File SAPHanaSR-monitor of Package SAPHanaSR.2773
#!/usr/bin/perl
#
# SAPHanaSR-monitor
# (c) 2014 SUSE Linux Products GmbH, Nuremberg, Germany
# (c) 2015 SUSE Linux GmbH, Nuremberg, Germany
# Author: Fabian Herschel <fabian.herschel@suse.com>
# License: GPL v2+
my $Version="0.17.2016.02.04.4";
#
##################################################################
use POSIX;
use strict;
use Sys::Syslog;
use Sys::Hostname;
use File::Path;
use Getopt::Long;
use lib '/usr/share/SAPHanaSR/tests';
use SAPHanaSRTools;
###################################
## this part is not for scale out and currently NOT zero-config
my $ClusterNodes=2;
my $ClusterPrimaries=1;
my $ClusterSecondaries=1;
#my %Name;
my %Host;
my $host = hostname();
my $varlib='/var/lib/SAPHanaTD';
my $testfile='SAPHanaTD.status';
my $testcount=0;
my $first_test=1;
my $sid="";
my @sids=();
my $ino="";
my $Format="txt";
my $Continuous="no";
my $HtmlPage="/srv/www/hawk/public/herschel/index.html";
my $Intervall=10;
my $cibFile="";
my $help=0;
sub init()
{
my $result = GetOptions ("sid=s" => \@sids,
"format=s" => \$Format,
"cont=s" => \$Continuous,
"out=s" => \$HtmlPage,
"cib=s" => \$cibFile,
"intervall=s" => \$Intervall,
"nodes=s" => \$ClusterNodes,
"help" => \$help,
);
if ( $help ) {
printf "SAPHanaSR-monitor {[--sid=<sid[:instNr]>]} [--format=(txt|html)] [--cont=(yes|no)] [--out=<file>] [--intervall=<time>] [--nodes=<nodeNr>] [--help]\n";
exit 0;
}
if ( $cibFile ) {
set_cibFile($cibFile);
}
return 0;
}
#
################ BEGIN CUT - MOVE THAT TO AN PERL-LIBRARY LATER
#
sub print_page($$$$$$$$$$$$$$$$$$$$$$$$$)
{
my ($format, $sid, $sid_color, $hana1stat, $node1, $hana2stat, $node2, $syncstat, $syncdir, $crm_out, $crm_col, $crm_errors, $attr_out, $attr_col, $attr_errors, $score_out, $score_col, $score_errors, $log_out, $log_col, $log_errors,
$ftd_out, $ftd_col, $ftd_errors, $check_lpa_col, $check_lpa_msg);
$format=shift();
$sid=shift(); $sid_color=shift();
$hana1stat=shift(); $node1=shift();
$hana2stat=shift(); $node2=shift();
$syncstat=shift(); $syncdir=shift();
$crm_out=shift(); $crm_col=shift(); $crm_errors=shift();
$attr_out=shift(); $attr_col=shift(); $attr_errors=shift();
$score_out=shift(); $score_col=shift(); $score_errors=shift();
$log_out=shift(); $log_col=shift(); $log_errors=shift();
$ftd_out=shift(); $ftd_col=shift(); $ftd_errors=shift();
$check_lpa_col=shift(); $check_lpa_msg=shift();
# in the page layout $score_col is ignored right now
# TODO: Check id we add more and more parameters to the print_page or if we will use an hash as parameter or
# id we (recover) more values here inside of the function like for the lpa values:
my $lpa_node1=get_lpa_by_host($sid, $node1);
my $lpa_node2=get_lpa_by_host($sid, $node2);
my $mydate=`date`; # TODO: do it with perl ;-)
chomp($crm_errors);
chomp($attr_errors);
chomp($log_errors);
if ($format eq "txt") {
my $syncArrow="??";
my $SID=uc($sid);
if ( $syncdir eq "right" ) { $syncArrow = "=>"; }
if ( $syncdir eq "left" ) { $syncArrow = "<="; }
my $myPage = <<EOF;
