Ctrl-C while an rpc handler has not finished causes errors

[varunchodanker]

In #13.

• Reproducible on Linux and Windows.
• Tested by @moodyhunter and @varunchodanker.
• In-depth knowledge of 'asyncio' and event loops will likely be necessary for a fix.

The errors produced are given by the following:

(venv) PS C:\Users\varun\OneDrive - University of Warwick\Learning\grpc-io\proactor_eventsn-context> python .\account_server.py
Account Service Server started. Listening on port: 50051                 o\proactor_events
1970-01-01 00:00:00+00:00
Account Service Server stopped.                                          o\base_events.py"
Exception ignored in: <function _ProactorBasePipeTransport.__del__ at 0x000001A384647E20>                                                         o\base_events.py"
Traceback (most recent call last):
  File "C:\Users\varun\AppData\Local\Programs\Python\Python310\lib\asyncio\proactor_events.py", line 116, in __del__                              n-context> 
  File "C:\Users\varun\AppData\Local\Programs\Python\Python310\lib\asyncio\proactor_events.py", line 108, in close
  File "C:\Users\varun\AppData\Local\Programs\Python\Python310\lib\asyncio\base_events.py", line 745, in call_soon
  File "C:\Users\varun\AppData\Local\Programs\Python\Python310\lib\asyncio\base_events.py", line 510, in _check_closed
RuntimeError: Event loop is closed

[moodyhunter]

probably fixed, by c3e5a84

[varunchodanker]

Will check on Windows now. Can you confirm whether it is fixed on Linux?

[moodyhunter]

yes on Linux,

I set up a 20-second sleep, during that time I press Ctrl+C, the code now says: "Stopping..."
After 20secs it stops properly with correct data returned back to the client.

[varunchodanker]

Still getting the exact same error on Windows.

[moodyhunter]

confirmed that c07d1ba fixed the problem

[varunchodanker]

c07d1ba uses selector-based event loops that do not have the bug that is present in Proactor event loops. This bug is exemplified in encode/httpx#914.

[varunchodanker]

See #20. This was reintroduced to remove a more severe bug. This issue can be mitigated by ensuring all gRPC clients have finished before stopping the server. This can be achieved by closing the Next component of the application first.

[ksanganee]

Are we good to delete this issue and just know that we close the next client first?

[varunchodanker]

Yes I think so.