implemented availability calculation with Monitoring::Availability

Makefile.PL

@@ -31,7 +31,7 @@ requires 'Moose';
 requires 'MooseX::MethodAttributes::Inheritable';
 requires 'MooseX::Role::WithOverloading';
 requires 'Monitoring::Livestatus'                 => '0.34';
-requires 'Monitoring::Availability'               => '0.04';
+requires 'Monitoring::Availability'               => '0.06';
 requires 'HTTP::Body';
 requires 'LWP';
 requires 'Config::Any';

docs/development/TODO

@@ -20,8 +20,8 @@ config:
     - add missing values
 
 avail:
-    - missing completly
-    - remove duplicates from timeperiods/groups list
+    - CSV output
+    - trends graph
 
 trends:
     - missing completly
@@ -31,14 +31,23 @@ histogram:
 
 livestatus:
     - StatsSum: ( needed by tac and extinfo 4 )
-    - Contacts: Service Notification Options, Host Notification Options, Retention Options, Service Notification Commands, Host Notification Commands
-    - Hosts: Freshness Threshold, Contact Groups, Notification Options, Event Handler, Stalking Options, Flap Detection Options, Enable Failure Prediction, Retention Options
-    - Services: Parallelize, Volatile, Check Freshness, Freshness Threshold, Contact Groups, Notification Options, Stalking Options, Flap Detection Options, Enable Failure Prediction, Failure Prediction Options, Retention Options
+    - Contacts: Service Notification Options, Host Notification Options,
+                Retention Options, Service Notification Commands,
+                Host Notification Commands
+    - Hosts:    Freshness Threshold, Contact Groups, Notification Options,
+                Event Handler, Stalking Options, Flap Detection Options,
+                Enable Failure Prediction, Retention Options
+    - Services: Parallelize, Volatile, Check Freshness, Freshness Threshold,
+                Contact Groups, Notification Options, Stalking Options,
+                Flap Detection Options, Enable Failure Prediction,
+                Failure Prediction Options, Retention Options
+    - Timeperiods:
+                Monday - Sunday, Exceptions
     - Tables:
-        - Host Dependencies
-        - Host Escalations
-        - Service Dependencies
-        - Service Escalations
+                Host Dependencies
+                Host Escalations
+                Service Dependencies
+                Service Escalations
 
 =========================================================
 after 1.0

lib/Thruk/Controller/avail.pm

@@ -3,6 +3,7 @@ package Thruk::Controller::avail;
 use strict;
 use warnings;
 use Data::Dumper;
+use Monitoring::Availability;
 use parent 'Catalyst::Controller';
 
 =head1 NAME
@@ -35,7 +36,7 @@ sub index :Path :Args(0) :MyAction('AddDefaults') {
     # lookup parameters
     my $report_type    = $c->{'request'}->{'parameters'}->{'report_type'}  || '';
     my $get_date_parts = $c->{'request'}->{'parameters'}->{'get_date_parts'};
-    my $timeperiod     = $c->{'request'}->{'parameters'}->{'timeperiod'}   || '';
+    my $timeperiod     = $c->{'request'}->{'parameters'}->{'timeperiod'};
     my $host           = $c->{'request'}->{'parameters'}->{'host'}         || '';
     my $hostgroup      = $c->{'request'}->{'parameters'}->{'hostgroup'}    || '';
     my $service        = $c->{'request'}->{'parameters'}->{'service'}      || '';
@@ -197,6 +198,7 @@ sub _create_report {
         $c->stash->{template}   = 'avail_report_hosts.tt';
     }
     else {
+        $c->log->debug("unknown report type");
         return;
     }
 
@@ -217,6 +219,7 @@ sub _create_report {
     my $t1           = $c->{'request'}->{'parameters'}->{'t1'};
     my $t2           = $c->{'request'}->{'parameters'}->{'t2'};
 
+    $timeperiod = 'last24hours' if(!defined $timeperiod and !defined $t1 and !defined $t2);
     my($start,$end) = Thruk::Utils::get_start_end_for_timeperiod($timeperiod,$smon,$sday,$syear,$shour,$smin,$ssec,$emon,$eday,$eyear,$ehour,$emin,$esec,$t1,$t2);
 
     $c->log->debug("start: ".$start." - ".(scalar localtime($start)));
@@ -248,6 +251,24 @@ sub _create_report {
     $backtrack = 4 unless defined $backtrack;
     $backtrack = 4 if $backtrack < 0;
 
