ONLamp.com
oreilly.comSafari Books Online.Conferences.

advertisement


Mail Server Performance Monitoring with Mailgraph

by David Ky
08/12/2004

Many seasoned administrators are more than familiar with how mail servers mature quickly. As things scale and grow, maintainability becomes much more difficult. Tailing a logfile is no longer an option to understand what your MTA is doing when it processes hundreds or thousands of requests every minute. At that point, automated monitoring tools — text-based or graphical — become useful. Mailgraph is one such tool.

The Mailgraph web site explains: "Mailgraph is a very simple mail statistics RRDtool front-end for Postfix that produces daily, weekly, monthly, and yearly graphs of received/sent and bounced/rejected mail."

David Schweikert, the author of Mailgraph, also provides an example of Mailgraph in action.

In a nutshell, installing Mailgraph will allow us to see how our mail server performs through neatly laid-out graphical and numerical representations of mail traffic flowing through a particular mail server. If you've ever used a similar tool that can display graphs, such as MRTG, you know that graphs often speak volumes of invaluable information when trying to diagnose a problem quickly. Graphs can portray information about the past, present, and sometimes even the future.

You may have noticed that we've also mentioned RRDtool. RRDtool is a piece of software created by Tobi Oetiker that has the ability to store and display data that changes over time in a Round Robin Database (RRD) that stays fixed in size over time. If you ever feel like Mailgraph isn't enough in terms of monitoring, you can always start monitoring other server variables as well, like your server-load average by using RRDtool as a back-end.

Installing Mailgraph

We'll use Postfix as an MTA here. This article assumes that you already have this running in one way or another, via a source installation or via your local package manager (portage, apt, yum, etc.). If you don't already have it installed, download the source from the Postfix Web site.

It's worth taking note that the example setup already has amavisd-new coupled with SpamAssassin and ClamAV to tag spam and viruses. Mailgraph will still function without a virus- and spam-checking facility, but it will not graph spam or virus data if there's nothing to flag them.

Mailgraph requires a couple of other packages:

  • RRDtool.
  • The Perl modules Time::HiRes and File::Tail. Download these from CPAN with:

    perl -MCPAN -e shell
  • Mailgraph itself is of course downloadable from the Mailgraph downloads page.

First install RRDtool, which will do the actual data storage and manipulation of the graphs for us.

Download the latest RRDtool release. At the time of this writing, it's 1.0.48. Check the RRDtool web site to see if there is a newer version).

david@toys david $ wget http://people.ee.ethz.ch/~oetiker/webtools/rrdtool/pub/rrdtool-1.0.48.tar.gz

Verify your download against the MD5 checksum:

david@toys david $ wget http://people.ee.ethz.ch/~oetiker/webtools/rrdtool/pub/rrdtool-1.0.48.tar.gz.md5
david@toys david $ md5sum -c rrdtool-1.0.48.tar.gz.md5
rrdtool-1.0.48.tar.gz: OK

Unpack the RRDtool release and switch to its directory.

david@toys david $ tar -xvzf rrdtool-1.0.48.tar.gz
david@toys david $ cd rrdtool-1.0.48

Do the general, configure, make, and make install loop. The make site-perl-install step makes other steps easier as well.

david@toys rrdtool-1.0.48 $ sh configure
david@toys rrdtool-1.0.48 $ make
david@toys rrdtool-1.0.48 $ sudo make install
david@toys rrdtool-1.0.48 $ sudo make site-perl-install

With that done, we can installing the required Perl modules, Time::HiRes and File::Tail. Mailgraph uses them to keep track of our mail logs. Installing these is quite simple with the CPAN module.

david@toys david $ sudo perl -MCPAN -e shell
cpan> install Time::HiRes
cpan> install File::Tail
cpan> quit

Now that we've installed those, we can finally start picking a bit more at Mailgraph itself.

Make sure you have the latest Mailgraph package.

david@toys david $ wget http://people.ee.ethz.ch/~dws/software/mailgraph/pub/mailgraph-1.8.tar.gz

Unpack the Mailgraph release and jump into its directory.

david@toys david $ tar -xvzf mailgraph-1.8.tar.gz
david@toys david $ cd mailgraph-1.8

Configuring Mailgraph

We first need to edit the example mailgraph-init script to do our bidding. Edit the MAIL_LOG and RRD_DIR directories to point to your mail log and the web directory where you'll put mailgraph.cgi.

david@toys mailgraph-1.8 $ vi mailgraph-init
MAIL_LOG=/var/log/mail.log
RRD_DIR=/var/www/virtual/cgi-bin/toys.example.com/cgi-bin/

