(feat) - provider budget improvements - ensure provider budgets work …

…with multiple proxy instances + improve latency to ~90ms (#6886)

* use 1 file for duration_in_seconds

* add to readme.md

* re use duration_in_seconds

* fix importing _extract_from_regex, get_last_day_of_month

* fix import

* update provider budget routing

* fix - remove dup test

* add support for using in multi instance environments

* test_in_memory_redis_sync_e2e

* test_in_memory_redis_sync_e2e

* fix test_in_memory_redis_sync_e2e

* fix code quality check

* fix test provider budgets

* working provider budget tests

* add fixture for provider budget routing

* fix router testing for provider budgets

* add comments on provider budget routing

* use RedisPipelineIncrementOperation

* add redis async_increment_pipeline

* use redis async_increment_pipeline

* use lower value for testing

* use redis async_increment_pipeline

* use consistent key name for increment op

* add handling for budget windows

* fix typing async_increment_pipeline

* fix set attr

* add clear doc strings

* unit testing for provider budgets

* test_redis_increment_pipeline

docs/my-website/docs/proxy/provider_budget_routing.md

@@ -16,25 +16,27 @@ model_list:
         api_key: os.environ/OPENAI_API_KEY
 
 router_settings:
-  redis_host: <your-redis-host>
-  redis_password: <your-redis-password>
-  redis_port: <your-redis-port>
   provider_budget_config: 
-	openai: 
-		budget_limit: 0.000000000001 # float of $ value budget for time period
-		time_period: 1d # can be 1d, 2d, 30d, 1mo, 2mo
-	azure:
-		budget_limit: 100
-		time_period: 1d
-	anthropic:
-		budget_limit: 100
-		time_period: 10d
-	vertex_ai:
-		budget_limit: 100
-		time_period: 12d
-	gemini:
-		budget_limit: 100
-		time_period: 12d
+    openai: 
+      budget_limit: 0.000000000001 # float of $ value budget for time period
+      time_period: 1d # can be 1d, 2d, 30d, 1mo, 2mo
+    azure:
+      budget_limit: 100
+      time_period: 1d
+    anthropic:
+      budget_limit: 100
+      time_period: 10d
+    vertex_ai:
+      budget_limit: 100
+      time_period: 12d
+    gemini:
+      budget_limit: 100
+      time_period: 12d
+
+  # OPTIONAL: Set Redis Host, Port, and Password if using multiple instance of LiteLLM
+  redis_host: os.environ/REDIS_HOST
+  redis_port: os.environ/REDIS_PORT
+  redis_password: os.environ/REDIS_PASSWORD
 
