viewvc.cgi/mon-modules/parse_log.cgi

#!/usr/bin/perl -w

# parse file.alert mon logs and report (up|down)time of services
#
# 2003-09-03 Dobrica Pavlinusic <dpavlin@rot13.org>
# 2003-10-05 converted to CGI script
#

use strict;
use POSIX qw(strftime);
use CGI qw/:standard *table/;
use CGI::Carp qw(fatalsToBrowser);
use Data::Sorting qw(:arrays);
use Time::ParseDate;
use Time::Available;

use Data::Dumper;

my $date_fmt = "%Y-%m-%d";
my $date_time_fmt = "%Y-%m-%d %H:%M:%S";

my $from_date = "now - 6 months";
my $to_date = "now";

# working days definition (1-7; mon=1)
my $dayMask = Time::Available::DAY_WEEKDAY;
# working hours
my $from_time_interval = "7:00";
my $to_time_interval = "17:00";

my $debug=1;
$debug++ if (grep(/-v/,@ARGV));
$debug++ if (grep(/-d/,@ARGV));

my %days = (
        Time::Available::DAY_MONDAY=>'Mo',
        Time::Available::DAY_TUESDAY=>'Tu',
        Time::Available::DAY_WEDNESDAY=>'We',
        Time::Available::DAY_THURSDAY=>'Th',
        Time::Available::DAY_FRIDAY=>'Fr',
        Time::Available::DAY_SATURDAY=>'Sa',
        Time::Available::DAY_SUNDAY=>'Su'
);

my $q = new CGI;

my $print_orphans = $q->param('print_orphans') || 0;
my $rep_reset = $q->param('rep_reset') || 0;
my @sg_selected = $q->param('sg_filter');

# init misc sort parametars
my @sort;
my $order;
my %sort_param;
my ($usort,$dsort);
if ($q->param('usort')) {
        $sort_param{'usort'} = $q->param('usort');
        $q->delete('usort');
        @sort = ( -compare => 'numeric', $sort_param{'usort'} );
}
if ($q->param('dsort')) {
        $sort_param{'dsort'} = $q->param('dsort');
        $q->delete('dsort');
        @sort = ( -compare => 'numeric', -order=>'reverse', $sort_param{'dsort'} );
}

# make interval
my $working_days;
if ($q->param('use_time_limit')) {
        $dayMask=0;
        foreach my $dm ($q->param('day_interval')) {
                $dayMask |= $dm;
        }
        $working_days=new Time::Available(start=>$q->param('from_time_interval'),end=>$q->param('to_time_interval'),dayMask=>$dayMask);
}

#
# This option (activated via command switch -r) will reset failure duration
# if repeated failure on same group/service happend.
# If you want honest reporting (or grouped only by group and service),
# you souldn't turn it on :-) However, if you have just failure events in your
# log, this will produce output which will show duration BETWEEN two failures
#

# pretty format date
sub d {
        my $utime = shift || return "?";
        if ($debug) {
                return strftime($date_time_fmt." [%s]",localtime($utime));
        } else {
                return strftime($date_time_fmt,localtime($utime));
        }
}
# pretty format duration
sub dur {
        my $dur = shift || return "0";
        my $out = "";

        my $s = $dur;
        my $d = int($s/(24*60*60));
        $s = $s % (24*60*60);
        my $h = int($s/(60*60));
        $s = $s % (60*60);
        my $m = int($s/60);
        $s = $s % 60;

        $out .= $d."d " if ($d > 0);
        if ($debug) {
                $out .= sprintf("%02d:%02d:%02d [%d]",$h,$m,$s, $dur);
        } else {
                $out .= sprintf("%02d:%02d:%02d",$h,$m,$s);
        }

        return $out;
}

# read log and calculate
#

my %fail;
my %downtime;   # total downtime
my %sg_filter;  # filter for service/group
my %sg_count;   # count number of downtimes

my $log_file="/home/dpavlin/mon-log/sap.log";

my @data;

open(LOG, $log_file) || die "$log_file: $!";

