ContextVar MemoryLeak

Basically in your case the finished tasks weren't cleaned. You can see reference to them in asyncio.tasks._all_tasks:

from asyncio import tasks async def main(): task1, task2 = run(1), run(2) await asyncio.gather(task1, task2) futures = tasks._all_tasks for future in futures: print(future, future.done()) 

This code will output three tasks:

<Task finished name='Task-2' coro=<run() done...> True <Task finished name='Task-3' coro=<run() done...> True <Task pending name='Task-1' coro=<main() running ...> False 

So, because tasks still referenced in tasks._all_tasks the context wasn't cleaned. I guess this is one of the examples where Python memory management is not very efficient. You may solve this problem in one of the following ways:

Solution 1
You can add additional async short sleep, this will allow to collect the garbage from the memory.

async def main(): await asyncio.gather(run(1), run(2)) await asyncio.sleep(-0.01) gc.collect() 

Solution 2
Another option is to do additional 'wrapper' for 'asyncio.gather'. This will also allow to properly clean the memory in the context.

import asyncio import gc from contextvars import ContextVar import objgraph ctx = ContextVar('ctx') class MyObj: def __init__(self, value): self.value = value async def run(value): ctx.set(MyObj(value)) async def both_run(): task1 = asyncio.create_task(run(1)) task2 = asyncio.create_task(run(2)) await asyncio.gather(task1, task2) async def main(): task = asyncio.create_task(both_run()) await asyncio.gather(task) print('# of MyObj=', objgraph.count('MyObj')) for obj in objgraph.by_type('MyObj'): print('MyObj.value-',obj.value) print('outer ctx=', ctx.get(None)) objgraph.show_backrefs(objgraph.by_type('MyObj'), filename='ctx.png', max_depth=5) asyncio.run(main())