Some improvements
This commit is contained in:
Benjamin Renard
parent
6ed85bafad
commit
9e507bc48b
1 changed files with 76 additions and 49 deletions
125
check_backuppc
125
check_backuppc
|
|
@ -54,6 +54,13 @@ my $goodOpt = 0;
|
|||
my @ownerOnly;
|
||||
my @hostsDesired;
|
||||
my @hostsExcluded;
|
||||
my %Status;
|
||||
my $statusCode = 'OK';
|
||||
my $ok_count = 0;
|
||||
my $unknown_count = 0;
|
||||
my $warning_count = 0;
|
||||
my $critical_count = 0;
|
||||
|
||||
|
||||
# Process options
|
||||
$goodOpt = GetOptions(
|
||||
|
|
@ -114,8 +121,6 @@ if ($err)
|
|||
exit $ERRORS{'UNKNOWN'};
|
||||
}
|
||||
|
||||
my %Status;
|
||||
|
||||
# query the BackupPC server for host status
|
||||
my $status_raw = $server->ServerMesg('status hosts');
|
||||
my $hosts_infos = $server->HostInfoRead();
|
||||
|
|
@ -125,19 +130,16 @@ eval $status_raw;
|
|||
|
||||
# check the dumped output
|
||||
my $hostCount = 0;
|
||||
my $errorLevel='OK';
|
||||
|
||||
foreach my $host (@hostsDesired, @hostsExcluded)
|
||||
{
|
||||
if (not grep {/$host/} keys(%Status))
|
||||
if (not grep {/^$host$/} keys(%Status))
|
||||
{
|
||||
print("BACKUPPC UNKNOWN - Unknown host ($host)\n");
|
||||
exit $ERRORS{'UNKNOWN'};
|
||||
}
|
||||
}
|
||||
|
||||
my @problems;
|
||||
|
||||
# host status checks
|
||||
foreach my $host (sort(keys(%Status)))
|
||||
{
|
||||
|
|
@ -145,10 +147,12 @@ foreach my $host (sort(keys(%Status)))
|
|||
my $owner = $hosts_infos->{$host}->{user};
|
||||
next if (@ownerOnly and not grep {/$owner/} @ownerOnly);
|
||||
my %host_conf = %{$server->ConfigDataRead($host)};
|
||||
next if ( $host_conf{BackupsDisable} );
|
||||
next if (@hostsDesired and not grep {/$host/} @hostsDesired);
|
||||
next if (@hostsExcluded and grep {/$host/} @hostsExcluded);
|
||||
$Status{$host}{BackupsDisable} = $host_conf{BackupsDisable};
|
||||
next if ( $Status{$host}{BackupsDisable} );
|
||||
next if (@hostsDesired and not grep {/^$host$/} @hostsDesired);
|
||||
next if (@hostsExcluded and grep {/^$host$/} @hostsExcluded);
|
||||
next if ($Status{$host}{'type'} eq 'archive');
|
||||
$Status{$host}{'statusCode'} = 'OK';
|
||||
$hostCount++;
|
||||
# Debug
|
||||
if ($verbose == 2)
|
||||
|
|
@ -161,50 +165,73 @@ foreach my $host (sort(keys(%Status)))
|
|||
# Check host error
|
||||
if ($Status{$host}{'error'})
|
||||
{
|
||||
# Check connectivity errors with greater care
|
||||
if ($Status{$host}{'error'} ne 'ping too slow' &&
|
||||
$Status{$host}{'error'} ne 'no ping response' &&
|
||||
$Status{$host}{'error'} ne 'no ping response' &&
|
||||
$Status{$host}{'error'} ne 'host not found' &&
|
||||
$Status{$host}{'reason'} !~ /Reason_restore_failed/) {
|
||||
push @problems, "$host error : ".$Status{$host}{'error'}." / ".$Status{$host}{'reason'};
|
||||
next;
|
||||
}
|
||||
$Status{$host}{statusMsg} = "error: ".$Status{$host}{'error'}." / ".$Status{$host}{'reason'};
|
||||
} else {
|
||||
$Status{$host}{statusMsg} = "status: ".$Status{$host}{'state'};
|
||||
}
|
||||
|
||||
# Check last good backup time
|
||||
my $difftime=difftime(time(), $Status{$host}{'lastGoodBackupTime'});
|
||||
