test(test_users.py): add testing for global proxy spend tracking

tests/test_users.py

@@ -4,6 +4,7 @@ import pytest
 import asyncio
 import aiohttp
 import time
+from openai import AsyncOpenAI
 
 
 async def new_user(session, i, user_id=None, budget=None, budget_duration=None):
@@ -105,7 +106,7 @@ async def test_user_update():
 
 
 @pytest.mark.asyncio
-async def test_users_with_budgets():
+async def test_users_budgets_reset():
     """
     - Create key with budget and 5s duration
     - Get 'reset_at' value
@@ -128,3 +129,63 @@ async def test_users_with_budgets():
         )
         reset_at_new_value = user_info["user_info"]["budget_reset_at"]
         assert reset_at_init_value != reset_at_new_value
+
+
+async def chat_completion(session, key, model="gpt-4"):
+    client = AsyncOpenAI(api_key=key, base_url="http://0.0.0.0:4000")
+    messages = [
+        {"role": "system", "content": "You are a helpful assistant"},
+        {"role": "user", "content": f"Hello! {time.time()}"},
+    ]
+
+    data = {
+        "model": model,
+        "messages": messages,
+    }
+    response = await client.chat.completions.create(**data)
+
+
+async def chat_completion_streaming(session, key, model="gpt-4"):
+    client = AsyncOpenAI(api_key=key, base_url="http://0.0.0.0:4000")
+    messages = [
+        {"role": "system", "content": "You are a helpful assistant"},
+        {"role": "user", "content": f"Hello! {time.time()}"},
+    ]
+
+    data = {"model": model, "messages": messages, "stream": True}
+    response = await client.chat.completions.create(**data)
+    async for chunk in response:
+        continue
+
+
+@pytest.mark.asyncio
+async def test_global_proxy_budget_update():
+    """
+    - Get proxy current spend
+    - Make chat completion call (normal)
+    - Assert spend increased
+    - Make chat completion call (streaming)
+    - Assert spend increased
+    """
+    get_user = f"litellm-proxy-budget"
+    async with aiohttp.ClientSession() as session:
+        user_info = await get_user_info(
+            session=session, get_user=get_user, call_user="sk-1234"
+        )
+        original_spend = user_info["user_info"]["spend"]
+        await chat_completion(session=session, key="sk-1234")
+        await asyncio.sleep(5)  # let db update
+        user_info = await get_user_info(
+            session=session, get_user=get_user, call_user="sk-1234"
+        )
+        new_spend = user_info["user_info"]["spend"]
+        print(f"new_spend: {new_spend}; original_spend: {original_spend}")
+        assert new_spend > original_spend
+        await chat_completion_streaming(session=session, key="sk-1234")
+        await asyncio.sleep(5)  # let db update
+        user_info = await get_user_info(
+            session=session, get_user=get_user, call_user="sk-1234"
+        )
+        new_new_spend = user_info["user_info"]["spend"]
+        print(f"new_spend: {new_spend}; original_spend: {original_spend}")
+        assert new_new_spend > new_spend