+    $assumeinitialstates          = 'no' unless $assumeinitialstates          eq 'yes';
+    $assumestateretention         = 'no' unless $assumestateretention         eq 'yes';
+    $assumestatesduringnotrunning = 'no' unless $assumestatesduringnotrunning eq 'yes';
+    $includesoftstates            = 'no' unless $includesoftstates            eq 'yes';
+
+    $initialassumedhoststate      = 0 unless $initialassumedhoststate ==  0  # Unspecified
+                                          or $initialassumedhoststate == -1  # Current State
+                                          or $initialassumedhoststate ==  3  # Host Up
+                                          or $initialassumedhoststate ==  4  # Host Down
+                                          or $initialassumedhoststate ==  5; # Host Unreachable
+
+    $initialassumedservicestate   = 0 unless $initialassumedservicestate ==  0  # Unspecified
+                                          or $initialassumedservicestate == -1  # Current State
+                                          or $initialassumedservicestate ==  6  # Service Ok
+                                          or $initialassumedservicestate ==  8  # Service Warning
+                                          or $initialassumedservicestate ==  7  # Service Unknown
+                                          or $initialassumedservicestate ==  9; # Service Critical
+
     $c->stash->{rpttimeperiod}                = $rpttimeperiod;
     $c->stash->{assumeinitialstates}          = $assumeinitialstates;
     $c->stash->{assumestateretention}         = $assumestateretention;
@@ -458,31 +479,27 @@ sub _create_report {
     $logs = $c->{'live'}->selectall_arrayref($log_query, { Slice => 1} );
     $c->stats->profile(end   => "avail.pm fetchlogs");
 
-    $logs = Thruk::Utils::sort($c, $logs, 'time', 'ASC');
-    $c->stash->{'logs'} = $logs;
-
     #$Data::Dumper::Indent = 1;
     #open(FH, '>', '/tmp/logs.txt') or die("cannot open logs.txt: $!");
-    #print FH Dumper($logs);
+    ##print FH Dumper($logs);
     #for my $line (@{$logs}) {
     #    print FH '['.$line->{'time'}.'] '.$line->{'type'};
     #    print FH ': '.$line->{'options'} if(defined $line->{'options'} and $line->{'options'} ne '');
     #    print FH "\n";
     #}
     #close(FH);
-    use Monitoring::Availability;
     $c->stats->profile(begin => "calculate availability");
     my $ma = Monitoring::Availability->new(
         'rpttimeperiod'                => $rpttimeperiod,
         'assumeinitialstates'          => $assumeinitialstates,
         'assumestateretention'         => $assumestateretention,
         'assumestatesduringnotrunning' => $assumestatesduringnotrunning,
         'includesoftstates'            => $includesoftstates,
-        'initialassumedhoststate'      => $initialassumedhoststate,
-        'initialassumedservicestate'   => $initialassumedservicestate,
+        'initialassumedhoststate'      => $self->_initialassumedhoststate_to_state($initialassumedhoststate),
+        'initialassumedservicestate'   => $self->_initialassumedservicestate_to_state($initialassumedservicestate),
         'backtrack'                    => $backtrack,
-        #'verbose'                      => 1,
-        #'logger'                       => $c->log,
+#        'verbose'                      => 1,
+#        'logger'                       => $c->log,
     );
     $c->stash->{avail_data} = $ma->calculate(
         'start'                        => $start,
@@ -494,13 +511,49 @@ sub _create_report {
     #$c->log->info(Dumper($c->stash->{avail_data}));
     $c->stats->profile(end => "calculate availability");
 
+    if($full_log_entries) {
+        $c->stash->{'logs'} = $ma->get_full_logs();
+        #$c->log->debug("got full logs: ".Dumper($c->stash->{'logs'}));
+    }
+    elsif($show_log_entries) {
+        $c->stash->{'logs'} = $ma->get_condensed_logs();
+        #$c->log->debug("got condensed logs: ".Dumper($c->stash->{'logs'}));
+    }
+
+    $c->stats->profile(end => "got logs");
+
     # finished
     $c->stash->{time_token} = time() - $start_time;
     $c->stats->profile(end => "_create_report()");
 
