[Mailman-Developers] New LockFile.py

Ron Jarrell jarrell@vt.edu
Mon, 01 May 2000 14:17:43 -0400


--=====================_257564536==_.ALT
Content-Type: text/plain; charset="us-ascii"

At 01:43 PM 5/1/00 -0400, Barry Warsaw wrote:

>The one questionable aspect of the new LockFile.py is that it uses the
>mtime of the file as the lock's lifetime.  This works great for
>avoiding race conditions but means that Mailman is setting mtime's of
>the files to some time in the future.  Is anybody aware of OSes where
>this will be a problem?  Linux and Solaris both seem to stay happy.

The things I can think of:

1) This makes mailman very sensitive to time synchonization problems
in environments with NFS mounted directories (possibly more a problem
with mailman than many others since it's so easy to just run it out of ~mailman
as a regular user).

2) Setting mtime in the future will probably piss off fsck; which in environments
with multiple systadmins, or the mailman person isn't a sysadmin, could be
horribly confusing.

Both can probably be addressed by adding warnings to the readme and/or
install files.


--=====================_257564536==_.ALT
Content-Type: text/html; charset="us-ascii"

<html>
<font size=3>At 01:43 PM 5/1/00 -0400, Barry Warsaw wrote:<br>
<br>
<blockquote type=cite cite>The one questionable aspect of the new
LockFile.py is that it uses the<br>
mtime of the file as the lock's lifetime.&nbsp; This works great 
for<br>
avoiding race conditions but means that Mailman is setting mtime's
of<br>
the files to some time in the future.&nbsp; Is anybody aware of OSes
where<br>
this will be a problem?&nbsp; Linux and Solaris both seem to stay
happy.</font></blockquote><br>
The things I can think of:<br>
<br>
1) This makes mailman very sensitive to time synchonization 
problems<br>
in environments with NFS mounted directories (possibly more a
problem<br>
with mailman than many others since it's so easy to just run it out of
~mailman<br>
as a regular user).<br>
<br>
2) Setting mtime in the future will probably piss off fsck; which in
environments<br>
with multiple systadmins, or the mailman person isn't a sysadmin, could
be<br>
horribly confusing.<br>
<br>
Both can probably be addressed by adding warnings to the readme
and/or<br>
install files.<br>
<br>
</html>

--=====================_257564536==_.ALT--