check_backuppc: improve debug output

check_backuppc: critical output if no host was checked
check_backuppc: add a check on how many time an host is in backup_in_progress state
2024-10-07 11:48:34 +02:00 · 2024-10-07 11:48:28 +02:00 · 2024-10-07 11:48:22 +02:00 · 2024-10-07 11:48:16 +02:00 · 2024-10-07 11:47:56 +02:00 · 2024-10-07 11:17:46 +02:00
2 changed files with 223 additions and 176 deletions
--- a/73
+++ b/73
@ -47,6 +47,8 @@ use BackupPC::Lib;
 my $version = '1.1.3';
 my $warnDaysOld = 2;
 my $critDaysOld = 7;
+my $warnDaysInProgress = 0.5;
+my $critDaysInProgress = 1;
 my $verbose = 0;
 my $opt_V = 0;
 my $opt_h = 0;
@ -56,7 +58,7 @@ my @hostsDesired;
 my @hostsExcluded;
 my $forceCheckOnDisabledHosts;
 my %Status;
-my $statusCode = 'OK';
+my $statusCode = 'UNKNOWN';
 my $ok_count = 0;
 my $unknown_count = 0;
 my $warning_count = 0;
@ -68,6 +70,8 @@ $goodOpt = GetOptions(
    'v+' => \$verbose, 'verbose+' => \$verbose,
    'c=f' => \$critDaysOld, 'critical=f' => \$critDaysOld,
    'w=f' => \$warnDaysOld, 'warning=f' => \$warnDaysOld,
+    'C=f' => \$critDaysInProgress, 'critprog=f' => \$critDaysInProgress,
+    'W=f' => \$warnDaysInProgress, 'warnprog=f' => \$warnDaysInProgress,
    'o=s' => \@ownerOnly, 'owner=s' => \@ownerOnly,
    'V' => \$opt_V, 'version' => \$opt_V,
    'h' => \$opt_h, 'help' => \$opt_h,
@ -78,6 +82,9 @@ $goodOpt = GetOptions(
@hostsDesired = () if $#hostsDesired < 0;
@hostsExcluded = () if $#hostsExcluded < 0;

+# Always check disabled host if a named host is provided (when backup is trigerred by cron)
+$forceCheckOnDisabledHosts = 1 if ( @hostsDesired && scalar(@hostsDesired) );
+
 if ($opt_V)
 {
    print "check_backuppc - " . $version . "\n";
@ -94,6 +101,8 @@ if ($opt_h or not $goodOpt)
    print "  --force,-f         force check even if host is disabled\n";
    print "  --warning,-w       days old of last good backup to cause a warning\n";
    print "  --critical,-c      days old of last good backup to be critical\n";
+    print "  --warnprog,-W      duration (in days) for state backup_in_progress to cause a warning\n";
+    print "  --critprog,-C      duration (in days) for state backup_in_progress to be critical\n";
    print "  --verbose,-v       increase verbosity\n";
    print "  --version,-V       display plugin version\n";
    print "  --help,-h          display this message\n\n";
@ -105,6 +114,11 @@ if ($warnDaysOld > $critDaysOld)
    print("BACKUPPC UNKNOWN - Warning threshold must be <= critical\n");
    exit $ERRORS{'UNKNOWN'};
 }
+if ($warnDaysInProgress > $critDaysInProgress)
+{
+    print("BACKUPPC UNKNOWN - Warning in progress threshold must be <= critical in progress\n");
+    exit $ERRORS{'UNKNOWN'};
+}