OVERALL: $SID: $sid_color $mydate
SR: $syncstat $node1 $syncArrow $node2
Cluster: $crm_col ($crm_errors)
HANA1: $node1 ${hana1stat}
HANA2: $node2 ${hana2stat}
SAPHanaSR: $attr_col ($attr_errors)
LPA: $check_lpa_col ($check_lpa_msg)
LPA: $node1 $lpa_node1
LPA: $node2 $lpa_node2
Syslog: $log_col ($log_errors)
EOF
printf( "%s\n", $myPage);
} elsif ($format eq "html") {
my $myPage = <<EOF;
<!DOCTYPE html>
<html lang="en">
<head>
<title>SAPHanaSR status</title>
<meta http-equiv="refresh" content="10; url=." />
<meta charset="utf-8"/>
</head>
<body class="">
<table cellpadding=5>
<tr><td valign="top" bgcolor="$sid_color"><h2>$sid</h2></td>
<td valign="top"><b>SAPHanaSR-monitor $mydate</b>
<table cellpadding=1>
<tr><td valign="top"><img width="100%" height="100%" src="final_icons/server_${hana1stat}.png"/></td>
<td><img width="100%" height="100%" src="final_icons/arrow_${syncdir}_${syncstat}.png"/></td>
<td valign="top"><img width="100%" height="100%" src="final_icons/server_${hana2stat}.png"/></td>
</tr>
<tr><td valign="top" align="center"><b>$node1</b></td>
<td valign="top" align="center"><b>$sid</b></td>
<td valign="top" align="center"><b>$node2</b></td>
<tr>
</table>
</td>
<td valign="top" width="5%"><img width="100%" height="100%" src="suse_logo_w-tag_color.jpg"/></td></tr>
<tr><td valign="top" bgcolor="$crm_col"><h2>Crm</h2></td>
<td valign="top" bgcolor="#B9B9B9"><pre>$crm_out</pre></td>
<td valign="top" bgcolor="#B9B9B9"><pre>$crm_errors</pre></td>
</tr>
<tr><td valign="top" bgcolor="$attr_col" rowspan="2"><h2>Attr</h2></td>
<td valign="top" bgcolor="#B9B9B9"><pre>$attr_out</pre></td>
<td valign="top" bgcolor="#B9B9B9"><pre>$attr_errors</pre></td>
</tr>
<tr><td bgcolor="#B9B9B9"><pre>$score_out</pre></td>
<td valign="top" bgcolor="#B9B9B9"><pre></pre></td>
</tr>
<tr><td valign="top" bgcolor="$log_col"><h2>Log</h2></td>
<td valign="top" bgcolor="#B9B9B9"><pre>$log_out</pre></td>
<td valign="top" bgcolor="#B9B9B9"><pre>$log_errors</pre></td>
</tr>
<tr><td valign="top" bgcolor="#B9B9B9"><h2>TD</h2></td>
<td valign="top" bgcolor="#B9B9B9"><pre>$ftd_out</pre></td>
<td valign="top" bgcolor="#B9B9B9"><pre></pre></td>
</tr>
</table>
</body>
</html>
EOF
open HTML, ">$HtmlPage.new";
print HTML $myPage;
close HTML;
system ("mv $HtmlPage.new $HtmlPage");
} # end html
}
my $ident = "SAPHanaSR-mon";
my $logopt = "pid";
my $facility = "LOCAL0";
my $priority = "info";
my %Colors;
$Colors{"green"}="#7AC142";
$Colors{"yellow"}="#fde944";
$Colors{"red"}="#BB0000";
$Colors{"gray"}="#B9B9B9";
openlog $ident, $logopt, $facility;
#( $sid, $ino ) = get_sid_and_InstNr();
mysyslog $priority, "%s", "Monitor startup";
sub processMonitor($) {
my $sid = shift();
my $message;
my $phost;
my $shost;
my $laststate="";
#
%Host=();
get_hana_attributes($sid);
my $crm_out=`crm_mon -r1 -D`;
my $attr_out=host_attr2string();
my $score_out=`crm_simulate -Ls | grep -i promo`;
my $log_out=`grep 'SAPHana.*INFO: RA' /var/log/messages | tail -20`; # TODO: Do we need to optimize the runtime of this command?