while(<LOG>) {
        chomp;
        if (/^(failure|up)\s+(\S+)\s+(\S+)\s+(\d+)\s+\(([^)]+)\)\s+(.+)$/) {
                my ($status,$group,$service,$utime,$date,$desc) = ($1,$2,$3,$4,$5,$6);
                my $id = "$group/$service";
                if ($status eq "up" && defined($fail{$id})) {
                        if (grep(m;$group/$service;,@sg_selected)) {
                                push @data, {
                                        'sg'=>"$group/$service",
                                        'from'=>$fail{$id},
                                        'to'=>$utime,
                                        'dur'=>($utime - $fail{$id}),
                                        'int'=>$working_days->interval($utime,$fail{$id}),
                                        'desc'=>$desc };
                                $downtime{"$group/$service"} += ($utime - $fail{$id}),
                                $sg_count{"$group/$service"}++;
                        }
                        $sg_filter{"$group/$service"}++;
                        delete $fail{$id};
                } elsif ($status eq "up") {
                        if ($print_orphans && grep(m;$group/$service;,@sg_selected)) {
                                push @data, {
                                        'sg'=>"$group/$service",
                                        'from'=>-1,
                                        'to'=>$utime,
                                        'desc'=>$desc };
                                $sg_count{"$group/$service"}++;
                        }
                        delete $fail{$id};
                        $sg_filter{"$group/$service"}++;
                } elsif (defined($fail{$id})) {
                        if ($rep_reset && grep(m;$group/$service;,@sg_selected)) {
                                push @data, {
                                        'sg'=>"$group/$service",
                                        'from'=>$fail{$id},
                                        'to'=>$utime,
                                        'dur'=>($utime - $fail{$id}),
                                        'int'=>$working_days->interval($utime,$fail{$id}),
                                        'desc'=>'[failure again]'};
                                $downtime{"$group/$service"} += ($utime - $fail{$id}),
                                $fail{$id} = $utime;
                                $sg_count{"$group/$service"}++;
                        }
                        $sg_filter{"$group/$service"}++;
                } else {
                        $fail{$id} = $utime;
                }
        }
}
close(LOG);

# generate output
#
print header,start_html("mon availiability report");

# make some filters
#

print start_form,
        start_table({-border=>0,-cellspacing=>0,-cellpadding=>0}),
        Tr(td(
        em("Show just service/group:"),br,
        checkbox_group(-name=>'sg_filter',
                -values=>[keys %sg_filter],
                -default=>[keys %sg_filter],
                -linebreak=>'true',
                ),
        ),td(
        em("Other options:"),br,
        $q->checkbox(-name=>'rep_reset',-checked=>0,
        -label=>"show repeated failures on same service as individual failures"),
        br,
        $q->checkbox(-name=>'print_orphans',-checked=>0,
        -label=>"show records which are not complete in this interval"),
        br,
        $q->checkbox(-name=>'use_date_limit',-checked=>1,
        -label=>"use date limit from:"),
        $q->textfield(-name=>'from_date',-size=>20,-default=>$from_date),
        " to: ",
        $q->textfield(-name=>'to_date',-size=>20,-default=>$to_date),
        small('Using <a href="http://search.cpan.org/search?mode=module&query=Time::ParseDate">Time::ParseDate</a>'),
        br,
        $q->checkbox(-name=>'use_time_limit',-checked=>1,
        -label=>"use time limit for each day:"),
        $q->textfield(-name=>'from_time_interval',-size=>8,-default=>$from_time_interval),
        " to: ",
        $q->textfield(-name=>'to_time_interval',-size=>8,-default=>$to_time_interval),
        br,"Days: ",
        $q->checkbox_group(-name=>'day_interval',
                -values=>[ sort { $a <=> $b } keys %days ],
                -labels=>\%days,
                -defaults=>[ 
                        Time::Available::DAY_MONDAY,
                        Time::Available::DAY_TUESDAY,
                        Time::Available::DAY_WEDNESDAY,
                        Time::Available::DAY_THURSDAY,
                        Time::Available::DAY_FRIDAY,
                ]
                ),
        $q->submit(-name=>'show',-value=>'Show report'),
        )),end_table;

