| mon(8) | Parallel Service Monitoring Daemon | mon(8) |
mon - monitor services for availability, sending alarms upon failures.
mon [-dfhlMSv] [-a dir] [-A authfile] [-b dir] [-B dir] [-c config] [-D dir] [-i secs] [-k num] [-l [statetype]] [-L dir] [-m num] [-p num] [-P pidfile] [-r delay] [-s dir]
mon is a general-purpose scheduler for monitoring service availability and triggering alerts upon detecting failures. mon was designed to be open in the sense that it supports arbitrary monitoring facilities and alert methods via a common interface, which are easily implemented through programs (in C, Perl, shell, etc.), SNMP traps, and special Mon (UDP packet) traps.
When the mon scheduler starts, it reads a configuration file to determine the services it needs to monitor. The configuration file defaults to /etc/mon.cf, and can be specified using the -c parameter. If the -M option is specified, then the configuration file is pre-processed with m4. If the configuration file ends with .m4, the file is also processed by m4 automatically.
The scheduler enters a loop which handles client connections, monitor invocations, and failure alerts. Each service has a timer, specified in the configuration file as the interval variable, which tells the scheduler how frequently to invoke a monitor process. The scheduler may be temporarily stopped. While it is stopped, client access still functions, but it just doesn't schedule things. This is useful in conjunction while resetting the server, because you can do this: save the hosts and services which are disabled, reset the server with the scheduler stopped, re-disabled those hosts and services, then start the scheduler. It also allows making atomic changes across several client connections. See the moncmd man page for more information.
Monitor processes are invoked with the arguments specified in the
configuration file, appended by the hosts from the applicable host group.
For example, if the watch group is "servers", which contain the
hostnames "smtp", "nntp", and "ns", and the
monitor line reads as follows,
monitor fping.monitor -t 4000 -r 2
then the exectuable "fping.monitor" will be executed with these
parameters:
MONITOR_DIR/fping.monitor -t 4000 -r 2 smtp nntp ns
MONITOR_DIR is actually a search path, by default /usr/local/lib/mon/mon.d then /usr/lib/mon/mon.d, but it can be overridden by the -s option or in the configuration file. If all hosts in the hostgroup have been disabled, then a warning is sent to syslog and the monitor is not run. This behavior may be overridden with the "allow_empty_group" option in the service definition. If the final argument to the "monitor" line is ";;" (it must be preceded by whitespace), then the host list will not be appended to the parameter list.
In addition to environment variables defined by the user in the service definition, mon passes certain variables to monitor process.
"fping.monitor" should return an exit status of 0 if it completed successfully (found no problems), or nonzero if a problem was detected. The first line of output from the monitor script has a special meaning: it is used as a brief summary of the exact failure which was detected, and is passed to the alert program. All remaining output is also passed to the alert program, but it has no required interpretation.
If a monitor for a particular service is still running, and the time comes for mon to run another monitor for that service, it will not start another monitor. For example, if the interval is 10s, and the monitor does not finish running within 10 seconds, then mon will wait until the first monitor exits before running another one.
Upon a non-zero or zero exit status, the associated alert or upalert program (respectively) is started, pending the following conditions: If an alert for a specific service is disabled, do not send an alert. If dep_behavior is set to 'a', or alertdepend is set, and a parent dependency is failing, then suppress the alert. If the alert has previously been acknowledged, do not send the alert, unless it is an upalert. If an alert is not within the specified period, record the failure via syslog(3) and do not send an alert. If the failure does not fall within a defined period, do not send an alert. No upalerts are sent without corresponding down alerts, unless no_comp_alerts is defined in the period section. An upalert will only be sent if the previous state is a failure. If an alert was already sent within the last alertevery interval, do not send another alert, unless the summary output from the current monitor program differs from the last monitor process. Otherwise, send an alert using each alert program listed for that period. The observe_detail argument to alertevery affects this behavior by observing the changes in the detail part of the output in addition to the summary line. If a monitor has successive failures and the summary output changes in each of them, alertevery will not suppress multiple consecutive alerts. The reasoning is that if the summary output changes, then a significant event occurred and the user should be alerted. The "strict" argument to alertevery will suppress both comparing the output from the previous monitor run to the current and prevent a successful return value of the monitor from resetting the alertevery timer. For example, "alertevery 24h strict" will only send out an alert once every 24 hours, regardless of whether the monitor output changes, or if the service stops and then starts failing.
