It looks like instantiated extensions passed to Schema are racy.

Example of racy usage:

For OpenTelemetryExtension a single instance of _span_holder is going to be created and reused.

Beyond just OpenTelemetryExtension, the setting of execution_context in ExtensionsRunner is racy:

You can trivially show this using the following example:

1. Create app.py

   import asyncio
   from typing import Optional
   
   import strawberry
   from strawberry.extensions import Extension
   from strawberry.types.execution import ExecutionContext
   
   
   class TestExtension(Extension):
       def __init__(
           self,
           *,
           execution_context: Optional[ExecutionContext] = None,
       ):
           if execution_context:
               self.execution_context = execution_context
   
       def on_request_start(self):
           print(f"on_request_start: self={id(self)} execution_context={id(self.execution_context)}")
   
       def on_request_end(self):
           print(f"on_request_end: self={id(self)} execution_context={id(self.execution_context)}")
   
   
   @strawberry.type
   class Query:
       @strawberry.field
       async def test(self, sleep: int) -> int:
           await asyncio.sleep(sleep)
           return sleep
   
   
   schema = strawberry.Schema(query=Query, extensions=[TestExtension()])

2. Then run the following:

   strawberry server app --port 3030

3. In a new window make the following request:

   curl 127.0.0.1:3030/graphql -H 'content-type: application/json' --data '{"query": "{ test(sleep: 30) }"}'

4. And while that request is running, make the concurrent request:

   curl 127.0.0.1:3030/graphql -H 'content-type: application/json' --data '{"query": "{ test(sleep: 1) }"}'

5. If you look at the output of app.py you'll see the following:

   Running strawberry on http://0.0.0.0:3030/graphql 🍓
   [2022-08-03 13:04:24]: No operation name
   { test(sleep: 30) }
   
   on_request_start: self=4478445024 execution_context=4478640672
   [2022-08-03 13:04:28]: No operation name
   { test(sleep: 1) }
   
   on_request_start: self=4478445024 execution_context=4478932016
   on_request_end: self=4478445024 execution_context=4478932016
   on_request_end: self=4478445024 execution_context=4478932016
   

   You'll notice there is no on_request_end for 4478640672 and a double execution of 4478932016.