An infinite loop happens when we use SysLogHandler with eventlet

[KazutakaMorita]

BPO	15179
Nosy	@vsajip, @pitrou
Files	syslog.patch: fix a close leak in SysLogHandler

^{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 2012-06-25.22:20:39.411>
created_at = <Date 2012-06-25.11:29:31.349>
labels = ['type-bug', 'library']
title = 'An infinite loop happens when we use SysLogHandler with eventlet'
updated_at = <Date 2012-06-25.22:20:39.410>
user = 'https://bugs.python.org/KazutakaMorita'

bugs.python.org fields:

activity = <Date 2012-06-25.22:20:39.410>
actor = 'python-dev'
assignee = 'none'
closed = True
closed_date = <Date 2012-06-25.22:20:39.411>
closer = 'python-dev'
components = ['Library (Lib)']
creation = <Date 2012-06-25.11:29:31.349>
creator = 'Kazutaka.Morita'
dependencies = []
files = ['26154']
hgrepos = []
issue_num = 15179
keywords = ['patch']
message_count = 4.0
messages = ['163939', '163977', '163981', '164020']
nosy_count = 4.0
nosy_names = ['vinay.sajip', 'pitrou', 'python-dev', 'Kazutaka.Morita']
pr_nums = []
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = 'behavior'
url = 'https://bugs.python.org/issue15179'
versions = ['Python 2.7', 'Python 3.2', 'Python 3.3']

[KazutakaMorita]

If we stop a syslog daemon when running the following program, it
leads to an infinite loop.

==
#!/usr/bin/env python

import eventlet
from logging.handlers import SysLogHandler
import time
import logging

eventlet.patcher.monkey_patch(all=False, socket=True)

logger = logging.getLogger('log')
logger.addHandler(SysLogHandler('/dev/log'))

while True:
    print "send a message to logger"
    logger.error("syslog test")
    time.sleep(1)
==

It is because there is a close leak in the python logging module, and
SysLogHandler continues to send a log message against the closed
connection forever.

The following patch seems to fix this problem:

diff -r 3b5545ba6432 Lib/logging/handlers.py
--- a/Lib/logging/handlers.py	Wed Jun 13 22:15:26 2012 -0400
+++ b/Lib/logging/handlers.py	Mon Jun 25 20:27:46 2012 +0900
@@ -801,7 +801,11 @@
         except socket.error:
             self.socket.close()
             self.socket = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
-            self.socket.connect(address)
+            try:
+                self.socket.connect(address)
+            except socket.error:
+                self.socket.close()
+                raise
 
     def encodePriority(self, facility, priority):
         """

[vsajip]

Why should the socket need closing if the connect fails? This seems a flawed contract; either connect() should work (in which case the socket will need closing when it's finished with) or it should fail, not connect and not require a close call (as it isn't connected). Could this be a problem with eventlet? One could argue that the leak is actually in whatever implements socket.connect in this case (presumably eventlet).

[pitrou]

This seems a flawed contract; either connect() should work (in which
case the socket will need closing when it's finished with) or it should
fail, not connect and not require a close call (as it isn't connected)

socket.close() closes the file descriptor, not only the underlying network connection. Otherwise the file descriptor will stay open until the socket object is garbage collected.

[python-dev]

New changeset 99f0c0207faa by Vinay Sajip in branch '2.7':
Issue bpo-15179: Closed socket on connection failure. Thanks to Kazutaka Morita for the patch.
http://hg.python.org/cpython/rev/99f0c0207faa

New changeset 1bc1a14feb70 by Vinay Sajip in branch '3.2':
Issue bpo-15179: Closed socket on connection failure. Thanks to Kazutaka Morita for the patch.
http://hg.python.org/cpython/rev/1bc1a14feb70

New changeset 6af0535b5e3a by Vinay Sajip in branch 'default':
Closes bpo-15179: Merged fix from 3.2.
http://hg.python.org/cpython/rev/6af0535b5e3a