6379))
-    redis_client = redis.Redis(
+    redis_client = redis.Redis(host=redis_host, port=redis_port, db=0, decode_responses=True)
-        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")
@@ -88,10 +84,10 @@ def generate_api_base_hash(api_base):
     try:
         # 1. 预处理：去空格
         s = api_base.strip()
-
+        
         # 2. 解析 URL
         parsed = urlparse(s)
-
+        
         # 3. 提取 Scheme 和 Netloc (域名+端口)
         # 如果解析成功且包含 scheme 和 netloc，则重组
         if parsed.scheme and parsed.netloc:
@@ -99,23 +95,20 @@ def generate_api_base_hash(api_base):
         else:
             # 兜底：如果不是标准 URL，或者解析失败，尝试简单的去斜杠和小写化
             # 这种情况比较少见，通常 litellm 会校验 URL 格式
-            clean_base = s.rstrip("/").lower()
+            clean_base = s.rstrip('/').lower()
-
+            
     except Exception:
         # 极端情况下的兜底
-        clean_base = api_base.strip().rstrip("/").lower()
+        clean_base = api_base.strip().rstrip('/').lower()
 
     # 4. 计算 Hash
     h = hashlib.md5(clean_base.encode()).hexdigest()[:8]
-
+    
     if DEBUG_MODE:
         # Debugging hash mismatch
         # 打印原始值和清洗后的值，方便确认
-        print(
+        print(f"🔍 [Hash Debug] Base='{api_base}' -> Clean='{clean_base}' -> Hash='{h}'", flush=True)
-            f"🔍 [Hash Debug] Base='{api_base}' -> Clean='{clean_base}' -> Hash='{h}'",
+        
-            flush=True,
-        )
-
     return h
 
 
@@ -139,74 +132,6 @@ def get_redis_key(clean_name, api_base=None):
 # 🎯 限额查找逻辑 (支持精确匹配)
 # ==========================================
 def get_daily_limit(clean_model, api_base):
-    # 0. Check Dynamic Limit (Redis Priority) for THIS model
-    # Logic: Limit = Current_Usage + Remaining_Balance_After_Buffer
-    if redis_client:
-        try:
-            # Check specific model limit first (from volc_limit_sync breakdown)
-            # Key: limit:dynamic:volcengine:remaining:{clean_model}
-            # Example: deepseek-v3
-            # Fallback to global if needed (though sync script now sets per-model)
-
-            # Since normalized names in sync script match clean_model (ideally),
-            # we try to find a matching key.
-            # But wait, clean_model in daily_limit might have -251201 suffix,
-            # while sync script key might be just "deepseek-v3".
-            # We need to iterate or check fuzzy?
-            # Actually, sync script sets keys based on mapped names.
-            # Let's rely on the fuzzy match logic we used in syncing?
-            # No, sync script sets specific keys.
-
-            # Let's try to find a key that is a substring of clean_model
-            # Or just check if there is a key.
-            # For simplicity and performance, we can iterate over the keys set by sync script?
-            # No, that's slow.
-
-            # Better approach:
-            # The sync script sets keys like "limit:dynamic:volcengine:remaining:deepseek-v3"
-            # We check if we can construct that key.
-            # clean_model is "deepseek-v3-2-251201"
-
-            # Let's use the scan match approach? No, slow.
-            # Let's try direct lookups for likely prefixes.
-
-            # Actually, if we use the same VOLC_MODEL_MAP logic here? No, code duplication.
-
-            # Pragmatic approach:
-            # Iterate over known dynamic keys?
-            # Or, just check "limit:dynamic:volcengine:remaining:global" if we want a catch-all?
-            # But user wants per-model.
-
-            # Let's try to match the keys that exist.
-            # Sync script sets keys. We can just list them once or use a known pattern.
-            # But we don't know the exact key name here easily without the map.
-
-            # Wait, the user said "ensure alignment".
-            # If VOLC_MODEL_MAP maps "deepseek-v3" -> "deepseek-v3-241226",
-            # then sync script sets key "limit:dynamic:volcengine:remaining:deepseek-v3-241226"
-            # And clean_model IS "deepseek-v3-241226".
-            # So it SHOULD be an exact match!
-
-            remaining_key = f"limit:dynamic:volcengine:remaining:{clean_model}"
-            remaining_balance = redis_client.get(remaining_key)
-
-            # If not found, maybe it's a prefix match case?
-            if remaining_balance is None:
-                # Try finding a key that is a prefix of clean_model
-                # This is tricky without listing all keys.
-                # For now, let's assume the alignment from Task 2 (Mapping) works.
-                pass
-
-            if remaining_balance is not None:
-                usage_key = get_redis_key(clean_model, api_base)
-                current_usage = redis_client.get(usage_key)
-                current_val = float(current_usage) if current_usage else 0.0
-                dynamic_limit = current_val + float(remaining_balance)
-                return dynamic_limit
-
-        except Exception:
-            pass
-
     # 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)
@@ -272,18 +197,13 @@ def soft_evict_from_router(target_upstream_model, target_api_base=None):
                 )
 