# dump report
#

sub sort_link {
        my $q = shift || return;
        my $col = shift || return;
        my $dir = lc(shift) || return;
        if ($sort_param{$dir.'sort'} && $sort_param{$dir.'sort'} eq $col) {
                return '&'.$dir.'Arr;';
        } else {
                return '<a href="'.$q->url(-query=>1).'&'.$dir.'sort='.$col.'">&'.$dir.'Arr;</a>';
        }
}


my ($from_time,$to_time,$from_html,$to_html);
if ($q->param('use_date_limit')) {
        $from_time = parsedate($q->param('from_date'), UK=>1);
        $to_time = parsedate($q->param('to_date'), UK=>1);
        $from_html = strftime($date_fmt,localtime($from_time));
        $to_html = strftime($date_fmt,localtime($to_time));
        $from_html .= " [$from_time] " if ($debug);
        $to_html .= " [$to_time] " if ($debug);
}

# sort data
#
my @sorted = sorted_array(@data, @sort);
#my @sorted = @data;

print "-- sort: ",Dumper(@sort)," (data: ".@data." sorted: ".@sorted.") --\n",br,"-- dayMask: $dayMask --\n",br if ($debug);

print start_table({-border=>1,-cellspacing=>0,-cellpadding=>2,-width=>'100%'});

print Tr(
                th("group/service"),
                th({-bgcolor=>'#f0f0f0'},
                        &sort_link($q,'from','u').' from '.
                        &sort_link($q,'from','d'),
                        br,$from_html
                ),
                th(
                        &sort_link($q,'to','u').' to '.
                        &sort_link($q,'to','d'),
                        br,$to_html
                ),
                th({-bgcolor=>'#e0e0e0'},
                        &sort_link($q,'dur','u').' duration '.
                        &sort_link($q,'dur','d')
                ),
                th("description")
        ) if (scalar @sorted > 0);

foreach my $row (@sorted) {
        next if ($q->param('use_date_limit') && ($row->{from} < $from_time || $row->{to} > $to_time));
        my ($from,$dur,$int) = ('unknown','unknown','unknown');
        if ($row->{from} != -1 ) {
                $from = d($row->{from});
                $dur = dur($row->{dur});
                $int = dur($row->{int});
        }
        print Tr(
                td({-align=>'left',-valign=>'center'},$row->{sg}),
                td({-align=>'right',-bgcolor=>'#f0f0f0'},$from),
                td({-align=>'right'},d($row->{to})),
                td({-align=>'center',-bgcolor=>'#e0e0e0'},$dur),
                td({-align=>'center',-bgcolor=>'#e0e0e0'},$int),
                td({-align=>'left'},$row->{desc}),
                ),"\n";
}

# dump totals
#

foreach my $sg (keys %downtime) {
print Tr(td({-colspan=>3,-align=>'right'},"total for $sg:"),
        td({-bgcolor=>'#e0e0e0',-align=>'right'},dur($downtime{$sg})),
        td(small("in ".$sg_count{$sg}." failures"))),"\n";
}

print end_table,
        end_form;