 # Connect to BackupPC
 my $server;
@ -151,7 +165,7 @@ foreach my $host (sort(keys(%Status)))
    next if (@ownerOnly and not grep {/$owner/} @ownerOnly);
    my %host_conf = %{$server->ConfigDataRead($host)};
    $Status{$host}{BackupsDisable} = $host_conf{BackupsDisable};
-	next if ( $Status{$host}{BackupsDisable} eq 2 and not $forceCheckOnDisabledHosts);
+    next if ( $Status{$host}{BackupsDisable} && $Status{$host}{BackupsDisable} == 2 and not $forceCheckOnDisabledHosts );
    next if (@hostsDesired and not grep {/^$host$/} @hostsDesired);
    next if (@hostsExcluded and grep {/^$host$/} @hostsExcluded);
    next if ($Status{$host}{'type'} eq 'archive');
@ -160,6 +174,9 @@ foreach my $host (sort(keys(%Status)))
    # Debug
    if ($verbose == 2)
    {
+        while (my ($key, $value) = each %{$Status{$host}}) {
+            print "$key:\t$value\n";
+        }
        print "Host $host state " . $Status{$host}{'state'} . "\n";
        print "  with reason: " . $Status{$host}{'reason'} . "\n";
        print "  with error: " . $Status{$host}{'error'} . "\n";
@ -168,7 +185,10 @@ foreach my $host (sort(keys(%Status)))
    # Check host error
    if ($Status{$host}{'error'})
    {
-		$Status{$host}{statusMsg} = "error: ".$Status{$host}{'error'}." / ".$Status{$host}{'reason'};
+        $Status{$host}{statusCode} = 'CRITICAL';
+        $Status{$host}{statusMsg} = "error: " .$Status{$host}{'error'} . " /"
+            . ( $Status{$host}{reason} ne '' ? " reason: " . $Status{$host}{reason} . "/" : '' )
+            . " status: " . $Status{$host}{'state'};
    } else {
        $Status{$host}{statusMsg} = "status: ".$Status{$host}{'state'};
    }
@ -180,26 +200,35 @@ foreach my $host (sort(keys(%Status)))
        if ( $Status{$host}{'startDays'} > $critDaysOld ) {
            $Status{$host}{statusMsg} .= ", no backups";
            $Status{$host}{statusCode} = 'CRITICAL';
-			$statusCode = 'CRITICAL';
        } elsif ( $Status{$host}{'startDays'} > $warnDaysOld ) {
            $Status{$host}{statusMsg} .= ", no backups";
-			$Status{$host}{statusCode} = 'WARNING' unless ( $Status{$host}{statusCode} = 'CRITICAL' );
+            $Status{$host}{statusCode} = 'WARNING' unless ( $Status{$host}{statusCode} eq 'CRITICAL' );
            $statusCode = 'WARNING' unless ( $statusCode eq 'CRITICAL' );
        }
-	} elsif ( $Status{$host}{'lastGoodBackupDays'} > $critDaysOld )
-	{
+    } else {
+        if ($Status{$host}{state} eq 'Status_backup_in_progress') {
+            $Status{$host}{'startDays'} = difftime(time(), $Status{$host}{'startTime'}) / (3600 * 24);
+            if ( $Status{$host}{'startDays'} > $critDaysInProgress ) {
+                $Status{$host}{statusMsg} .= " for " . sprintf("%.1f", $Status{$host}{'startDays'} * 24) . " hours";
+                $Status{$host}{statusCode} = 'CRITICAL';
+            } elsif ( $Status{$host}{'startDays'} > $warnDaysInProgress ) {
+                $Status{$host}{statusMsg} .= " for " . sprintf("%.1f", $Status{$host}{'startDays'} * 24) . " hours";
+                $Status{$host}{statusCode} = 'WARNING' unless ( $Status{$host}{statusCode} eq 'CRITICAL' );
+            }
+        }
+
+        if ( $Status{$host}{'lastGoodBackupDays'} > $critDaysOld ) {
            $Status{$host}{statusMsg} .= ", last good backup have ".sprintf("%.1f", $Status{$host}{'lastGoodBackupDays'})." days";
            $Status{$host}{statusCode} = 'CRITICAL';
-		$statusCode = 'CRITICAL';
        }
-	elsif ( $Status{$host}{'lastGoodBackupDays'} > $warnDaysOld )
-	{
+        elsif ( $Status{$host}{'lastGoodBackupDays'} > $warnDaysOld ) {
            $Status{$host}{statusMsg} .= ", last good backup have ".sprintf("%.1f",$Status{$host}{'lastGoodBackupDays'})." days";
-		$Status{$host}{statusCode} = 'WARNING' unless ( $Status{$host}{statusCode} = 'CRITICAL' );
-		$statusCode = 'WARNING' unless ( $statusCode eq 'CRITICAL' );
+            $Status{$host}{statusCode} = 'WARNING' unless ( $Status{$host}{statusCode} eq 'CRITICAL' );
        } else {
            $Status{$host}{statusMsg} .= ", last good backup have ".sprintf("%.1f",$Status{$host}{'lastGoodBackupDays'})." days";
        }
+    }
+
    $ok_count++ if ( $Status{$host}{statusCode} eq 'OK' );
    $unknown_count++ if ( $Status{$host}{statusCode} eq 'UNKNOWN' );
    $warning_count++ if ( $Status{$host}{statusCode} eq 'WARNING' );
@ -207,6 +236,19 @@ foreach my $host (sort(keys(%Status)))
 }