                 # Backup original state
-                model_entry["_daily_limit_backup"] = {
+                model_entry["_daily_limit_backup"] = {"model_name": current_alias, "api_base": current_api_base}
-                    "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"] = (
+                llm_router.model_list[i]["litellm_params"]["api_base"] = "http://0.0.0.0"
-                    "http://0.0.0.0"
-                )
 
                 evicted = True
 
@@ -348,9 +268,7 @@ def restore_blocked_models():
 
                 # Restore IN PLACE
                 llm_router.model_list[i]["model_name"] = backup["model_name"]
-                llm_router.model_list[i]["litellm_params"]["api_base"] = (
+                llm_router.model_list[i]["litellm_params"]["api_base"] = original_api_base
-                    original_api_base
-                )
                 del llm_router.model_list[i]["_daily_limit_backup"]
                 restored_count += 1
 
@@ -427,22 +345,6 @@ class DailyLimitHandler(CustomLogger):
                 current_total = redis_client.incrby(key, int(total_tokens))
                 redis_client.expire(key, 172800)
 
-                # Increment global counter for Volcengine sync
-                try:
-                    # Ensure date_str is defined for the global key
-                    now_ts_global = time.time()
-                    # Adjust for timezone if needed, or consistent with existing date_str logic
-                    # Assuming local time is acceptable for sync triggers
-                    date_str_global = time.strftime(
-                        "%Y-%m-%d", time.localtime(now_ts_global)
-                    )
-
-                    global_key = f"daily_usage:{date_str_global}:global_total"
-                    redis_client.incrby(global_key, int(total_tokens))
-                    redis_client.expire(global_key, 172800)  # 48h
-                except Exception:
-                    pass
-
                 # Get Limit for display
                 daily_limit = get_daily_limit(clean_model, api_base)
                 display_limit = f"/{int(daily_limit):,}" if daily_limit else ""
@@ -456,25 +358,17 @@ class DailyLimitHandler(CustomLogger):
 
                 # Enforcement Logic
                 if daily_limit and current_total >= float(daily_limit):
-                    log(
+                    log(f"🚫 [BLOCK] {clean_model} Limit: {current_total}/{daily_limit}", level="WARNING")
-                        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(
+                        log(f"⚠️ [Usage Skip] Total tokens is {total_tokens}", level="DEBUG")
-                            f"⚠️ [Usage Skip] Total tokens is {total_tokens}",
-                            level="DEBUG",
-                        )
 
         except Exception as e:
-            log(
+            log(f"❌ _record_usage failed: {e}\n{traceback.format_exc()}", level="ERROR")
-                f"❌ _record_usage failed: {e}\n{traceback.format_exc()}", level="ERROR"
-            )
 
 
 proxy_handler_instance = DailyLimitHandler()