1	dpavlin	1.1	#!/usr/bin/perl -w
2
3			# parse file.alert mon logs and report (up\|down)time of services
4			#
5			# 2003-09-03 Dobrica Pavlinusic <dpavlin@rot13.org>
6			# 2003-10-05 converted to CGI script
7			#
8
9			use strict;
10			use POSIX qw(strftime);
11			use CGI qw/:standard *table/;
12			use CGI::Carp qw(fatalsToBrowser);
13			use Data::Sorting qw(:arrays);
14	dpavlin	1.2	use Time::ParseDate;
15	dpavlin	1.3	use Time::Available;
16	dpavlin	1.2
17	dpavlin	1.1	use Data::Dumper;
18
19	dpavlin	1.2	my $date_fmt = "%Y-%m-%d";
20			my $date_time_fmt = "%Y-%m-%d %H:%M:%S";
21
22			my $from_date = "now - 6 months";
23			my $to_date = "now";
24	dpavlin	1.1
25			# working days definition (1-7; mon=1)
26	dpavlin	1.3	my $dayMask = Time::Available::DAY_WEEKDAY;
27	dpavlin	1.1	# working hours
28	dpavlin	1.3	my $from_time_interval = "7:00";
29			my $to_time_interval = "17:00";
30	dpavlin	1.1
31	dpavlin	1.2	my $debug=1;
32	dpavlin	1.1	$debug++ if (grep(/-v/,@ARGV));
33			$debug++ if (grep(/-d/,@ARGV));
34
35	dpavlin	1.3	my %days = (
36			Time::Available::DAY_MONDAY=>'Mo',
37			Time::Available::DAY_TUESDAY=>'Tu',
38			Time::Available::DAY_WEDNESDAY=>'We',
39			Time::Available::DAY_THURSDAY=>'Th',
40			Time::Available::DAY_FRIDAY=>'Fr',
41			Time::Available::DAY_SATURDAY=>'Sa',
42			Time::Available::DAY_SUNDAY=>'Su'
43			);
44
45	dpavlin	1.1	my $q = new CGI;
46
47			my $print_orphans = $q->param('print_orphans') \|\| 0;
48			my $rep_reset = $q->param('rep_reset') \|\| 0;
49			my @sg_selected = $q->param('sg_filter');
50
51	dpavlin	1.3	# init misc sort parametars
52	dpavlin	1.1	my @sort;
53			my $order;
54			my %sort_param;
55			my ($usort,$dsort);
56			if ($q->param('usort')) {
57			$sort_param{'usort'} = $q->param('usort');
58			$q->delete('usort');
59			@sort = ( -compare => 'numeric', $sort_param{'usort'} );
60			}
61			if ($q->param('dsort')) {
62			$sort_param{'dsort'} = $q->param('dsort');
63			$q->delete('dsort');
64			@sort = ( -compare => 'numeric', -order=>'reverse', $sort_param{'dsort'} );
65			}
66
67	dpavlin	1.3	# make interval
68			my $working_days;
69			if ($q->param('use_time_limit')) {
70			$dayMask=0;
71			foreach my $dm ($q->param('day_interval')) {
72			$dayMask \|= $dm;
73			}
74			$working_days=new Time::Available(start=>$q->param('from_time_interval'),end=>$q->param('to_time_interval'),dayMask=>$dayMask);
75			}
76
77	dpavlin	1.1	#
78			# This option (activated via command switch -r) will reset failure duration
79			# if repeated failure on same group/service happend.
80			# If you want honest reporting (or grouped only by group and service),
81			# you souldn't turn it on :-) However, if you have just failure events in your
82			# log, this will produce output which will show duration BETWEEN two failures
83			#
84
85			# pretty format date
86			sub d {
87			my $utime = shift \|\| return "?";
88	dpavlin	1.3	if ($debug) {
89			return strftime($date_time_fmt." [%s]",localtime($utime));
90			} else {
91			return strftime($date_time_fmt,localtime($utime));
92			}
93	dpavlin	1.1	}
94			# pretty format duration
95			sub dur {
96	dpavlin	1.3	my $dur = shift \|\| return "0";
97	dpavlin	1.1	my $out = "";
98
99			my $s = $dur;
100			my $d = int($s/(246060));
101			$s = $s % (246060);
102			my $h = int($s/(60*60));
103			$s = $s % (60*60);
104			my $m = int($s/60);
105			$s = $s % 60;
106
107			$out .= $d."d " if ($d > 0);
108	dpavlin	1.2	if ($debug) {
109			$out .= sprintf("%02d:%02d:%02d [%d]",$h,$m,$s, $dur);