     return 1;
 }
 
+##########################################################
+sub _initialassumedhoststate_to_state {
+    my $self                    = shift;
+    my $initialassumedhoststate = shift;
+
+    return 'unspecified' if $initialassumedhoststate ==  0; # Unspecified
+    return 'current'     if $initialassumedhoststate == -1; # Current State
+    return 'up'          if $initialassumedhoststate ==  3; # Host Up
+    return 'down'        if $initialassumedhoststate ==  4; # Host Down
+    return 'unreachable' if $initialassumedhoststate ==  5; # Host Unreachable
+}
+
+##########################################################
+sub _initialassumedservicestate_to_state {
+    my $self                       = shift;
+    my $initialassumedservicestate = shift;
+
+    return 'unspecified' if $initialassumedservicestate ==  0; # Unspecified
+    return 'current'     if $initialassumedservicestate == -1; # Current State
+    return 'ok'          if $initialassumedservicestate ==  6; # Service Ok
+    return 'warning'     if $initialassumedservicestate ==  8; # Service Warning
+    return 'unknown'     if $initialassumedservicestate ==  7; # Service Unknown
+    return 'critical'    if $initialassumedservicestate ==  9; # Service Critical
+}
+
 =head1 AUTHOR
 
 Sven Nierlein, 2009, <nierlein@cpan.org>

lib/Thruk/Utils.pm

@@ -769,7 +769,7 @@ sub get_start_end_for_timeperiod {
         $start = Mktime($year-1,1,1,  0,0,0);
         $end   = Mktime($year,1,1,  0,0,0);
     }
-    elsif($timeperiod eq 'custom') {
+    elsif($timeperiod eq 'custom' or $timeperiod eq 'current') {
         $start = $t1;
         $end   = $t2;
         if(!defined $start) {

templates/avail_report_form_box.tt

@@ -5,12 +5,13 @@
 [% IF servicegroup %]<input type='hidden' name='servicegroup' value='[% servicegroup %]'>[% END %]
 <input type='hidden' name='t1' value='[% start %]'>
 <input type='hidden' name='t2' value='[% end %]'>
-<input type='hidden' name='show_log_entries' value='[% show_log_entries %]'>
+[% IF show_log_entries %]<input type='hidden' name='show_log_entries' value='[% show_log_entries %]'>[% END %]
+[% IF full_log_entries %]<input type='hidden' name='full_log_entries' value='[% full_log_entries %]'>[% END %]
 <input type='hidden' name='assumeinitialstates' value='[% assumeinitialstates %]'>
 <input type='hidden' name='assumestateretention' value='[% assumestateretention %]'>
 <input type='hidden' name='assumestatesduringnotrunning' value='[% assumestatesduringnotrunning %]'>
 <input type='hidden' name='includesoftstates' value='[% includesoftstates %]'>
-<input type='hidden' name='report_type' value='[% report_type %]'>
+[% IF report_type %]<input type='hidden' name='report_type' value='[% report_type %]'>[% END %]
 <table border="0" class='optBox'>
   <tr>
     [% UNLESS service %]

templates/avail_report_hostgroup.tt

@@ -33,9 +33,9 @@
     <br>
     <br>
 
-    [% FOR groupname IN groups.keys.sort  %]
-        [% SET hostgroupname = groupname      %]
-        [% SET hosts         = g.hosts        %]
+    [% FOR groupname IN groups.keys.sort %]
+        [% SET hostgroupname = groupname %]
+        [% SET hosts         = groups.$groupname.hosts %]
         [% PROCESS avail_report_hosts_part.tt %]
     [% END %]
 

