There are a lot of excellent monitoring tools (Big Brother, Nagios and so on), and some of them allow recovery from dead services, but with great complexity in their configuration, which becomes even more complicated when you want to supervise local services that are not remotely accessible, such as syslog, xinet, mrtg, iptables or Nagios itself.

The purpose with SMART was to have a simple, flexible and quick-to-implement application for monitoring the most critical system dæmons that made it possible to add new ones without modifying the code and to avoid installation and configuration complexities. It also needed to be capable of making decisions and solving problems (or at least trying to do that).

After a first version of “passive” monitoring, we tried to go a step further and obtain an “active” application, that is to say, to add the possibility of auto-recovery. By executing the application periodically through crond, it should detect dæmons that were down and boot them without the intervention of the system administrator.

Later, we considered the possibility that a nonprivileged user could execute this application from a console or remotely (via Telnet or SSH). Centralization of detection and error recovery in only one script made integration with sudo easier. Furthermore, it allowed delegating some stronger recovery actions needed in critical situations, such as rebooting the whole system, to this nonroot-privileged user.

With the ps command, we can list all the active processes in the system, but being “active” is not the same as being “operative”, so this led us to include the check scripts, which are small programs to test services and determine whether they really are operative and answering requests. The difficulties we found suggested that we not waste efforts re-inventing the wheel and profit from plugins included in Nagios (monitoring software that we were using satisfactorily for almost three years).

The distribution of SMART has two shell scripts (smart and check-service), two configuration files (host.conf and services.conf) and two directories (scripts and plugins), which contain the check scripts and the plugins (Listing 1).

[root@server /]# ls -la /home/sysman/
drwxr-x---  4 root sysman    4096 May 27 11:49 .
drwxr-xr-x  3 root root      4096 Jul  8  2003 ..
-rwxr-x---  1 root sysman    1448 May 27 11:51 smart
-rwx------  1 root root      7815 May 27 11:51 check-service
-rw-r--r--  1 root root       242 May 27 11:49 host.conf
drwx------  2 root root      4096 Apr 29 13:38 plugins
drwx------  2 root root      4096 Apr 29 13:39 scripts
-rw-r--r--  1 root root       883 May 17 10:40 services.conf

Permissions of files and directories allow a nonprivileged user called sysman to execute the application, but deny sysman the ability to modify the contents to use it in an inadequate way.

The SMART program reads the configuration files services.conf and host.conf and executes check-service for each defined service. If a check script has been assigned to a service, for example, services 1 and 2 in Figure 1, check-service will execute it, passing the needed parameters and then will wait for the exit status to determine whether the service is alive. If this check script executes some other external script (plugin), such as service 1 in Figure 1, this one will be responsible for checking the service status.

Figure 1. SMART Program

If no check script has been assigned to a service (service 3 in Figure 1), the check-service file will determine the service status by getting the number of active processes. According to this information, the SMART command-line parameters and the configuration parameters, it will decide what actions to carry out.

Integration with the sudo (superuser do) tool allows the system administrator to permit another user (sysman) to start dead services, restart all the services or reboot the whole system. Advantages of this are:

For a user sysman, who needs privileges on the host server, the configuration file of sudo (/etc/sudoers) should be as shown in Listing 2.

# Defaults specification
Defaults:root !syslog

# User privilege specification
root    ALL=(ALL) ALL
sysman  server=(root) NOPASSWD: /home/sysman/check-service
sysman  server=(root) NOPASSWD: /sbin/reboot

This way, we disable syslog logging when sudo is executed by user root, and we assign root privileges to user sysman, at the host server, only for the execution of commands /home/sysman/check-service and /sbin/reboot, without asking sysman for the password every time.

Through the PID file defined in the configuration file, we obtain the parent process identifier (PID), and we determine the number of active processes generated by this service. Next we check whether:

Considering the results obtained in former verifications, we classify the service status:

We group the above five situations in three more general cases:

When executing the program with no parameters, it simply will determine the status of services defined in the configuration file and will display the results. If we want the program to work in an active way, we need to use some of the following parameters:

Once the service status has been determined, and according to the parameter specified in the execution, the action carried out for each service will consist of that shown in Table 1.