Alert programs are found in the path supplied with the -a parameter, or in the /usr/local/lib/mon/alert.d and directories if not specified. They are invoked with the following command-line parameters:
The remaining arguments are supplied from the trailing parameters in the configuration file, after the "alert" service parameter.
As with monitor programs, alert programs are invoked with environment variables defined by the user in the service definition, in addition to the following which are explicitly set by the server:
The first line from standard input must be used as a brief summary of the problem, normally supplied as the subject line of an email, or text sent to an alphanumeric pager. Interpretation of all subsequent lines read from stdin is left up to the alerting program. The usual parameters are a list of recipients to deliver the notification to. The interpretation of the recipients is not specified, and is up to the alert program.
The configuration file consists of zero or more global variable definitions, zero or more hostgroup definitions, and one or more watch definitions. Each watch definition may have one or more service definitions. A watch definition is terminated by a blank line, another definition, or the end of the file. A line beginning with optional leading whitespace and a pound ("#") is regarded as a comment, and is ignored.
Lines are parsed as they are read. Long lines may be continued by ending them with a backslash ("\"). If a line is continued, then the backslash, the trailing whitespace after the backslash, and the leading whitespace of the following line are removed. The end result is assembled into a single line.
Typically the configuration file has the following layout:
1. Global variable definitions
2. Hostgroup definitions
3. Watch definitions
See the "etc/example.cf" file which comes for the distribution for an example.
The following variables may be set to override compiled-in defaults. Command-line options will have a higher precedence than these definitions.
Multiple alert paths may be specified by separating them with a colon. Non-absolute paths are taken to be relative to the base directory (/usr/lib/mon by default).
When the configuration file is read, all alerts referenced from the configuration will be looked up in each of these paths, and the full path to the first instance of the alert found is stored in a hash. This hash is only generated upon startup or after a "reset" command, so newly added alert scripts will not be recognized until a "reset" is performed.
Multiple alert paths may be specified by separating them with a colon. All paths must be absolute.
When the configuration file is read, all monitors referenced from the configuration will be looked up in each of these paths, and the full path to the first instance of the monitor found is stored in a hash. This hash is only generated upon startup or after a "reset" command, so newly added monitor scripts will not be recognized until a "reset" is performed.
If type is getpwnam, then the standard Unix passwd file authentication method will be used (calls getpwnam(3) on the user and compares the crypt(3)ed version of the password with what it gets from getpwnam). This will not work if shadow passwords are enabled on the system.
If type is userfile, then usernames and hashed passwords are read from userfile, which is defined via the userfile configuration variable.
If type is pam, then PAM (pluggable authentication modules) will be used for authentication. The service specified by the pamservice global will be used. If no global is given, the PAM passwd service will be used.
If type is trustlocal, then if the client connection comes from locahost, the username passed from the client will be trusted, and the password will be ignored. This can be used when you want the client to handle the authentication for you. I.e. a CGI script using one of the many apache authentication methods.
Blank lines and lines beginning with # are ignored.
serverbind and trapbind specify which address to bind the server and trap ports to, respectively. If these are not defined, the default address is INADDR_ANY, which allows connections on all interfaces. For security reasons, it could be a good idea to bind only to the loopback interface.
timenoticed group service firstfail downtime interval summary.
timenoticed is the time(2) the service came back up.
group service is the group and service which failed.
firstfail is the time(2) when the service began to fail.
downtime is the number of seconds the service failed.
interval is the frequency (in seconds) that the service is polled.
summary is the summary line from when the service was failing.
Turns downtime logging on or off. The default is off.
This is a global setting which controls the default settings for the service-specified variable.
This is a global setting which controls the default settings for the service-specified variable.
If set to "yes", startupalerts will be invoked when the reset client command is executed. The default is "no".
If set, this external program will be called by Mon when various client requests are processed. This can be used to propagate those changes from one Mon server to another, if you have multiple monitoring machines. An example script, monremote.pl is available in the clients directory.
Hostgroup entries begin with the keyword hostgroup, and are followed by a hostgroup tag and one or more hostnames or IP addresses, separated by whitespace. The hostgroup tag must be composed of alphanumeric characters, a dash ("-"), a period ("."), or an underscore ("_"). Non-blank lines following the first hostgroup line are interpreted as more hostnames. The hostgroup definition ends with a blank line. For example:
hostgroup servers nameserver smtpserver nntpserver nfsserver httpserver smbserver hostgroup router_group cisco7000 agsplus
View entries begin with the keyword view, and are followed by a view tag and the names of one or more hostgroups. The view tag must be composed of alphanumeric characters, a dash ("-"), a period ("."), or an underscore ("_"). Non-blank lines following the first view line are interpreted as more hostgroup names. The view definition ends with a blank line. For example:
view servers dns-servers web-servers file-servers
mail-servers view network-services routers switches vpn-servers
Watch entries begin with a line that starts with the keyword watch, followed by whitespace and a single word which normally refers to a pre-defined hostgroup. If the second word is not recognized as a hostgroup tag, a new hostgroup is created whose tag is that word, and that word is its only member.
Watch entries consist of one or more service definitions.
A watch group is terminated by a blank line, the end of the file, or by a subsequent definition, "watch", "hostgroup", or otherwise.
There may be a special watch group entry called "default". If a default watch group is defined with a service entry named "default", then this definition will be used in handling traps received for an unrecognized watch and service.
The components of a service are an interval, monitor, and one or more time period definitions, as defined below.
If a service name of "default" is defined within a watch group called "dafault" (see above), then the default/default definition will be used for handling unknown mon traps.
The following configuration parameters are valid only following a service definition:
Before evaluation, the following substitutions on the expression occur: phrases which look like "group:service" are substituted with the value of the current operational status of that specified service. These opstatus substitutions are computed recursively, so if service A depends upon service B, and service B depends upon service C, then service A depends upon service C. Successful operational statuses (which evaluate to "1") are "STAT_OK", "STAT_COLDSTART", "STAT_WARMSTART", and "STAT_UNKNOWN". The word "SELF" (in all caps) can be used for the group (e.g. "SELF:service"), and is an abbreviation for the current watch group.
This feature can be used to control alerts for services which are dependent on other services, e.g. an SMTP test which is dependent upon the machine being ping-reachable.
Alert suppression. If this option is set to "a", then the dependency expression will be evaluated after the monitor for the service exits or after a trap is received. An alert will only be sent if the evaluation succeeds, meaning that none of the nodes in the dependency graph indicate failure.
Monitor suppression. If it is set to "m", then the dependency expression will be evaulated before the monitor for the service is about to run. If the evaulation succeeds, then the monitor will be run. Otherwise, the monitor will not be run and the status of the service will remain the same.
Host suppression. If it is set to "hm" then Mon will extract the list of "parent" services from the dependency expression. (In fact the expression can be just a list of services.) Then when the monitor for the service is about to be run, for each host in the current hostgroup Mon will search all the parent services which are currently failing and look for the hostname in the current summary output. If the hostname is found, this host will be excluded from this run of the monitor. This can be used to e.g. allow an SMTP test on a group of hosts to still be run even when a single host is not ping-reachable. If all the rest of the hosts are working fine, the service will be in an OK state, but if another host fails the SMTP test Mon can still alert about that host even though the parent dependency was failing. The dependency expression will not be used recursively in this case.
Periods are used to define the conditions which should allow alerts to be delivered.
The second form requires a label followed by a period specification, as defined above. The label is a tag consisting of an alphabetic character or underscore followed by zero or more alphanumerics or underscores and ending with a colon. This form allows multiple periods with the same period definition. One use is to have a period definition which has no alertafter or alertevery parameters for a particular time period, and another for the same time period with a different set of alerts that does contain those parameters.
Period definitions, in either the first or second form, must be unique within each service definition. For example, if you need to define two periods both for "wd {Sun-Sat}", then one or both of the period definitions must specify a label such as "period t1: wd {Sun-Sat}" and "period t2: wd {Sun-Sat}".
In the second form, the arguments are a positive integer followed by an interval, as described by the interval variable above. If these parameters are specified, then the alerts for that period will only be called after that many failures happen within that interval. For example, if alertafter is given the arguments "3 30m", then the alert will be called if 3 failures happen within 30 minutes.
In the third form, the argument is an interval, as described by the interval variable above. Alerts for that period will only be called if the service has been in a failure state for more than the length of time desribed by the interval, regardless of the number of failures noticed within that interval.
