fixed compile problem with Time::HiRes no being installed in server.

sleep time can be individually set for each program via $SPONGSLEEP{} hash.
changed escalation code to be a subloop that rechecks all services reported
as down after main check loop

diff --git a/src/spong-network.pl b/src/spong-network.pl
index 819e98c3ead44f94c592104ef2c3e24f7ce881fa..295a2f5838adf02080d02b9564ed5221021ae90a 100755 (executable)
--- a/src/spong-network.pl
+++ b/src/spong-network.pl
@@ -1,4 +1,4 @@
-#!@@PERL@@
+#!/usr/bin/perl
 #
 # Spong network monitoring script.  This runs various tests against IP based
 # network services (like nntp, smtp, http, pop, etc...)  If it can not
@@ -33,7 +33,7 @@ use Spong::Status qw(status);
 use Spong::Log;
 
 # Check to see if the Time::HiRes module is available
-eval "use Time::HiRes qw( time );";
+eval { require Time::HiRes; import Time::HiRes qw(time); };
 if (! $@ ) { $hires = 1; } else { $hires = 0; };
 
 srand( time() ^ ($$ + ($$ << 15 )) );
@@ -71,32 +71,60 @@ $SPONGSLEEP = $SPONGSLEEP{'spong-network'} || $SPONGSLEEP{'DEFAULT'} ||
 %PLUGINS = {};
 &config_funcs();
 
+my @bad_checks;
+
 # Do the various network tests for each host.
 
+my $lastcheck = time();
+
 while( 1 ) {
+   @bad_checks = ();  # Clear out the bad check list
+
+   # Main checking loop, check everything.
    foreach $host ( @HOSTS_LIST ) {
       &debug( "checking network services on $host" );
 
-      my $check;
-      foreach $check ( split(/\s+/,$HOSTS{$host}->{'services'}) ) {
+      foreach my $check ( split(/\s+/,$HOSTS{$host}->{'services'}) ) {
            $0 = "spong-network (checking $host/$check)";
            &do_check($host,$check);
       }
-  }
+   }
+
 
-   # If we are suppose to stay alive, then sleep about $SPONGSLEEP seconds, we
-   # add a little randomness so that things don't get in sync and pound the 
-   # spong-server.  Otherwise, just exit.
+   # Now recheck the services that were down, but need to be rechecked
+   while ( @bad_checks ) {
+      my @tmp_checks = @bad_checks;
+      @bad_checks = ();
+      foreach $tmp (@tmp_checks) {
+         my ( $host, $check ) = split / /,$tmp;
+         &do_check($host,$check);
+      }
+
+      sleep $RECHECKSLEEP;
+   }
+ 
+   # If we are suppose to stay alive, then calculate the time for the next
+   # main time loop which is $lastcheck + $SPONGSLEEP, add a little randomness
+   # so that things don't get in sync and pound the spong-server. If we need
+   # to sleep off the difference.  Otherwise, just exit.
 
    if( $nosleep ) {
       last;
    } else {
-      my $sleep = int($SPONGSLEEP - (.05 * $SPONGSLEEP) + 
-                 rand(.1 * $SPONGSLEEP));
-      &debug( "sleeping for $sleep seconds" );
-      $0 = "spong-network (sleeping)";
-      sleep $sleep;
+
+      # Calculate the time for the next loop
+      my $nexttime = $lastcheck + int($SPONGSLEEP - (.05 * $SPONGSLEEP) + 
+                       rand(.1 * $SPONGSLEEP));
+   
+      if ( $nexttime > time() ) {
+         my $sleep = $nexttime - time();
+         &debug( "sleeping for $sleep seconds" );
+         $0 = "spong-network (sleeping)";
+         sleep $sleep;
+      }
+      $lastcheck = $nexttime;  # Save the curent check time as the last time
    }
+
 }
 
 unlink( "$SPONGTMP/spong-network.pid" ) unless $nosleep;
@@ -136,7 +164,8 @@ sub do_check {
    if ($status eq 'red') {
       $crit_count += 1;
       $color = ($crit_count < $CRIT_WARN_LEVEL) ? 'yellow' : 'red';
-      $summary = "($crit_count/$CRIT_WARN_LEVEL) " . $summary
+      $summary = "($crit_count/$CRIT_WARN_LEVEL) " . $summary;
+      push @bad_checks, "$host $service" if ($crit_count < $CRIT_WARN_LEVEL);
    } else {
       $crit_count = 0;
       $color = $status;
@@ -145,11 +174,7 @@ sub do_check {
    &debug("$status - $crit_count - $CRIT_WARN_LEVEL - $color - $summary");
 
    # Save the critical counter in the host for the service
-   if ($color ne "green") {
-      $HOSTS{$host}->{'service'}->{$service}->{'count'} = $crit_count;
-   } else {
-      undef $HOSTS{$host}->{'service'}->{$service}->{'count'};
-   }
+   $HOSTS{$host}->{'service'}->{$service}->{'count'} = $crit_count;
 
    eval {
       &status( $SPONGSERVER, $host, $service, $color, $summary, $message );
@@ -170,9 +195,9 @@ sub check_simple {
    my( $attempt, $start, $message, $diff, $errcd );
 
    for $timeout ( 3, 5, 12 ) {
-      $start = main::time();
+      $start = time();
       ($errcd,$message) = &check_tcp( $host, $port, $send, $timeout );
-      $diff    = main::time() - $start;
+      $diff    = time() - $start;
 
       $attempt++;
       if( $message =~ /$check/ ) { $color = "green"; last; }