Per the mailgraph-init example script, copy files over accordingly.

david@toys mailgraph-1.8 $ sudo cp mailgraph.pl /usr/local/bin/mailgraph.pl

Then move the mailgraph.cgi program, which actually displays the graphs, into the appropriate cgi-bin directory.

david@toys mailgraph-1.8 $ cp mailgraph.cgi /var/www/virtual/toys.example.com/cgi-bin/

Next, tell mailgraph-init to start up everything. In the version I downloaded, this script wasn't marked as executable, so you may have to change its permissions mailgraph-init.

david@toys mailgraph-1.8 $ chmod 755 mailgraph-init
david@toys mailgraph-1.8 $ sudo ./mailgraph-init start

Open a browser that can view images and visit your the cgi-bin directory that now holds mailgraph.cgi. You should see your Mailgraph images. If you did, congratulations! If not, double-check your work carefully to see if you made a typo along the installation — that happens quite often.

If you want mailgraph-init to start up on boot, you'll have to add it into something like your rc.local file or copy and edit an init script from your own distribution. Usually, starting mailgraph-init from rc.local is easiest. Add a line like this:

/etc/init.d/mailgraph-init start

If all goes well, you should start to see graphs that resemble Figure 1.

an example graph
Figure 1. An example graph from Mailgraph.

Don't be alarmed if you see none of those blue or green lines. Good graphs take time, so leave Mailgraph running for a few hours and try again later. If your MTA has processed any mail, Mailgraph should have by then graphed a few lines and reported some numbers at the bottom.

If you still can't seem to make Mailgraph work, or if you think you've found a bug, you can subscribe to its mailing list and ask questions by sending an email with the subject subscribe to mailgraph-request at list.ee.ethz.ch. Be sure to provide specific information about your setup.

Installed, Working, Ready to Go: What am I Looking for?

Now that you've installed Mailgraph, it should become a valuable resource in doing an analysis of your MTA. While it may not be evident immediately, Mailgraph can help you test the effectiveness of your filtering setup.

For example, the graph in Figure 2 may indicate a few things.

weekly graph of errors
Figure 2. A weekly graph of errors.

  • The reject rate for an MTA of this size is pretty high. The cause could come from several factors -- perhaps our extensive set of header_checks or from connecting hosts listed in the RBLs that we use. If this rate continues to increase, it might be good to check the mail log for something fishy, particularly anomalies in mail delivery.
  • The bounces don't set off any alarms, but if we become suspicious, it might be worth finding their causes.
  • The amount of viruses received is normal for this MTA and there are no signs that there is a virus storm, but it might be a good idea to update our ClamAV databases using freshclam just to be sure we're up-to-date.
  • There is very little spam this week. This isn't a cause for alarm, but should turn on a light bulb. The spam count could be low due to the high reject rate — we might be rejecting legitimate mail via our header_checks. If we start hearing high rates of spam reports from users or questions about missing mail, it's worth investigating and possibly reconfiguring SpamAssassin to tag at a lower threshold, adding new rules, or removing portions of our header_checks to allow more mail to pass.

Mailgraph can also help you identify and see attacks as they're happening and help make a post-assessment of any damage. For instance, if you're running a dedicated mail server and start seeing load averages jump, looking at Mailgraph can easily show what kind of mail your MTA has received, be it spam, rejects, bounces, or viruses. You might see indications of an attack, which should prompt you to look at your mail log for suspicious hosts and then possibly take action, perhaps firewalling off a host or using an RBL specifically tailored to the attacking host(s). That way, your users don't have to tell you that email is slow. You'll see it ahead of time and take action early.

More Ideas

To monitor many hosts all at once, you can also create a single index page that contains all the images from your various Mailgraph installations so that they're always a click away. To make the single index page, simply view the source of mailgraph.cgi and "steal" the image links. It can end up looking as simple as Figure 3.

a unified server status page
Figure 3. A unified server status page.

Showing Mailgraph to others like the hard-balled people in management can help sway decisions as well. If your mail server starts to suffer under a heavy load of mail and management is wondering where their emails are, show them a few graphs with a few toppings of sar output. The graphs and numbers should be more than enough for anybody to understand that your systems are overloaded and need attention. If not, show how mail traffic looked before and how it looks now.

Please remember that adding Mailgraph to your list of weapons is no substitute for occasional and true vigilance. Mailgraph is only there to suggest that you check things out.

David Ky is a part-time server administrator and full-time geek.


Return to ONLamp.com.



Sponsored by: