Process Resource Monitor (PRM) v1.1.4
             (C) 2002-2012, R-fx Networks <[email protected]>
             (C) 2012, Ryan MacDonald <[email protected]>
 This program may be freely redistributed under the terms of the GNU GPL v2

::::::::::::::::::::::::::::::::::

:: CONTENTS ::
.: 1 [ DESCRIPTION ]
.: 2 [ FEATURES ]
.: 3 [ CONFIGURATION ]
.: 4 [ RULE FILES ]
.: 5 [ IGNORE OPTIONS ]
.: 6 [ CRON ]

::::::::::::::::::::::::::::::::::

.: 1 [ DESCRIPTION ]
Process Resource Monitor (PRM) is a CPU, Memory, Processes & Run (Elapsed) Time 
resource monitor for Linux & BSD. The flexibility of PRM is achieved through
global scoped resource limits or rule-based per-process / per-user limits. A
great deal of control of PRM is maintained through a number of ignore options,
ability to configure soft/hard kill triggers, wait/recheck timings and to send
kill signals to parent/children process trees. Additionally, the status output
is very verbose by default with support for sending log data through syslog.

There is no shortage of usage methods for PRM, especially when leveraging the
ignore and kill options with the rule based system. You can easily take control
of those run-away aspell/pspell processes eating 80% of memory, errant exim
processes threading off 500 processes or prevent scripts executed under apache
from running for hours at a time.

.: 2 [ FEATURES ]
- global resource limits
- per-process/per-user rule based resource limits
- rules only run mode
- alert only run mode
- ignore root processes option
- process list global ignore file
- user based global ignore file
- command based global ignore file
- regex based global and/or per-rule ignore variable
- global scoped resource limits
- per-process or per user rule based resource limits
- set custom kill signals (i.e: SIGHUP)
- parent process kill option to terminate an entire process tree
- kill trigger/wait times allow rechecking usage over a period of time
- kill restart option to execute a custom restart command
- all kill/resource/ignore options can be global or rule defined
- easy to configure percentage of total CPU & MEM limits
- total number of processes limits
- elapsed run time limits

.: 3 [ CONFIGURATION ]
The configuration file is very well commented so there is not a whole lot to
cover here but for the sake of clarity, I will list some of the more important
variables and expand on there usage (the values here are the defaults).

RULES_ONLY="1"
This tells PRM that we will ONLY check limits against processes that we can find
a matching rule for. When this is not set, all processes will be checked against
the global resource limits defined in conf.prm, however, if a rule match for a
process is found, it will still be processed under the respective rule.

IGNORE_ROOT="1"
This tells PRM to ignore any root owned processes, however, when the KILL_PARENT
option is set, root owned parent processes will still be subject to kills. As
a saftey measure there is a KILL_MINPID variable in the internals.conf file that
controls the lowest possible PID that PRM will kill, default is 150. This is
intended to prevent PRM from trying to kill important parent threaders such as
init.

IGNORE=""
This is an important option and is the recommended method for ignoring when
using the rules system. The accepted values are basic and extended regexp, which
use pipes (|) as a spacer. For example, IGNORE="^httpd$|^sendmail$", when placed
in a rule called nobody.user, would ignore any processes under user nobody that
have the command name of exactly "httpd" or "sendmail". It is important that you
try to use the ^NAME$ convention as this limits the scope of the ignore option
and can prevent some rather weird and unexpected behavior.

MAX_ETIME=""
This is the elapsed run time for a process, there really isnt much to this value
except that the format requires a bit of explaining. The accepted format is in
the form of DD-HH:MM:ss, this can be broken apart or kept in full length such
as the following example values:
1h = 1:00:00, 1h = 00-01:00:00, 1d = 1-00:00:00, 30m = 30:00

KILL_TRIG="3"
KILL_WAIT="10"
These values control the soft rechecks of a process, allowing for a process to
have its resources rechecked TRIG times with WAIT time between checks, giving a
bit of margin for a process to "burst" resources and come back into normal use.
The max time to kill a process is equal to TRIG*WAIT, these values can be set 0
if you want to instantly kill offending processes.

KILL_PARENT="1"
This is an important option that should in most cases be enabled, it allows for
the parent process and children of the parent to be killed. This is important as
when a process is created by a parent threader (such as apache), when the child
disappears, the parent will simply fork off a new child thread to replace it.
With this enabled, the parent process (ppid) is killed and all child processes
of the parent. As a saftey measure, ppid's lower than KILL_MINPID set in the
internals.conf file (default 150), are never killed. In situations where uptime
is of importance, such as an apache rule, you can use KILL_RESTART_CMD to have
a command executed after the kills are completed.

KILL_SIG="9"
In some custom situations, it may be desired to send signals other than SIGTERM
(9) to a process or parent/child process tree. An example of this would be a
service such as apache or named that will take a SIGHUP and complete all pending
requests before conducting a reload. For a list of signals please see 'kill -l',
only the numeric representation of a signal is accepted.

