[Mailman-Developers] Mailman throwing lost-datafile errors with every message

[Brad Knowles]

At 4:17 PM -0400 2004-06-24, foobar at eh.net wrote:

>  Jun 24 15:49:39 2004 (353) lost data files for filebase:
>  1088100914.7161551+9a9200d2cba7be877a02453e1f812069b0c78761
>  <rinse and repeat about once a second>
>  Mailman 2.1.2 (I'll upgrade if you INSIST...)
>  Python 2.2
>  SunOS eh.net 5.7 Generic_106541-24 sun4u sparc SUNW,Ultra-5_10

	Upgrading to 2.1.5 may help solve this problem.  Certainly, this 
is one of the issues that Barry mentioned in his message at 
<http://mail.python.org/pipermail/mailman-developers/2004-January/016396.html>.

	However, before you do that, you might want to try Barry's other 
suggestion in this message, namely to enable SYNC_AFTER_WRITE.  It 
may (or may not) kill your performance, but it may help you avoid 
these errors.


	Out of curiosity, have you added the "logging" option to your 
filesystem mount where you process mail?  My experience is that this 
helps improve the performance of systems running Solaris 7 and above, 
and will greatly reduce/eliminate your time to run fsck when you're 
booting the system.

	Of course, replacing the filesystem with Veritas VxFS would be a 
bigger performance improvement, as would replacing the disk subsystem 
with an external high-performance RAID array, but these are not 
likely to be options you can take advantage of.

>  I've been getting this line thrown in my error logfile frequently.
>  Additionally, mailman has about 1200 files open when the (aging) computer
>  locks up after taking quite a bit of abuse (it tends to end up with the
>  CPU at 100% iowait, out of swap, with more processes running than is
>  healthy. (The machine only has 256 megabytes of RAM-- and the replacement
>  server was initially due to arrive over a month ago, but Hewlett-Packard
>  has delayed that date to about a month from now).

	If possible, you should increase the amount of swap on the 
system.  You probably can't make any hardware upgrades in your 
current condition, but I'd be surprised if you can't add more swap 
space, even if the new swap space is not contiguous with the old.

	That might at least help you survive the problems a bit better.

-- 
Brad Knowles, <brad.knowles at skynet.be>

"They that can give up essential liberty to obtain a little temporary
safety deserve neither liberty nor safety."
     -Benjamin Franklin, Historical Review of Pennsylvania.

   SAGE member since 1995.  See <http://www.sage.org/> for more info.