my $diffdays=$difftime/(3600 * 24);
|
||||
$Status{$host}{'lastGoodBackupTime'} = $Status{$host}{'startTime'} if (not $Status{$host}{'lastGoodBackupTime'});
|
||||
if ($difftime > ($critDaysOld * 3600 * 24))
|
||||
{
|
||||
push @problems, "$host : last good backup have ".sprintf("%.1f",$diffdays)." days";
|
||||
$errorLevel='CRITICAL';
|
||||
}
|
||||
elsif ($difftime > ($warnDaysOld * 3600 * 24))
|
||||
{
|
||||
push @problems, "$host : last good backup have ".sprintf("%.1f",$diffdays)." days";
|
||||
$errorLevel='WARNING' if ($errorLevel eq 'OK');
|
||||
}
|
||||
}
|
||||
|
||||
my $problemTxt="";
|
||||
if (scalar(@problems) > 0) {
|
||||
if ($verbose > 0) {
|
||||
foreach my $pbl (@problems) {
|
||||
if ($problemTxt ne "") {
|
||||
$problemTxt.=" , ";
|
||||
}
|
||||
else {
|
||||
$problemTxt=" ( ";
|
||||
}
|
||||
$problemTxt.=$pbl;
|
||||
$Status{$host}{'lastGoodBackupDays'} = difftime(time(), $Status{$host}{'lastGoodBackupTime'}) / (3600 * 24) if ( defined $Status{$host}{'lastGoodBackupTime'} );
|
||||
if ( ! $Status{$host}{'lastGoodBackupDays'} ) {
|
||||
$Status{$host}{'startDays'} = difftime(time(), $Status{$host}{'startTime'}) / (3600 * 24);
|
||||
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' );
|
||||
$statusCode = 'WARNING' unless ( $statusCode eq 'CRITICAL' );
|
||||
}
|
||||
$problemTxt.=" )";
|
||||
} elsif ( $Status{$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 )
|
||||
{
|
||||
$Status{$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' );
|
||||
} else {
|
||||
$Status{$host}{statusMsg} .= ", last good backup have ".sprintf("%.1f",$Status{$host}{'lastGoodBackupDays'})." days";
|
||||
}
|
||||
else {
|
||||
$problemTxt=" (".scalar(@problems)." problems)";
|
||||
$ok_count++ if ( $Status{$host}{statusCode} eq 'OK' );
|
||||
$unknown_count++ if ( $Status{$host}{statusCode} eq 'UNKNOWN' );
|
||||
$warning_count++ if ( $Status{$host}{statusCode} eq 'WARNING' );
|
||||
$critical_count++ if ( $Status{$host}{statusCode} eq 'CRITICAL' );
|
||||
}
|
||||
|
||||
|
||||
my $statusMsg = "BACKUPPC $statusCode";
|
||||
|
||||
if ( $statusCode eq 'OK' ) {
|
||||
if ( $verbose && scalar(@hostsDesired) == 1 ) {
|
||||
$statusMsg .= " (".$Status{$hostsDesired[0]}{statusMsg}.")";
|
||||
} else {
|
||||
$statusMsg .= " ($ok_count OK)";
|
||||
}
|
||||
} else {
|
||||
if ( $verbose ) {
|
||||
$statusMsg .= " (";
|
||||
my $first_host = 1;
|
||||
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}{'type'} eq 'archive');
|
||||
if ( $Status{$host}{statusCode} ne 'OK' ) {
|
||||
$statusMsg .= ", " unless ( $first_host );
|
||||
$statusMsg .= "$host: ".$Status{$host}{statusCode}." - ".$Status{$host}{statusMsg};
|
||||
$first_host = 0 if ( $first_host );
|
||||
}
|
||||
}
|
||||
$statusMsg .= ")";
|
||||
} else {
|
||||
$statusMsg .= " ( $ok_count OK, $unknown_count UNKNOWN, $warning_count WARNING, $critical_count CRITICAL)";
|
||||
}
|
||||
}
|
||||
|
||||
print "BACKUPPC $errorLevel$problemTxt\n";
|
||||
exit $ERRORS{$errorLevel};
|
||||
print "$statusMsg\n";
|
||||
exit $ERRORS{$statusCode};
|
||||
|
|
|
|||
Loading…
Reference in a new issue