#!/usr/bin/perl
#
# Copyright 2012 Chris Wilson
# Copyright 2006 Holger Levsen
#
# This plugin monitors ALL processes on a system. No exceptions. It can
# produce very big graphs! But if you want to know which processes are
# killing your system by page faulting, without knowing what to monitor
# in advance, this can help; or in addition to one of the more specific
# plugins to monitor just Apache or MySQL, for example.
#
# Each counter is a DERIVE (difference since the last counter reading)
# of the number of major page faults, usually 4k each, read in by a
# process. Memory mapped files probably contribute to this. The process
# cannot continue until the page fault is served, so this is a
# high-priority read that usually indicates memory starvation.
# Processes with no page faults at all are ignored. Processes
# that die may not appear on the graph, and anyway their last chunk of
# CPU usage before they died is lost. You could modify this plugin to
# read SAR/psacct records if you care about that.
#
# This program is free software; you can redistribute it and/or
# modify it under the terms of the GNU General Public License
# as published by the Free Software Foundation; version 2 dated June,
# 1991.

use strict;
use warnings;

my $scriptname = $0;
$scriptname =~ s|.*/||;
my $fieldname = ($scriptname =~ /^total_by_process_(.*)_(.*)/) ? $1 : undef;
my $fieldtype = ($scriptname =~ /^total_by_process_(.*)_(.*)/) ? $2 : undef;

if (defined($ARGV[0]) and $ARGV[0] eq "suggest")
{
	system("ps L | cut -d' ' -f1");
	exit(0);
}

if (!$fieldname)
{
	print STDERR "Must be used with a PS format specifier name; try '$0 suggest'";
	exit(2);
}

unless ($fieldtype =~ /^(GAUGE|DERIVE)$/)
{
	print STDERR "Unknown field type $fieldtype: should be GAUGE or DERIVE";
	exit(2);
}

my $cmd = "ps -eo $fieldname,comm h";
open PS, "$cmd|" or die "Failed to run ps command: $cmd: $!";

# my $header_line = <PS>;
my %total_by_process;

while (<PS>)
{
	my @fields = split;
	my $value = $fields[0];
	my $process = $fields[1];

	# remove any / and everything after it from the process name,
	# e.g. kworker/0:2 -> kworker
	$process =~ s|/.*||;

	# remove any . at the end of the name (why does this appear?)
	# $process =~ s|\.$||;

	# change any symbol that's not allowed in a munin variable name to _
	$process =~ tr|a-zA-Z0-9|_|c;

	$total_by_process{$process} += $value;
}

foreach my $process (keys %total_by_process)
{
	# remove all processes with 0 faults
	if (not $total_by_process{$process})
	{
		delete $total_by_process{$process};
	}
}

close(PS);

if (defined($ARGV[0]) and $ARGV[0] eq "config")
{
	print <<END;
graph_title $fieldname by Process
graph_category processes
graph_info Shows total of $fieldname (reported by ps) for each process name
graph_vlabel $fieldname (from ps)
END

	# graph_args --base 1000
	# graph_vlabel seconds

	my $stack = 0;
	sub draw() { return $stack++ ? "STACK" : "AREA" }
	print map
	{
		"$_.label $_\n" .
		"$_.min 0\n" .
		"$_.type $fieldtype\n" .
		"$_.draw " . draw() . "\n"
	} sort keys %total_by_process;
}
else
{
	print map
	{
		"$_.value $total_by_process{$_}\n"
	} sort keys %total_by_process;
}

exit(0);