templates/avail_report_log_entries.tt

@@ -3,7 +3,7 @@
 [% IF show_log_entries or full_log_entries %]
     <div align="center" class='dataTitle'>[% IF service %]Service[% ELSE %]Host[% END %] Log Entries:</div><br>
     <div align="center" class='infoMessage'>
-      <a href='avail.cgi?t1=[% start %]&amp;t2=[% end %][% IF host %]&amp;host=[% host %][% END %][% IF service %]&amp;service=[% service %][% END %]&amp;backtrack=[% backtrack %]&amp;assumestateretention=[% assumestateretention %]&amp;assumeinitialstates=[% assumeinitialstates %]&amp;assumestatesduringnotrunning=[% assumestatesduringnotrunning %]&amp;initialassumedhoststate=[% initialassumedhoststate %]&amp;initialassumedservicestate=[% initialassumedservicestate %]&amp;[% IF show_log_entries %]full_log_entries[% END %]&amp;[% IF full_log_entries %]show_log_entries[% END %]&amp;showscheduleddowntime=[% showscheduleddowntime %]'>[ View [% IF full_log_entries %]condensed[% ELSE %]full[% END %] log entries ]</a>
+      <a href='avail.cgi?t1=[% start %]&amp;t2=[% end %][% IF host %]&amp;host=[% host %][% END %][% IF service %]&amp;service=[% service %][% END %]&amp;backtrack=[% backtrack %]&amp;assumestateretention=[% assumestateretention %]&amp;assumeinitialstates=[% assumeinitialstates %]&amp;assumestatesduringnotrunning=[% assumestatesduringnotrunning %]&amp;initialassumedhoststate=[% initialassumedhoststate %]&amp;initialassumedservicestate=[% initialassumedservicestate %]&amp;[% IF full_log_entries %]show_log_entries[% ELSE %]full_log_entries[% END %]&amp;showscheduleddowntime=[% showscheduleddowntime %]'>[ View [% IF full_log_entries %]condensed[% ELSE %]full[% END %] log entries ]</a>
     </div>
     <div align="center">
       <table border="1" cellspacing="0" cellpadding="3" class='logEntries'>
@@ -15,67 +15,14 @@
           <th class='logEntries'>Event/State Information</th>
         </tr>
         [% FOR log IN logs %]
-
-
-        [%# filter some logs out #%]
-        [%# logentry outside our timedefinition #%]
-        [% IF log.time > end or log.time < start %][% NEXT %][% END %]
-
-        [% IF host and !service %]
-          [%# filter out anything besides the host on hostmode #%]
-          [% IF log.class != 2 and log.host_name != host         %][% NEXT %][% END %]
-          [% IF log.class != 2 and log.service_description != '' %][% NEXT %][% END %]
-        [% END %]
-
-
         [% SET class = 'logEntriesEven' IF loop.odd  %]
         [% SET class = 'logEntriesOdd'  IF loop.even %]
-        [% SET message       = log.type %]
-        [% SET plugin_output = log.plugin_output %]
-        [% IF log.type == 'CURRENT SERVICE STATE' or log.type == 'INITIAL SERVICE STATE' or log.type == 'SERVICE ALERT' %]
-          [% IF log.state == 0 %][% SET mclass = "OK"       message = 'SERVICE OK'       %]
-          [% ELSIF log.state == 1 %][% SET mclass = "WARNING"  message = 'SERVICE WARNING'  %]
-          [% ELSIF log.state == 2 %][% SET mclass = "UNKNOWN"  message = 'SERVICE UNKNOWN'  %]
-          [% ELSIF log.state == 3 %][% SET mclass = "CRITICAL" message = 'SERVICE CRITICAL' %]
-          [% END %]
-
-          [% IF log.options.search(';HARD;') %][% SET message = message _ ' (HARD)' %][% ELSE %][% SET message = message _ ' (SOFT)' %][% END %]
-        [% ELSIF log.type == 'CURRENT HOST STATE' or log.type == 'INITIAL HOST STATE' or log.type == 'HOST ALERT' %]
-          [% IF log.state == 0 %][% SET mclass = "UP"          message = 'HOST UP'          %]
-          [% ELSIF log.state == 1 %][% SET mclass = "DOWN"        message = 'HOST DOWN'        %]
-          [% ELSIF log.state == 2 %][% SET mclass = "UNREACHABLE" message = 'HOST UNREACHABLE' %]
-          [% END %]
-
-          [% IF log.options.search(';HARD;') %][% SET message = message _ ' (HARD)' %][% ELSE %][% SET message = message _ ' (SOFT)' %][% END %]
-        [% ELSIF log.type == 'HOST DOWNTIME ALERT' %]
-          [% IF log.options.search(';STARTED;') %][% SET mclass = "INDETERMINATE" message = 'HOST DOWNTIME START'     plugin_output = 'Start of scheduled downtime' %][% END %]
-          [% IF log.options.search(';STOPPED;') %][% SET mclass = "INDETERMINATE" message = 'HOST DOWNTIME STOP'      plugin_output = 'End of scheduled downtime'   %][% END %]
-        [% ELSIF log.type == 'SERVICE DOWNTIME ALERT' %]
-          [% IF log.options.search(';STARTED;') %][% SET mclass = "INDETERMINATE" message = 'SERVICE DOWNTIME START'  plugin_output = 'Start of scheduled downtime' %][% END %]
-          [% IF log.options.search(';STOPPED;') %][% SET mclass = "INDETERMINATE" message = 'SERVICE DOWNTIME STOP'   plugin_output = 'End of scheduled downtime'   %][% END %]
-        [% ELSIF log.type.search(' starting\.\.\.') or log.type.search(' restarting\.\.\.') %]
-                                                  [% SET mclass = "INDETERMINATE" message = 'PROGRAM (RE)START'       plugin_output = 'Program start'               %]
-        [% ELSIF log.type.search(' shutting down\.\.\.') %]
-                                                  [% SET mclass = "INDETERMINATE" message = 'PROGRAM END'             plugin_output = 'Normal program termination'  %]
-        [% ELSIF log.type.search('Bailing out') %]
-                                                  [% SET mclass = "INDETERMINATE" message = 'PROGRAM END'             plugin_output = 'Abnormal program termination'  %]
-        [% ELSE %]
-          [%# SET message = 'SKIPPED!! ' _ message #%]
-          [%# SET plugin_output = 'SKIPPED!! ' _ plugin_output #%]
-          [% NEXT %]
-        [% END %]
-
-        [%# no process restarts in condensed mode #%]
-        [% IF show_log_entries and log.class == 2 %]
-          [% NEXT %]
-        [% END %]
-
         <tr class='[% class %]'>
-          <td class='[% class %]'>[% date.format(log.time, '%Y-%m-%d  %H:%M:%S') %]</td>
-          <td class='[% class %]'>2010-01-09 14:11:33</td>
-          <td class='[% class %]'>0d 14h 11m 33s</td>
-          <td class='logEntries[% mclass %]'>[% message %]</td>
-          <td class='[% class %]'>[% plugin_output %]</td>
+          <td class='[% class %]'>[% date.format(log.start, '%Y-%m-%d  %H:%M:%S') %]</td>
+          <td class='[% class %]'>[% date.format(log.end, '%Y-%m-%d  %H:%M:%S') %]</td>
+          <td class='[% class %]'>[% log.duration %]</td>
+          <td class='logEntries[% log.class %]'>[% log.type %]</td>
+          <td class='[% class %]'>[% log.plugin_output %]</td>
         </tr>
         [% END %]
       </table>

