test_create_server_ssl_over_ssl times out on some systems

[zooba]

BPO	46822
Nosy	@pfmoore, @tjguk, @asvetlov, @zware, @1st1, @zooba
PRs	bpo-46822: Increase timeout for test_create_server_ssl_over_ssl to match underlying timeouts #31502

^{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 2022-02-22.17:42:58.344>
created_at = <Date 2022-02-22.00:02:05.894>
labels = ['3.11', 'type-bug', 'tests', 'OS-windows', 'expert-asyncio']
title = 'test_create_server_ssl_over_ssl times out on some systems'
updated_at = <Date 2022-02-22.17:42:58.344>
user = 'https://github.com/zooba'

bugs.python.org fields:

activity = <Date 2022-02-22.17:42:58.344>
actor = 'steve.dower'
assignee = 'none'
closed = True
closed_date = <Date 2022-02-22.17:42:58.344>
closer = 'steve.dower'
components = ['Tests', 'Windows', 'asyncio']
creation = <Date 2022-02-22.00:02:05.894>
creator = 'steve.dower'
dependencies = []
files = []
hgrepos = []
issue_num = 46822
keywords = ['patch']
message_count = 4.0
messages = ['413687', '413711', '413712', '413715']
nosy_count = 6.0
nosy_names = ['paul.moore', 'tim.golden', 'asvetlov', 'zach.ware', 'yselivanov', 'steve.dower']
pr_nums = ['31502']
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = 'behavior'
url = 'https://bugs.python.org/issue46822'
versions = ['Python 3.11']

[zooba]

This causes a failure on one of my test machines where the firewall settings forbid it.

However, the test itself seems designed to only listen on localhost. Even tracing all call through socket, I don't see when or where it is attempting to listen on 0.0.0.0, and yet TCP Monitor (and my firewall) claim that it is.

This seems to be fairly recent, though I haven't done a bisect yet. Anyone have any ideas?

[zooba]

Looks like my issue is actually related to the timeouts added in bpo-44011.

I'm guessing it's an SSL shutdown timeout, because those are notoriously weird on Windows/OpenSSL.

The timeout is now 30s, but the test aborts after 10s. If I increase the test timeout to 50s, it passes reliably.

However, if I pass ssl_shutdown_timeout to the protocol created in the test (which I think is the right place?), it fails with a timeout somewhere else.

So my guess is that my system (VM in a somewhat protected network) is legitimately taking that long to shut down the connection? And so the test timeout ought to be increased?

[zooba]

In fact, I only have to increase the test timeout to 15s to get it to pass reliably.

I don't see any reason the test should time out quicker than the operations it's testing though, right? So set it to 30s?

[zooba]

New changeset 77f31a9 by Steve Dower in branch 'main':
bpo-46822: Increase timeout for test_create_server_ssl_over_ssl to match underlying timeouts (GH-31502)
77f31a9