[cpx] data corruption update

[Scott Wiersdorf <scottw@xxxxxxxxxxxx>]

Friends,

Our mutual friend, Alvin Tan, has identified what I believe the cause
of the latest cpx.conf corruption issues.

Two fixes have been pushed out in the past 4 months or so that have
fixed almost all cases of this; these fixes addressed the inherent
race condition when two or more administrators are accessing the
control panel at the same time.

This new problem that Alvin has found (at least, I'm pretty sure about
it also), is where the VPS itself is out of processes. You might see
entries like this in your messages log:

Jul 20 16:30:28 server inetd[80024]: fork: Resource temporarily unavailable
Jul 20 16:30:29 server sm-mta[80031]: NOQUEUE: SYSERR(root): daemon: cannot fork: Resource temporarily unavailable

This indicates that your server is rather busy! Each server has a
different number of available processes. My own VPS v2 (a B2) has 150
processes. You can see how many processes you have available by typing
'limits' at your shell prompt and look at the 'maxprocesses' line.

We'll be working on a fix in the coming weeks, but this will help
anyone else who might be having the same problem.

*What can I do about it?*

I'm glad you asked. One easy solution is to upgrade your server. This
condition means that you're really pushing the envelope of your server
and could use some more horses.

A second solution would be to profile your server and see what it's
doing. Using top and ps, you can see where all the processes are
going; with this information you can optimize your server's forking
profile to balance between what your server is actually doing with
what it should be doing.

If you have a busy web server, for example, you'll see a bunch of
httpsd processes and maybe some mysql processes. If you have a busy
mail server, on the other hand, you might see a bunch of sendmail,
procmail, spamassassin, and/or clamav processes.

If you primarily use your server for serving web, you might scale back
inetd requests with the "-c max", "-C rate", "-R rate", and "-s max"
options (see the inetd manpage for details). You can also throttle
sendmail with some easy config changes. This will slow your server
response time for people checking and sending mail, doing ftp, etc.

If you primarily use your server for mail, you might consider lowering
your httpd.conf MaxClients setting to reduce the number of Apache
children running at peak traffic. This will slow your web response
time to web visitors.

In short, you'll likely see other problems when your server is maxed
out, but here are a few ways to squeeze a little more out of them
until we have the cpx issue fixed.

Scott
-- 
Scott Wiersdorf
scottw@xxxxxxxxxxxx
======================================================================
This is <cpx@xxxxxxxxxxxxx>      <http://www.groupmail.org/lists/cpx/>
Before posting a question, please search the archives (see above URL).