Documentation

Requirements

• A computer with linux 2.4/linux 2.6/FreeBSD 5
• rrdtool
• perl
• bash
• A webserver with php support

Since the program is quite modular some collectors require extra programs. It is impossible to collect mysql statistics if there is no mysql server running.

Installation for a single host

Step 1: Fix the collectors

To install run the install script. Use 'install --prefix=/installdir' to install to /installdir. The install script will create a cron.sh file in the installation dir. Put this in a crontab to run every 5 minutes:

*/5 * * * * /installdir/cron.sh

Step 2: Fix the viewing

/installdir now contains a 'site' directory. Make this directory accessible from your webserver. I will not go into details since there are a lot of things that can go wrong for every particular webserver setup. The most simple thing you can do is create a symlink from your webroot to the site directory:

ln -s /installdir/site /var/www/rjstats

You might want to put a .htaccess file in there which limits access to the page. If there are any problems look in the logs.

Step 3: View the graphs

Point your browser to the location you've made rjstats available for and you should see a list of available graphs (Do this after the cron.sh entry has run). Now get some coffee and wait for 30 minutes to get some real values in there.

Installation for a network

It is possible to have 1 rjstats server where other computers send there data to. Currently there is no form of security whatsoever so be warned. At my work we run this setup on all our linux servers inside a private network where only company administrators have access, so we consider it secure enough. Do not put this on an internet accessible host unless you know what you're doing.

I suppose you already have a setup with 1 computer as described above.

Step 1: Fix the server

At this time it only works with xinetd. inetd is not supported. You will have to add an entry to /etc/services:

rjstats 2500/tcp

A xinetd config file is created in /installdir. Read it, adjust it to your situation and copy it to the correct location (probably /etc/xinetd.d). Restart the xinetd server and check your logs for errors. If everything went well xinetd is now listening on tcp port 2500.

Now do something creative with security. I suggest you use tcpwrappers or iptables to deny all connections to port 2500 except for the ip's you want to see graphs for. Use stunnel if you don't want others to see your computers' status.

Step 2: Fix the collectors

Instead of the cron file setup earlier above, you will need to create a new one one each host you want to monitor:

*/5 * * * * * /installdir/cron-client.sh

Inside this file you have to put the ip or hostname of the rjstats server.

Step 3: View the graphs.

If everything worked out you can now see more computers in your browser (wait untill after the first rrdstats_send has run).

Known issues

• When you have a lot of servers you want to monitor and they are all synchronized with ntp your rjstats server will have too many inbound connections every 5 minutes. Look at the cps setting from xinetd.conf
• Some plugins are prone to "hanging". For example the diskusage program will hang if it is checking a dead mount (cifs/smb or nfs is a good candidate). We circumvented this by a lockfile. Unfortunately other graphs do not get updates when 1 process hangs.
• peak values. Some of the plugins are prone to peak values. I have not yet figured out whether this is a rjstats or rrdtool bug. What i do know is that rrdtool 1.2 (released april 2005) has support for filtering out peak values.
  Until then your graphs will sometimes get ruined by peaks. There is a solution, but it is not a very nice one. On the server:

        rrdtool dump ruined_graph.rrd > ruined_graph.xml
        edit ruined_graph.xml and change the abnormal high value.
        rrdtool restore ruined_graph.xml ruined_graph.rrd
        

TODO

• Make it portable to other unices (Only works on linux/FreeBSD) (think snmp here).
• Think of more things to monitor
• Add support for other rrd-functions then AVERAGE. (users should be MAX)
• Add change capability to rrd datastores. (now they can only be created or updated, if one changes you will have to throw away the graph)
• Write a dedicated server instead of relying on xinetd.