KILL_RESTART_CMD=""
This is an optional command executed after kill signals are sent, the intended
goal of this feature is to send restart commands to a service for the purpose
of maintaining uptime. An example here would be on an apache rule, you can have
the option set something like: KILL_RESTART_CMD="/etc/init.d/httpd restart"

.: 4 [ RULE FILES ]
The rules system has two methods of use, the first is a user based rule and the
second is a process command based rule. The naming of the rules is very specific
and PRM will look for only the rules that match a specific processes user or cmd
values.

The rules path is located at /usr/local/prm/rules/ and the naming conventions
are as follows:

USERNAME.user
COMMAND.cmd

For example, to create a rule for user "mike" you would create rules/mike.user,
to create a rule for apache processes you would create rules/httpd.cmd or for
the apache user "nobody" you would create rules/nobody.user. It is important to
remember that .cmd rules only apply to the specific command process but user
rules apply to ALL processes running under that user. 

The following variables can be used in rule files, please see conf.prm comments
for description of what each variable does, any variable not included in a rule
file will have its value set from conf.prm (# can be used as comments):

IGNORE
MAX_CPU
MAX_MEM
MAX_PROC
MAX_ETIME
KILL_TRIG
KILL_WAIT
KILL_PARENT
KILL_SIG
KILL_RESTART_CMD

Lets look at a few example rules. The first is an apache user rule under user
"nobody", the goal of this rule is to control processes that fork from scripts
on user websites, to make sure they stay within certain resource limits and to
make sure they do not run beyond 10 minutes. Since we only care about what is
being forked off of apache, we want to ignore the httpd threads themselves and
any other trusted apache related processes such as suphp, cgiwrap and suexec.

file rules/nobody.user:
IGNORE="^httpd$|^suexec$|^suphp$|^cgiwrap$|^spamd$"
MAX_CPU="50"
MAX_MEM="10"
MAX_PROC="25"
MAX_ETIME="10:00"
KILL_TRIG="3"
KILL_WAIT="10"
# We need to kill parent here otherwise the HTTP Request that spawned the
# script we are trying to kill, will probably just respawn it.
KILL_PARENT="1"
KILL_SIG="9"
KILL_RESTART_CMD="/etc/init.d/httpd restart"

The second rule, is one intended to stop run away java processes on a custom
application from consuming all memory by forking off lots of java processes,
each using a bunch of memory.

file rules/java.cmd:

IGNORE=""
MAX_CPU="99"
MAX_MEM="30"
MAX_PROC="100"
# we dont care about the process run time, set value 0 to disable check
MAX_ETIME="0"
KILL_TRIG="3"
# we want to set a bit longer soft rechecks as sometimes the problem fixes
# itself
KILL_WAIT="20"
KILL_PARENT="1"
KILL_SIG="9"
KILL_RESTART_CMD="/etc/init.d/openfire restart"

The third and final example is for a sendmail server where web scripts that
call the sendmail executable directly can sometimes cause hundres of processes
to spawn.

file rules/sendmail.cmd
IGNORE=""
MAX_CPU="0"
MAX_MEM="0"
# we only care about process counts, so we zero the other MAX_ settings
MAX_PROC="100"
MAX_ETIME="0"
KILL_TRIG="3"
KILL_WAIT="10"
KILL_PARENT="1"
KILL_SIG="9"
KILL_RESTART_CMD="/etc/init.d/sendmail restart"

.: 5 [ IGNORE OPTIONS ]
There is a number of ignore options available for PRM, they fall into the static
ignore files and then the dynamic IGNORE variable that can be included into rule
files.

The static ignore files and there usage are as follows:

ignore_pslist
This is a top level ignore file applied when the /bin/ps command is run, this is
not recommended to be used unless you absolutely must. Ignore strings in this
file should be very specific and caution taken as short strings may get broadly
applied across the process list. 

ignore_cmd
This ignore file applys directly to the command value of processes and avoids
broadly applied ignore values that are an issue with the _pslist ignore file.
When a process is found to match an entry in this file, it will stop processing
and PRM will move on.

ignore_users 
This ignore file applys directly to the username value of processes and avoids
broadly applied ignore values that are an issue with the _pslist ignore file.
When a process is found to match an entry in this file, it will stop processing
and PRM will move on.

The hands down best way to create ignore conditions is with the IGNORE variable
inside of rule files, this variable allows for exact definition of what should
be ignored. The IGNORE variable is applied against the user and cmd values of
the process, whichever matches first, in the order of user>command. 

IGNORE=""
This is an important option and is the recommended method for ignoring when
using the rules system. The accepted values are basic and extended regexp, which
use pipes (|) as a spacer. For example, IGNORE="^httpd$|^sendmail$", when placed
in a rule called nobody.user, would ignore any processes under user nobody that
have the command name of exactly "httpd" or "sendmail". It is important that you
try to use the ^NAME$ convention as this limits the scope of the ignore option
and can prevent some rather weird and unexpected behavior.

.: 6 [ CRON ]
The default execution of PRM is handled through /etc/cron.d/prm and set to run
at 5 minute intervals, you may increase this at your own discretion. You are
warned though, that if your system has hundreds of processes with many process
events on a run of PRM, it can take a couple of minutes to complete.

As a measure of protection from PRM stepping on its own toes, there is a lock
file feature that times out after 10 minutes, you can change the lock timeout
in internals.conf using the LOCK_TIMEOUT variable.