templates/avail_report_services_part.tt

@@ -23,7 +23,7 @@
           <th class='data'>Host</th>
           [% END %]
           <th class='data'>Service</th>
-          <th class='data'>% Time O </th>
+          <th class='data'>% Time Ok</th>
           <th class='data'>% Time Warning</th>
           <th class='data'>% Time Unknown</th>
           <th class='data'>% Time Critical</th>

templates/extinfo_type_1.tt

@@ -105,7 +105,7 @@
                               [% ELSIF host.state == 2 %]
                               <div class='hostUNREACHABLE'>&nbsp;&nbsp;UNREACHABLE&nbsp;&nbsp;</div>
                               [% END %]
-                              &nbsp;(for [% duration(0 - host.last_state_change) %])
+                              &nbsp;(for [% duration(date.now - host.last_state_change) %])
                               [% IF host.acknowledged %]
                               &nbsp;(Has been acknowledged)
                               [% END %]

templates/extinfo_type_2.tt

@@ -96,7 +96,7 @@
                               [% ELSIF service.state == 3 %]
                               <div class='serviceUNKNOWN'>&nbsp;&nbsp;UNKNOWN&nbsp;&nbsp;</div>
                               [% END %]
-                              &nbsp;(for [% duration(0 - service.last_state_change) %])
+                              &nbsp;(for [% duration(date.now - service.last_state_change) %])
                               [% IF service.acknowledged %]
                               &nbsp;(Has been acknowledged)
                               [% END %]

templates/outages.tt

@@ -1,3 +1,4 @@
+[% USE date %]
 [% PROCESS _header.tt %]
 
     <table border="0" width="100%">
@@ -39,7 +40,7 @@
             N/A
             [% END %]
           </td>
-          <td class='[% class %]'>[% duration(0 - host.last_state_change) %]</td>
+          <td class='[% class %]'>[% duration(date.now - host.last_state_change) %]</td>
           <td class='[% class %]'>[% host.affected_hosts %]</td>
           <td class='[% class %]'>[% host.affected_services %]</td>
           <td class='[% class %]'>