110			} else {
111			$out .= sprintf("%02d:%02d:%02d",$h,$m,$s);
112			}
113	dpavlin	1.1
114			return $out;
115			}
116
117			# read log and calculate
118			#
119
120			my %fail;
121			my %downtime; # total downtime
122			my %sg_filter; # filter for service/group
123	dpavlin	1.2	my %sg_count; # count number of downtimes
124	dpavlin	1.1
125			my $log_file="/home/dpavlin/mon-log/sap.log";
126
127			my @data;
128
129			open(LOG, $log_file) \|\| die "$log_file: $!";
130
131			while(<LOG>) {
132			chomp;
133			if (/^(failure\|up)\s+(\S+)\s+(\S+)\s+(\d+)\s+\(([^)]+)\)\s+(.+)$/) {
134			my ($status,$group,$service,$utime,$date,$desc) = ($1,$2,$3,$4,$5,$6);
135			my $id = "$group/$service";
136			if ($status eq "up" && defined($fail{$id})) {
137			if (grep(m;$group/$service;,@sg_selected)) {
138			push @data, {
139			'sg'=>"$group/$service",
140	dpavlin	1.3	'from'=>$fail{$id},
141			'to'=>$utime,
142			'dur'=>($utime - $fail{$id}),
143			'int'=>$working_days->interval($utime,$fail{$id}),
144	dpavlin	1.1	'desc'=>$desc };
145			$downtime{"$group/$service"} += ($utime - $fail{$id}),
146	dpavlin	1.2	$sg_count{"$group/$service"}++;
147	dpavlin	1.1	}
148			$sg_filter{"$group/$service"}++;
149			delete $fail{$id};
150			} elsif ($status eq "up") {
151	dpavlin	1.2	if ($print_orphans && grep(m;$group/$service;,@sg_selected)) {
152	dpavlin	1.1	push @data, {
153			'sg'=>"$group/$service",
154	dpavlin	1.3	'from'=>-1,
155			'to'=>$utime,
156	dpavlin	1.1	'desc'=>$desc };
157	dpavlin	1.2	$sg_count{"$group/$service"}++;
158	dpavlin	1.1	}
159			delete $fail{$id};
160			$sg_filter{"$group/$service"}++;
161			} elsif (defined($fail{$id})) {
162	dpavlin	1.2	if ($rep_reset && grep(m;$group/$service;,@sg_selected)) {
163	dpavlin	1.1	push @data, {
164			'sg'=>"$group/$service",
165	dpavlin	1.3	'from'=>$fail{$id},
166			'to'=>$utime,
167			'dur'=>($utime - $fail{$id}),
168			'int'=>$working_days->interval($utime,$fail{$id}),
169	dpavlin	1.1	'desc'=>'[failure again]'};
170			$downtime{"$group/$service"} += ($utime - $fail{$id}),
171			$fail{$id} = $utime;
172	dpavlin	1.2	$sg_count{"$group/$service"}++;
173	dpavlin	1.1	}
174			$sg_filter{"$group/$service"}++;
175			} else {
176			$fail{$id} = $utime;
177			}
178			}
179			}
180			close(LOG);
181
182			# generate output
183			#
184			print header,start_html("mon availiability report");
185
186			# make some filters
187			#
188
189			print start_form,
190			start_table({-border=>0,-cellspacing=>0,-cellpadding=>0}),
191			Tr(td(
192			em("Show just service/group:"),br,
193			checkbox_group(-name=>'sg_filter',
194			-values=>[keys %sg_filter],
195			-default=>[keys %sg_filter],
196			-linebreak=>'true',
197			),
198			),td(
199			em("Other options:"),br,
200			$q->checkbox(-name=>'rep_reset',-checked=>0,
201			-label=>"show repeated failures on same service as individual failures"),
202			br,
203			$q->checkbox(-name=>'print_orphans',-checked=>0,
204			-label=>"show records which are not complete in this interval"),
205			br,
206	dpavlin	1.2	$q->checkbox(-name=>'use_date_limit',-checked=>1,
207			-label=>"use date limit from:"),
208			$q->textfield(-name=>'from_date',-size=>20,-default=>$from_date),
209			" to: ",
210			$q->textfield(-name=>'to_date',-size=>20,-default=>$to_date),
211			small('Using <a href="http://search.cpan.org/search?mode=module&query=Time::ParseDate">Time::ParseDate</a>'),
212			br,
213	dpavlin	1.3	$q->checkbox(-name=>'use_time_limit',-checked=>1,
214			-label=>"use time limit for each day:"),
