write() method in Transport should not buffer data

[BossKwei]

BPO	32763
Nosy	@asvetlov, @1st1

^{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 2018-02-04.14:59:31.483>
created_at = <Date 2018-02-04.14:30:37.096>
labels = ['invalid', 'expert-asyncio']
title = 'write() method in Transport should not buffer data'
updated_at = <Date 2018-02-04.14:59:31.482>
user = 'https://bugs.python.org/BossKwei'

bugs.python.org fields:

activity = <Date 2018-02-04.14:59:31.482>
actor = 'asvetlov'
assignee = 'none'
closed = True
closed_date = <Date 2018-02-04.14:59:31.483>
closer = 'asvetlov'
components = ['asyncio']
creation = <Date 2018-02-04.14:30:37.096>
creator = 'Boss Kwei'
dependencies = []
files = []
hgrepos = []
issue_num = 32763
keywords = []
message_count = 2.0
messages = ['311600', '311601']
nosy_count = 3.0
nosy_names = ['asvetlov', 'yselivanov', 'Boss Kwei']
pr_nums = []
priority = 'normal'
resolution = 'not a bug'
stage = 'resolved'
status = 'closed'
superseder = None
type = None
url = 'https://bugs.python.org/issue32763'
versions = ['Python 3.4', 'Python 3.5', 'Python 3.6']

[BossKwei]

write() method implemented in https://github.com/python/cpython/blob/master/Lib/asyncio/selector_events.py#L830 is not stable in somecases. If this method was called too quickly, separate data will be packed and sent in same tcp package, which may be considered as unexpected behavior.

[asvetlov]

Transport buffers writes if kernel buffer is full, the behavior is intentional and present starting from very beginning of asyncio development.

Moreover, two plain socket.send() calls can be joined into single TCP packet, TCP protocol is a STREAM of data by design, not a sequence of packets.