websocket client: when the network is bad, it shows fatal read error message

[yjqiang]

This is a msg in PYTHONISTA on ios (python 3.6.1). So maybe it is a bug in pythonista.

Fatal read error on socket transport
protocol: <asyncio.sslproto.SSLProtocol object at 0x1177ce940>
transport: <_SelectorSocketTransport fd=47 read=polling write=<idle, bufsize=0>>
Traceback (most recent call last):
File "/var/containers/Bundle/Application/9F565D30-C54B-4210-902B-874D7A5AB814/Pythonista3.app/Frameworks/Py3Kit.framework/pylib/asyncio/selector_events.py", line 725, in _read_ready
data = self._sock.recv(self.max_size)
TimeoutError: [Errno 60] Operation timed out

[kxepal]

Definitely, not a bug of aiohttp nor Pythonista, but remote server which couldn't respond within specified timeout interval. There is nothing to do from aiohttp side.

[yjqiang]

@kxepal "remote server which couldn't respond within specified timeout interval"? I wrap timeout with the receive function like https://github.com/yjqiang/bilibili-live-tools/blob/master/bilibiliCilent.py#L92 .

[asvetlov]

I cannot understand how did you get TimeoutError (not asyncio.TimeoutError)?
The former is never used by asyncio nor aiohttp.
Bare TimeoutError with errno 60 is for synchronous sockets only.

[yjqiang]

The server said we must send a special heartbeat package 1 time per 30 seconds, and it would send a confirm msg to client after receiving the msg. So I wrote that to make sure that when the server closes the connection but without talking to client, after 35s, the client will find that fact.

[yjqiang]

Can I handle the error "TimeoutError: [Errno 60] Operation timed out" by myself? I used try and exception, but it did not work at all.

[yjqiang]

Sorry my English is not good so I think maybe I misunderstood your words. You mean why the msg is TimeoutError instead of asyncio.TimeoutError? I can't get it too. Maybe it is a bug in pythonista?

[asvetlov]

@yjqiang yes. I mean that TimeoutError is never raised by asyncio/aiohttp in normal usage.
It could be raised if a blocking socket was passed into event loop but this case is definitely an error.

I know nothing about pythonista thogh

[yjqiang]

omz/Pythonista-Issues#563
Looks like it is not a bug in pythonista, asyncio raises timeout error. Maybe aiohttp should catch and handle it?

[yjqiang]

@asvetlov

[yjqiang]

The reason why it raises TimeoutError is that the code in omz/Pythonista-Issues#563 (comment)
And 3 years ago, people discussed about that in python/asyncio#135

[asvetlov]

TimeoutError builtin is raised when a syscall returns ETIMEDOUT status code.
man errno says that this error code is for "Connection timed out (POSIX.1-2001)".

From my understanding, the error should be never raised for a nonblocking socket, only EAGAIN and EWOULDBLOCK are possible values if the socket is not ready.

UPD.
Looks like TCP keepalive facility can return ETIMEDOUT: https://stackoverflow.com/questions/16772519/socket-recv-on-selected-socket-failing-with-etimedout

I need to investigate more.

[yjqiang]

TimeoutError builtin is raised when a syscall returns ETIMEDOUT status code.
man errno says that this error code is for "Connection timed out (POSIX.1-2001)".

From my understanding, the error should be never raised for a nonblocking socket, only EAGAIN and EWOULDBLOCK are possible values if the socket is not ready.

UPD.
Looks like TCP keepalive facility can return ETIMEDOUT: https://stackoverflow.com/questions/16772519/socket-recv-on-selected-socket-failing-with-etimedout

I need to investigate more.

"a NAT gateway could time out after some minutes of not doing anything" So the server does not work well or the network is not normal, after I setup the connection. If it is that reason, maybe aiohttp/asyncio should raise this exception when I use recv() and let us handle it? But maybe they would hold the exception for a long time if I did not use recv() immediately.

[asvetlov]

The current behavior is:

1. TimeoutError is received on socket operation.
2. asyncio closes transport and logs the error by loop.call_exception_handler().

The logging can be suppressed in next Python release (as we do it for ConnectionError and several other exceptions).
But I see no reason to change the main reaction: close the socket and call protocol.connection_lost(exc) with exception info.

[yjqiang]

Thanks.

[yjqiang]

@asvetlov https://bugs.python.org/issue34148 I think it should be fixed by just add TimeoutError to base_events._FATAL_ERROR_IGNORE. But I didn't find the api about this issue in aiohttp.

[asvetlov]

Move discussion to Python bug tracker

[yjqiang]

Move discussion to Python bug tracker

No response yet for 2 months.
python/cpython#11576