 general_settings:
   master_key: sk-1234
@@ -132,6 +134,31 @@ This metric indicates the remaining budget for a provider in dollars (USD)
 litellm_provider_remaining_budget_metric{api_provider="openai"} 10
 ```
 
+## Multi-instance setup
+
+If you are using a multi-instance setup, you will need to set the Redis host, port, and password in the `proxy_config.yaml` file. Redis is used to sync the spend across LiteLLM instances.
+
+```yaml
+model_list:
+    - model_name: gpt-3.5-turbo
+      litellm_params:
+        model: openai/gpt-3.5-turbo
+        api_key: os.environ/OPENAI_API_KEY
+
+router_settings:
+  provider_budget_config: 
+    openai: 
+      budget_limit: 0.000000000001 # float of $ value budget for time period
+      time_period: 1d # can be 1d, 2d, 30d, 1mo, 2mo
+  
+  # 👇 Add this: Set Redis Host, Port, and Password if using multiple instance of LiteLLM
+  redis_host: os.environ/REDIS_HOST
+  redis_port: os.environ/REDIS_PORT
+  redis_password: os.environ/REDIS_PASSWORD
+
+general_settings:
+  master_key: sk-1234
+```
 
 ## Spec for provider_budget_config
 

litellm/caching/redis_cache.py

@@ -20,6 +20,7 @@
 import litellm
 from litellm._logging import print_verbose, verbose_logger
 from litellm.litellm_core_utils.core_helpers import _get_parent_otel_span_from_kwargs
+from litellm.types.caching import RedisPipelineIncrementOperation
 from litellm.types.services import ServiceLoggerPayload, ServiceTypes
 from litellm.types.utils import all_litellm_params
 
@@ -890,3 +891,92 @@ async def async_delete_cache(self, key: str):
 
     def delete_cache(self, key):
         self.redis_client.delete(key)
+
+    async def _pipeline_increment_helper(
+        self,
+        pipe: pipeline,
+        increment_list: List[RedisPipelineIncrementOperation],
+    ) -> Optional[List[float]]:
+        """Helper function for pipeline increment operations"""
+        # Iterate through each increment operation and add commands to pipeline
+        for increment_op in increment_list:
+            cache_key = self.check_and_fix_namespace(key=increment_op["key"])
+            print_verbose(
+                f"Increment ASYNC Redis Cache PIPELINE: key: {cache_key}\nValue {increment_op['increment_value']}\nttl={increment_op['ttl']}"
+            )
+            pipe.incrbyfloat(cache_key, increment_op["increment_value"])
+            if increment_op["ttl"] is not None:
+                _td = timedelta(seconds=increment_op["ttl"])
+                pipe.expire(cache_key, _td)
+        # Execute the pipeline and return results
+        results = await pipe.execute()
+        print_verbose(f"Increment ASYNC Redis Cache PIPELINE: results: {results}")
+        return results
+
+    async def async_increment_pipeline(
+        self, increment_list: List[RedisPipelineIncrementOperation], **kwargs
+    ) -> Optional[List[float]]:
+        """
+        Use Redis Pipelines for bulk increment operations
+        Args:
+            increment_list: List of RedisPipelineIncrementOperation dicts containing:
+                - key: str
+                - increment_value: float
+                - ttl_seconds: int
+        """
+        # don't waste a network request if there's nothing to increment
+        if len(increment_list) == 0:
+            return None
+
+        from redis.asyncio import Redis
+
+        _redis_client: Redis = self.init_async_client()  # type: ignore
+        start_time = time.time()
+
+        print_verbose(
+            f"Increment Async Redis Cache Pipeline: increment list: {increment_list}"
+        )
+
+        try:
+            async with _redis_client as redis_client:
+                async with redis_client.pipeline(transaction=True) as pipe:
+                    results = await self._pipeline_increment_helper(
+                        pipe, increment_list
+                    )
+
+            print_verbose(f"pipeline increment results: {results}")
+
+            ## LOGGING ##
+            end_time = time.time()
+            _duration = end_time - start_time
+            asyncio.create_task(
+                self.service_logger_obj.async_service_success_hook(
+                    service=ServiceTypes.REDIS,
+                    duration=_duration,
+                    call_type="async_increment_pipeline",
+                    start_time=start_time,
+                    end_time=end_time,
+                    parent_otel_span=_get_parent_otel_span_from_kwargs(kwargs),
+                )
+            )
+            return results
+        except Exception as e:
+            ## LOGGING ##
+            end_time = time.time()
+            _duration = end_time - start_time
+            asyncio.create_task(
+                self.service_logger_obj.async_service_failure_hook(
+                    service=ServiceTypes.REDIS,
+                    duration=_duration,
+                    error=e,
+                    call_type="async_increment_pipeline",
+                    start_time=start_time,
+                    end_time=end_time,
+                    parent_otel_span=_get_parent_otel_span_from_kwargs(kwargs),
+                )
+            )
+            verbose_logger.error(
+                "LiteLLM Redis Caching: async increment_pipeline() - Got exception from REDIS %s",
+                str(e),
+            )
+            raise e

litellm/proxy/proxy_config.yaml

@@ -2,8 +2,23 @@ model_list:
   - model_name: gpt-4o
     litellm_params:
       model: openai/gpt-4o
-      api_key: os.environ/OPENAI_API_KEY
+      api_base: https://exampleopenaiendpoint-production.up.railway.app/
+  - model_name: fake-anthropic-endpoint
+    litellm_params:
+      model: anthropic/fake
+      api_base: https://exampleanthropicendpoint-production.up.railway.app/
+
+router_settings:
+  provider_budget_config: 
+    openai: 
+      budget_limit: 0.3 # float of $ value budget for time period
+      time_period: 1d # can be 1d, 2d, 30d 
+    anthropic:
+      budget_limit: 5
+      time_period: 1d
+  redis_host: os.environ/REDIS_HOST
+  redis_port: os.environ/REDIS_PORT
+  redis_password: os.environ/REDIS_PASSWORD
 
-default_vertex_config:
-  vertex_project: "adroit-crow-413218"
-  vertex_location: "us-central1"
+litellm_settings:
+  callbacks: ["prometheus"]