From d1eedb04bd26805fffc41e5aa1060f72e8a6f072 Mon Sep 17 00:00:00 2001
From: Ishaan Jaff <ishaanjaffer0324@gmail.com>
Date: Sat, 10 Aug 2024 14:05:18 -0700
Subject: [PATCH] prometheus log_success_fallback_event

---
 litellm/integrations/prometheus.py | 33 ++++++++++++++++++++++++++++++
 litellm/proxy/proxy_config.yaml    |  1 +
 2 files changed, 34 insertions(+)

diff --git a/litellm/integrations/prometheus.py b/litellm/integrations/prometheus.py
index 1425b101c0..8e31bca137 100644
--- a/litellm/integrations/prometheus.py
+++ b/litellm/integrations/prometheus.py
@@ -170,6 +170,17 @@ class PrometheusLogger(CustomLogger):
                     labelnames=_logged_llm_labels,
                 )
 
+                self.llm_deployment_successful_fallbacks = Counter(
+                    "llm_deployment_successful_fallbacks",
+                    "LLM Deployment Analytics - Number of successful fallback workloads",
+                    ["primary_model", "fallback_model"],
+                )
+                self.llm_deployment_failed_fallbacks = Counter(
+                    "llm_deployment_failed_fallbacks",
+                    "LLM Deployment Analytics - Number of failed fallback workloads",
+                    ["primary_model", "fallback_model"],
+                )
+
         except Exception as e:
             print_verbose(f"Got exception on init prometheus client {str(e)}")
             raise e
@@ -479,6 +490,28 @@ class PrometheusLogger(CustomLogger):
             )
             return
 
+    async def log_success_fallback_event(self, original_model_group: str, kwargs: dict):
+        verbose_logger.debug(
+            "Prometheus: log_success_fallback_event, original_model_group: %s, kwargs: %s",
+            original_model_group,
+            kwargs,
+        )
+        _new_model = kwargs.get("model")
+        self.llm_deployment_successful_fallbacks.labels(
+            primary_model=original_model_group, fallback_model=_new_model
+        ).inc()
+
+    async def log_failure_fallback_event(self, original_model_group: str, kwargs: dict):
+        verbose_logger.debug(
+            "Prometheus: log_failure_fallback_event, original_model_group: %s, kwargs: %s",
+            original_model_group,
+            kwargs,
+        )
+        _new_model = kwargs.get("model")
+        self.llm_deployment_failed_fallbacks.labels(
+            primary_model=original_model_group, fallback_model=_new_model
+        ).inc()
+
     def set_deployment_state(
         self,
         state: int,
diff --git a/litellm/proxy/proxy_config.yaml b/litellm/proxy/proxy_config.yaml
index c16d09e72c..8d499c1c0f 100644
--- a/litellm/proxy/proxy_config.yaml
+++ b/litellm/proxy/proxy_config.yaml
@@ -37,5 +37,6 @@ general_settings:
   master_key: sk-1234
 
 litellm_settings:
+  fallbacks: [{"gemini-1.5-pro-001": ["gpt-4o"]}]
   success_callback: ["langfuse", "prometheus"]
   failure_callback: ["prometheus"]