asyncore does not check for POLLERR and POLLHUP if neither readable nor writable

[socketpair]

BPO	11267
Nosy	@vstinner, @giampaolo, @socketpair

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = None
closed_at = <Date 2014-08-27.16:29:53.358>
created_at = <Date 2011-02-21.05:54:31.497>
labels = ['invalid', 'type-bug', 'library', 'expert-IO']
title = 'asyncore does not check for POLLERR and POLLHUP if neither readable nor writable'
updated_at = <Date 2014-08-27.16:29:53.355>
user = 'https://github.com/socketpair'

bugs.python.org fields:

activity = <Date 2014-08-27.16:29:53.355>
actor = 'vstinner'
assignee = 'none'
closed = True
closed_date = <Date 2014-08-27.16:29:53.358>
closer = 'vstinner'
components = ['Library (Lib)', 'IO']
creation = <Date 2011-02-21.05:54:31.497>
creator = 'socketpair'
dependencies = []
files = []
hgrepos = []
issue_num = 11267
keywords = []
message_count = 3.0
messages = ['128941', '220903', '225999']
nosy_count = 4.0
nosy_names = ['vstinner', 'giampaolo.rodola', 'BreamoreBoy', 'socketpair']
pr_nums = []
priority = 'normal'
resolution = 'not a bug'
stage = None
status = 'closed'
superseder = None
type = 'behavior'
url = 'https://bugs.python.org/issue11267'
versions = ['Python 3.1', 'Python 3.4', 'Python 3.5']

[socketpair]

asyncore.py: poll2() :
--------------------

if flags:
    # Only check for exceptions if object was either readable
    # or writable.
    flags |= select.POLLERR | select.POLLHUP | select.POLLNVAL
    pollster.register(fd, flags)

---

This is unexpected behaviour. Why descriptor does not get polled if neither read nor write selected ? So I can not track dropping of connection if I do not want neither read nor write. I think "if flags:" should be removed.

[BreamoreBoy]

I'm assuming that any work on this is unlikely as asyncore is deprecated in favour of asyncio.

[vstinner]

Why descriptor does not get polled if neither read nor write selected ?

IMO it's a deliberate design choice, made for performances.

In the asyncio module, the high-level StreamReader API stops listening to read even if the buffer is too large (if we received more than limit*2 bytes, where limit is 64 KB by default). We listen again when the application reads enough data from the buffer.

When we stop listening to read event (and we don't listen to write events neither), we ignore completly the socket. So if the socket is closed, we are not notified immediatly. IMO it's a design choice for efficency. It's more efficient to ignore temporarily the socket while the application is busy, than notifying immediatly that the connection was closed.

In my test, I still got some bytes with socket.recv() even after the connection was already closed by the peer.

Pseudo-code:

• client: connect
• client: recv(10): get some bytes
• server: close the connection
• client: recv(10): get more bytes (even if the connection was already closed)
• client: notified that the connection was closed

I don't think that asyncore nor asyncio should be modified to be notified immediatly that a connection was closed, so I close the issue.

Thanks for your report, it was an interesting question :-)