Merge pull request #1604 from BerriAI/litellm_view_cache_key_in_spend_logs

[Feat] Add cache_key in SpendLogs Table

litellm/proxy/_types.py

@@ -343,8 +343,7 @@ class LiteLLM_SpendLogs(LiteLLMBase):
     endTime: Union[str, datetime, None]
     user: Optional[str] = ""
     modelParameters: Optional[Json] = {}
-    messages: Optional[Json] = []
-    response: Optional[Json] = {}
     usage: Optional[Json] = {}
     metadata: Optional[Json] = {}
     cache_hit: Optional[str] = "False"
+    cache_key: Optional[str] = None

litellm/proxy/proxy_server.py

@@ -593,6 +593,12 @@ async def track_cost_callback(
                 "user_api_key_user_id", None
             )
 
+            if kwargs.get("cache_hit", False) == True:
+                response_cost = 0.0
+                verbose_proxy_logger.info(
+                    f"Cache Hit: response_cost {response_cost}, for user_id {user_id}"
+                )
+
             verbose_proxy_logger.info(
                 f"response_cost {response_cost}, for user_id {user_id}"
             )
@@ -1382,8 +1388,6 @@ async def initialize(
                 verbose_proxy_logger.setLevel(
                     level=logging.DEBUG
                 )  # set proxy logs to debug
-                litellm.set_verbose = True
-
     dynamic_config = {"general": {}, user_model: {}}
     if config:
         (

litellm/proxy/schema.prisma

@@ -58,4 +58,5 @@ model LiteLLM_SpendLogs {
   usage               Json     @default("{}")
   metadata            Json     @default("{}")
   cache_hit           String   @default("")
+  cache_key           String   @default("")
 }

litellm/proxy/utils.py

@@ -995,11 +995,18 @@ def get_logging_payload(kwargs, response_obj, start_time, end_time):
     if api_key is not None and isinstance(api_key, str) and api_key.startswith("sk-"):
         # hash the api_key
         api_key = hash_token(api_key)
-
     if "headers" in metadata and "authorization" in metadata["headers"]:
         metadata["headers"].pop(
             "authorization"
         )  # do not store the original `sk-..` api key in the db
+    if litellm.cache is not None:
+        cache_key = litellm.cache.get_cache_key(**kwargs)
+    else:
+        cache_key = "Cache OFF"
+    if cache_hit == True:
+        import time
+
+        id = f"{id}_cache_hit{time.time()}"  # SpendLogs does not allow duplicate request_id
 
     payload = {
         "request_id": id,
@@ -1013,6 +1020,7 @@ def get_logging_payload(kwargs, response_obj, start_time, end_time):
         "modelParameters": optional_params,
         "usage": usage,
         "metadata": metadata,
+        "cache_key": cache_key,
     }
 
     json_fields = [

litellm/tests/test_key_generate_prisma.py

@@ -716,6 +716,9 @@ def test_call_with_key_over_budget(prisma_client):
             # update spend using track_cost callback, make 2nd request, it should fail
             from litellm.proxy.proxy_server import track_cost_callback
             from litellm import ModelResponse, Choices, Message, Usage
+            from litellm.caching import Cache
+
+            litellm.cache = Cache()
             import time
 
             request_id = f"chatcmpl-e41836bb-bb8b-4df2-8e70-8f3e160155ac{time.time()}"
@@ -763,6 +766,10 @@ def test_call_with_key_over_budget(prisma_client):
             assert spend_log.request_id == request_id
             assert spend_log.spend == float("2e-05")
             assert spend_log.model == "chatgpt-v-2"
+            assert (
+                spend_log.cache_key
+                == "a61ae14fe4a8b8014a61e6ae01a100c8bc6770ac37c293242afed954bc69207d"
+            )
 
             # use generated key to auth in
             result = await user_api_key_auth(request=request, api_key=bearer_token)

schema.prisma

@@ -61,4 +61,5 @@ model LiteLLM_SpendLogs {
   usage               Json     @default("{}")
   metadata            Json     @default("{}")
   cache_hit           String   @default("")
+  cache_key           String   @default("")
 }