215			$q->textfield(-name=>'from_time_interval',-size=>8,-default=>$from_time_interval),
216			" to: ",
217			$q->textfield(-name=>'to_time_interval',-size=>8,-default=>$to_time_interval),
218			br,"Days: ",
219			$q->checkbox_group(-name=>'day_interval',
220			-values=>[ sort { $a <=> $b } keys %days ],
221			-labels=>\%days,
222			-defaults=>[
223			Time::Available::DAY_MONDAY,
224			Time::Available::DAY_TUESDAY,
225			Time::Available::DAY_WEDNESDAY,
226			Time::Available::DAY_THURSDAY,
227			Time::Available::DAY_FRIDAY,
228			]
229			),
230	dpavlin	1.1	$q->submit(-name=>'show',-value=>'Show report'),
231			)),end_table;
232
233			# dump report
234			#
235
236			sub sort_link {
237			my $q = shift \|\| return;
238			my $col = shift \|\| return;
239			my $dir = lc(shift) \|\| return;
240			if ($sort_param{$dir.'sort'} && $sort_param{$dir.'sort'} eq $col) {
241			return '&'.$dir.'Arr;';
242			} else {
243			return '<a href="'.$q->url(-query=>1).'&'.$dir.'sort='.$col.'">&'.$dir.'Arr;</a>';
244			}
245			}
246
247	dpavlin	1.2
248			my ($from_time,$to_time,$from_html,$to_html);
249			if ($q->param('use_date_limit')) {
250			$from_time = parsedate($q->param('from_date'), UK=>1);
251			$to_time = parsedate($q->param('to_date'), UK=>1);
252			$from_html = strftime($date_fmt,localtime($from_time));
253			$to_html = strftime($date_fmt,localtime($to_time));
254			$from_html .= " [$from_time] " if ($debug);
255			$to_html .= " [$to_time] " if ($debug);
256			}
257
258			# sort data
259			#
260			my @sorted = sorted_array(@data, @sort);
261			#my @sorted = @data;
262
263	dpavlin	1.3	print "-- sort: ",Dumper(@sort)," (data: ".@data." sorted: ".@sorted.") --\n",br,"-- dayMask: $dayMask --\n",br if ($debug);
264	dpavlin	1.2
265			print start_table({-border=>1,-cellspacing=>0,-cellpadding=>2,-width=>'100%'});
266
267			print Tr(
268	dpavlin	1.1	th("group/service"),
269			th({-bgcolor=>'#f0f0f0'},
270	dpavlin	1.3	&sort_link($q,'from','u').' from '.
271			&sort_link($q,'from','d'),
272	dpavlin	1.2	br,$from_html
273	dpavlin	1.1	),
274			th(
275	dpavlin	1.3	&sort_link($q,'to','u').' to '.
276			&sort_link($q,'to','d'),
277	dpavlin	1.2	br,$to_html
278	dpavlin	1.1	),
279			th({-bgcolor=>'#e0e0e0'},
280	dpavlin	1.3	&sort_link($q,'dur','u').' duration '.
281			&sort_link($q,'dur','d')
282	dpavlin	1.1	),
283			th("description")
284	dpavlin	1.2	) if (scalar @sorted > 0);
285	dpavlin	1.1
286			foreach my $row (@sorted) {
287	dpavlin	1.3	next if ($q->param('use_date_limit') && ($row->{from} < $from_time \|\| $row->{to} > $to_time));
288			my ($from,$dur,$int) = ('unknown','unknown','unknown');
289			if ($row->{from} != -1 ) {
290			$from = d($row->{from});
291			$dur = dur($row->{dur});
292			$int = dur($row->{int});
293			}
294	dpavlin	1.1	print Tr(
295			td({-align=>'left',-valign=>'center'},$row->{sg}),
296	dpavlin	1.3	td({-align=>'right',-bgcolor=>'#f0f0f0'},$from),
297			td({-align=>'right'},d($row->{to})),
298			td({-align=>'center',-bgcolor=>'#e0e0e0'},$dur),
299			td({-align=>'center',-bgcolor=>'#e0e0e0'},$int),
300	dpavlin	1.1	td({-align=>'left'},$row->{desc}),
301			),"\n";
302			}
303
304			# dump totals
305			#
306
307			foreach my $sg (keys %downtime) {
308	dpavlin	1.2	print Tr(td({-colspan=>3,-align=>'right'},"total for $sg:"),
309			td({-bgcolor=>'#e0e0e0',-align=>'right'},dur($downtime{$sg})),
310			td(small("in ".$sg_count{$sg}." failures"))),"\n";
311	dpavlin	1.1	}
312
313			print end_table,
314			end_form;
315