my $ftd_out=`grep 'fhTD' /var/log/messages | tail -10`;
my $ftd_col="gray";
my $crm_col="green";
my $attr_col="green";
my $score_col="green";
my $log_col="green";
my $complete_col="green";
#
# for the grafic output
#
my @node_list;
if ( $cibFile eq "" ) {
open CRM_NODES, "crm_node -l |";
while (<CRM_NODES>) {
if (/^[0-9]+\W+(\w+)\W/) {
push (@node_list, $1);
}
}
close CRM_NODES;
} else {
# TODO: For the "cold" cib case we need to check standby, maintenance and online status
open CIB, "<$cibFile";
while (<CIB>) {
# <node id="gdepfhb2" uname="gdepfhb2">
if ( /<node .*uname="([^"]+)"/ ) {
push (@node_list, $1);
}
}
close CIB;
}
@node_list = ( sort (@node_list ));
my $sync_col="green";
my ( $first_node_name, $second_node_name ) = @node_list;
my %hanaStatus;
#unset %hanaStatus;
my $thenode;
foreach $thenode ( @node_list ) {
$hanaStatus{$thenode}="grey";
if ( check_node_status($sid, "34" , $thenode) ) { $hanaStatus{$thenode}="green"; }
if ( check_node_status($sid, "2" , $thenode) ) { $hanaStatus{$thenode}="yellow"; }
if ( check_node_status($sid, "1" , $thenode) ) { $hanaStatus{$thenode}="red"; }
}
if ( get_hana_sync_state($sid) ne "SOK" ) { $sync_col="red"; }
my $sync_direction="none";
if ( check_node_mode( $sid, "P" , $node_list[0]) && check_node_mode( $sid, "S" , $node_list[1]) ) {
$sync_direction="right";
} elsif ( check_node_mode( $sid, "S" , $node_list[0]) && check_node_mode( $sid, "P" , $node_list[1]) ) {
$sync_direction="left";
} else {
$sync_direction="none";
$sync_col="grey";
}
#printf "first node: %s \n", $first_node_name;
#printf "second node: %s \n", $second_node_name;
my $first_node_col="green";
my $second_node_col="green";
#
# crm checks
#
my $crm_errors="";
if ( get_nodes_online() != $ClusterNodes ) { $crm_col="yellow"; $crm_errors .= " C#n\n";}
# attr checks
my $attr_errors="";
if ( get_number_primary($sid, "1234") != 1 ) { $attr_col="yellow"; $attr_errors .= " A#p"; }
if ( get_number_primary($sid, "234") != 1 ) { $attr_col="yellow"; $attr_errors .= " A-P"; }
if ( get_number_secondary($sid, "1234") != 1 ) { $attr_col="yellow"; $attr_errors .= " A-s";}
if ( get_number_secondary($sid, "234") != 1 ) { $attr_col="yellow"; $attr_errors .= " A-S";}
if ( get_number_primary($sid, "234") > 1 ) { $attr_col="red"; $attr_errors .= " A-PP";}
if ( get_hana_sync_state($sid) ne "SOK" ) { $attr_col="yellow"; $attr_errors .= " A-SFAIL"; }
#printf "\n";
#printf "%-16s %-16s %-16s\n", $hanaStatus{$node_list[0]}, "$sync_direction : $sync_col", $hanaStatus{$node_list[1]};
#printf "%-16s %-16s %-16s\n", $node_list[0], $sid, $node_list[1];
#printf "\n";
# log checks
my $log_errors="";
my $check_lpa_status_col;
my $check_lpa_msg;
( $check_lpa_status_col, $check_lpa_msg)=check_lpa_status($sid, $node_list[0], $node_list[1]);
if ( ( $attr_col eq "yellow" ) || ( $crm_col eq "yellow" ) ||
( $log_col eq "yellow" ) || ( $attr_col eq "yellow" ) ||
( $score_col eq "yellow" ) || ( $check_lpa_status_col eq "yellow") )
{ $complete_col="yellow"; }
if ( ( $attr_col eq "red" ) || ( $crm_col eq "red" ) ||
( $log_col eq "red" ) || ( $attr_col eq "red" ) ||
( $score_col eq "red" ) || ( $check_lpa_status_col eq "red"))
{ $complete_col="red"; }
print_page($Format,$sid, $complete_col,
$hanaStatus{$node_list[0]}, $node_list[0],
$hanaStatus{$node_list[1]}, $node_list[1],
$sync_col, $sync_direction,
$crm_out, $crm_col, $crm_errors,
$attr_out, $attr_col, $attr_errors,
$score_out, $attr_col, $attr_errors,
$log_out, $log_col, $log_errors,
$ftd_out, $ftd_col, $check_lpa_status_col, $check_lpa_msg
);
#
$phost=get_host_primary($sid, "1234");
$shost=get_host_secondary($sid, "1234");
my ($checkOK, $failures ) = check_all_ok($sid, $ClusterNodes);
if ( $checkOK == 0 ) {
$message="All checks passed - primary=$phost secondary=$shost";
mysyslog $priority, "%s", $message;
} else {
if ( $checkOK == 1 ) {
$message="$checkOK check failed ($failures) - primary=$phost secondary=$shost";
} else {
$message="$checkOK checks failed ($failures) - primary=$phost secondary=$shost";
}
mysyslog $priority, "%s", $message;
}
}
init();
if ( 0 == @sids ) {
my $sid_ino_list;
( $sid_ino_list ) = get_sid_and_InstNr();
@sids = split(",", $sid_ino_list);
}
foreach $sid ( @sids ) {
( $sid, $ino ) = split(":", $sid);
processMonitor($sid);
}
while ( $Continuous eq "yes" ) {
sleep $Intervall;
foreach $sid ( @sids ) {
( $sid, $ino ) = split(":", $sid);
processMonitor($sid);
}
}
closelog;
#