This variable tells the server to call no more than num alerts during a failure. The alert counter is kept on a per-period basis, and is reset upon each success.
If this option is specified, then upalerts will be called whenever the service state changes from failure to success, rather than only after a corresponding "down" alert.
See the ALERT PROGRAMS section above for a list of the pramaeters mon will pass automatically to alert programs.
The file specified by the authfile variable in the configuration file (or passed via the -A parameter) will be loaded upon startup. This file defines restrictions upon which client commands may be executed by which users. It is a text file which consists of comments, command definitions, and trap authentication parameters. A comment line begins with optional whitespace followed by pound sign. Blank lines are ignored.
The file is separated into a command section and a trap section. Sections are specified by a single line containing one of the following statements:
command section
or
trap section
Lines following one of the above statements apply to that section until either the end of the file or another section begins.
A command definition consists of a command, followed by a colon, followed by a comma-separated list of users who may execute the command. The default is that no users may execute any commands unless they are explicitly allowed in this configuration file. For clarity, a user can be denied by prefixing the user name with "!". If the word "AUTH_ANY" is used for a username, then any authenticated user will be allowed to execute the command. If the word "all" is used for a username, then that command may be executed by any user, authenticated or not.
The trap section allows configuration of which users may send traps from which hosts. The syntax is a source host (name or ip address), whitespace, a username, whitespace, and a plaintext password for that user. If the source host is "*", then allow traps from any host. If the username is "*", then accept traps without regard for the username or password. If no hosts or users are specified, then no traps will be accepted.
An example configuration file:
command section list: all reset: root,admin loadstate: root savestate: root trap section 127.0.0.1 root r@@tp4sswrd
This means that all clients are able to perform the list command, "root" is able to perform "reset", "loadstate", "savestate", and "admin" is able to execute the "reset" command.
The server listens on TCP port 2583, which may be overridden using the -p port option. Commands are a single line each, terminated by a newline. The server can handle any number of simultaneous client connections.
See manual page for moncmd.
Mon has the facility to receive special "mon traps" from any local or remote machine. Currently, the only available method for sending mon traps are through the Mon::Client perl interface, though the UDP packet format is defined well enough to permit the writing of traps in other languages.
Traps are handled similarly to monitors: a trap sends an operational status, summary line, and description text, and mon generates an alert or upalert as necessary.
Traps can be caught by any watch/service group set up in the mon configuration file, however it is suggested that you configure watch/service groups specifically for the traps you expect to receive. When defining a special watch/service group for traps, do not include a "monitor" directive (as no monitor need be invoked). Since a monitor is not being invoked, it is not necessary for the watch definition to have a hostgroup which contains real host names. Just make up a useful name, and mon will automatically create the watch group for you.
Here is a simple config file example:
watch trap-service
service host1-disks
description TRAP: for host1 disk status
period wd {Sun-Sat}
alert mail.alert someone@your.org
upalert mail.alert -u someone@your.org
Since mon listens on a UDP port for any trap, a default facility is available for handling traps to unknown groups or services. To enable this facility, you must include a "default" watch group with a "default" service entry containing the specifics of alarms. If a default/default watch group and service are not configured, then unknown traps get logged via syslog, and no alarm is sent. NOTE: The default/default facility is a single entity as far as accounting and alarming go. Alarm programs which are not aware of this fact may send confusing information when a failure trap comes from one machine, followed by a success (ok) trap from a different machine. See the alarm environment variable MON_TRAP_INTENDED above for a possible way around this. It is intended that default/default be used as a facility to catch unknown traps, and should not be relied upon to catch all traps in a production environment. If you are lazy and only want to use default/default for catching all traps, it would be best to disable upalerts, and use the MON_TRAP_INTENDED environment variable in alert scripts to make the alerts more meaningful to you.
Here is an example default facility:
watch default
service default
description Default trap service
period wd {Sun-Sat}
alert mail.alert someone@your.org
upalert mail.alert -u someone@your.org
The mon distribution comes with an example configuration called example.cf. Refer to that file for more information.
mon was written because I couldn't find anything out there that did just what I needed, and nothing was worth modifying to add the features I wanted. It doesn't have a cool name, and that bothers me because I couldn't think of one.
Report bugs to the email address below.
Jim Trocki <trockij@arctic.org>
| $Date: 2007/06/25 13:10:07 $ | Linux |