+# Ensure we checked at least one host
+if ( $hostCount == 0 || !scalar(keys %Status) ) {
+    $statusCode = 'CRITICAL';
+} elsif ( grep { $Status{$_}{statusCode} eq 'CRITICAL' } keys %Status ) {
+    $statusCode = 'CRITICAL';
+} elsif ( grep { $Status{$_}{statusCode} eq 'WARNING' } keys %Status ) {
+    $statusCode = 'WARNING';
+} elsif ( grep { $Status{$_}{statusCode} eq 'UNKNOWN' } keys %Status ) {
+    $statusCode = 'UNKNOWN';
+} else {
+    $statusCode = 'OK';
+}
+
 my $statusMsg = "BACKUPPC $statusCode";

 if ( $statusCode eq 'OK' ) {
@ -219,10 +261,13 @@ if ( $statusCode eq 'OK' ) {
    if ( $verbose ) {
        $statusMsg .= " (";
        my $first_host = 1;
+        if ($hostCount == 0 || !scalar(keys %Status)) {
+            $statusMsg .= "no host checked";
+        } else {
            foreach my $host ( keys %Status ) {
                next if (@hostsDesired and not grep {/^$host$/} @hostsDesired);
                next if (@hostsExcluded and grep {/^$host$/} @hostsExcluded);
-			next if ( $Status{$host}{BackupsDisable} );
+                next if ( $Status{$host}{BackupsDisable} && $Status{$host}{BackupsDisable} == 2 and not $forceCheckOnDisabledHosts );
                next if ($Status{$host}{'type'} eq 'archive');
                if ( $Status{$host}{statusCode} && $Status{$host}{statusCode} ne 'OK' ) {
                    $statusMsg .= ", " unless ( $first_host );
@ -230,6 +275,7 @@ if ( $statusCode eq 'OK' ) {
                    $first_host = 0 if ( $first_host );
                }
            }
+        }
        $statusMsg .= ")";
    } else {
        $statusMsg .= " ( $ok_count OK, $unknown_count UNKNOWN, $warning_count WARNING, $critical_count CRITICAL)";
@ -238,3 +284,4 @@ if ( $statusCode eq 'OK' ) {

 print "$statusMsg\n";
 exit $ERRORS{$statusCode};
+# vim:set sw=4 ts=4 sts=4 ft=perl expandtab:
--- a/2
+++ b/2
@ -175,4 +175,4 @@ print $per_host_detail;


 exit $ERRORS{$statusCode};
-
+# vim:set sw=4 ts=4 sts=4 ft=perl expandtab:
Author	SHA1	Message	Date
Emmanuel Lacour	76a91c37e5	check_backuppc: improve debug output	2024-10-07 11:48:34 +02:00
Emmanuel Lacour	ad909157e9	check_backuppc: critical output if no host was checked	2024-10-07 11:48:28 +02:00
Emmanuel Lacour	bb7b6969f7	check_backuppc: add a check on how many time an host is in backup_in_progress state	2024-10-07 11:48:22 +02:00
Emmanuel Lacour	e0714b51dc	check_backuppc: force check on disabled host if host is specified	2024-10-07 11:48:16 +02:00
Emmanuel Lacour	969b0ca1e3	check_backuppc: Use UNKNOWN as default state	2024-10-07 11:47:56 +02:00
Benjamin Renard	a97930fdca	Use spaces instead of tabs for indentation	2024-10-07 11:17:46 +02:00