Furthermore, independently of the service's status, with the parameters --all and --reboot, a notification via e-mail is sent to the administrator about the performed action.

Listing 3 shows a sample of SMART in action, executed from a console with parameter -d (recovery of services in DOWN status).

There are some optional executables files, the check scripts, responsible for checking whether the monitored services really are operative and responding to petitions. These files are written in Shell (.sh extension) and Expect (.exp extension). Expect is a tool that requires Tcl and allows for automation of interactive applications that use textual representation.

These scripts could be written in any programming language, because only the exit status is taken into account. If it's not equal to 0, we suppose that there has been no answer or that the answer given by the service has not been the expected one. This means that a check script not only can monitor services, but it also can achieve any check that returns a Boolean value, for example, to check whether the size of a directory exceeds a certain value, whether the amount of logged users is greater than a desired number, whether a kernel module is loaded and so on (Listing 4).

[root@server /]# ls /home/sysman/scripts/
disk.nag       http-forb.nag  nfs.nag     pop3.nag     smtp.nag
disk.sh        http.nag       nfs.sh      printer.nag  snmp.nag
dns.nag        http.sh        nmb.sh      proxy.nag    ssh2.nag
dns.sh         imap.exp       ntp.sh      slap.nag     ssh.nag
ftp.exp        imap.nag       pgsql2.nag  slap.sh      ssh.sh
ftp.nag        mysql.nag      pgsql.nag   smb.nag
http-auth.nag  mysql.sh       pgsql.sh    smb.sh
http.exp       nagios.nag     pop3.exp    smtp.exp

Files with the .nag extension are also Shell scripts, but unlike the former ones, they call an external program (plugin) passing to it the parameters received from check-service, following the order and format that the plugin expects. This checks the service and returns the information gathered to the check script, which will interpret and convert it into the exit status that check-service is waiting for (Listing 5).

[root@server /]# ls /home/sysman/plugins/
check_disk  check_http    check_pgsql  check_snmp  check_udp
check_dns   check_imap    check_pop    check_ssh
check_ftp   check_nagios  check_smtp   check_tcp

Plugins are programmed in C, Perl and Shell and belong to Nagios. Their sources can be downloaded independently of the Nagios distribution, and some of them require the additional installation of certain programs and libraries.

Software requirements include the following:

With sudo you can permit another user to run SMART. If you're not interested in creating such a user, you can omit steps 1, 2 and 3 below.

Configuration is as follows. First, edit the SMART host configuration file host.conf, and modify it according to your preferences (hostname, mail addresses, commands paths and so on). Then, edit the SMART services configuration file services.conf, and uncomment/modify/add any service/dæmon you want to check. Every line describes one service, with the following semicolon-separated parameters:

Leave the parameters empty if they are not applicable, except NAME, process_name, max_procs, min_procs and start_mode, which can't be empty.

Now, you should be able to run SMART as user root or sysman:

/home/sysman/smart

Try using -h to get more information about available parameters. Running SMART through crond might be a good thing. You can run it as frequently as you want, but doing it every five minutes seems to be reasonable enough.

SMART is an easy-to-install application (simply copying the program), is much simpler to configure than Nagios (adding a new element to monitor involves adding only one line in the configuration file), and SMART is flexible, allowing you to monitor any service or aspect of the system, and it is very effective.

Our experience in a production environment with thousands of users tells us that it's inevitable that we will reach some peak periods in which the amount of requests received by a service goes beyond the capabilities of the system, and response time grows in a dramatic manner. The fact that the system detects this situation, before its own administrator, and solves it in five minutes, is a great problem solver and provides a perception of better service to users.

After two years of running SMART on about 15 servers, we can say that its main contribution has been our peace of mind. It's wonderful having a colleague who is checking that everything works correctly 24/7 and who informs you about troubles after they already have been solved (especially during the weekends).

SMART was created, developed, tested and enjoyed in the IT Department of the Universitat Internacional de Catalunya. Vicente Sangrador and Jordi Xavier Prat have collaborated on this project and encouraged me to write this article.

Resources for this article: /article/9268.