mirror of https://github.com/schoebel/mars
same small fixes.
This commit is contained in:
jmann
Thomas Schoebel-Theuer
parent
f1122eccf7
commit
4bc3076f7f
|
|
@ -1,7 +1,7 @@
|
|||
#!/usr/bin/perl -w
|
||||
# (c) 2012 Joerg Mann / 1&1 Internet AG
|
||||
#
|
||||
# $Id: 38d2a6826ed99f22cf5ca709ae31e925c8dde7d3 $
|
||||
# $Id: 68e1b49d98800e0343dac1d6512db8bebd6a7581 $
|
||||
# last update at now ...
|
||||
|
||||
# TODO:
|
||||
|
|
@ -10,10 +10,10 @@
|
|||
# check monitoring -> redesign statuscodes
|
||||
|
||||
# 20121201 - redesign debug-files and -messages
|
||||
# 20120204 - add display todo counter
|
||||
# 20120205 - add/update system entrys
|
||||
|
||||
|
||||
# 20121204 - add display todo counter
|
||||
# 20121205 - add/update system entrys
|
||||
# 20121206 - upgrade LogDelay/LogSpeed
|
||||
# 20121210 - optimize same code
|
||||
|
||||
|
||||
###
|
||||
|
|
@ -27,8 +27,8 @@ use POSIX qw(strftime);
|
|||
use File::Basename;
|
||||
|
||||
### defaults
|
||||
my $version = "0.070-12";
|
||||
my $alife_timeout = "99"; # sec
|
||||
my $version = "0.070-20";
|
||||
my $alife_timeout = "30"; # sec
|
||||
my $is_tty = 0;
|
||||
my $mars_dir = '/mars';
|
||||
my $himself = `uname -n` or die "cannot determine my network node name\n";
|
||||
|
|
@ -43,7 +43,7 @@ chomp $himself;
|
|||
### ARGV
|
||||
# Optionen in Hash-Ref parsen
|
||||
my $params = {};
|
||||
GetOptions( $params, 'help', 'h', 'version', 'v', 'monitor', 'role', 'cstate', 'dstate', 'resource=s', 'system', 'interval=i', 'history', 'debug=s' );
|
||||
GetOptions( $params, 'help', 'h', 'version', 'v', 'monitor', 'role', 'cstate', 'dstate', 'resource=s', 'system', 'interval=i', 'history', 'debug' );
|
||||
|
||||
### small help
|
||||
sub display_help {
|
||||
|
|
@ -52,16 +52,16 @@ sub display_help {
|
|||
print "Usage: mars-status [--help]\n";
|
||||
print "Usage: mars-status [--version]\n";
|
||||
print "Usage: mars-status (without specification of parameters, an abstract of all the informations spent)\n";
|
||||
print "Usage: mars-status [--resource <RESNAME>] [--interval <seconds>] | [[--history] | [--debug <level>] | [--system]]\n";
|
||||
print "Usage: mars-status [--resource <RESNAME>] [--interval <seconds>] | [--history] | [--debug ] | [--system]\n";
|
||||
print "Usage: mars-status [--resource <RESNAME>] --monitor \n";
|
||||
print "Usage: mars-status --resource <RESNAME> [--role | --cstate | --dstate]\n";
|
||||
print " --resource : limits the display to the specified resource\n";
|
||||
print " --interval : refreshes the display every second xxx\n";
|
||||
print " --history : shows information about the log files, version numbers and their status\n";
|
||||
print " --system : display mars-system informations\n";
|
||||
print " --monitor : indicator to use for monitoring on all state\n";
|
||||
print " --monitor : indicator to use for monitoring on all state (by local node only !)\n";
|
||||
print " --role|--cstate|--dstate single state on lokal node\n";
|
||||
print " --debug : additional display debug messages by level (level: debug,info,warn,error,fatal,total)\n";
|
||||
print " --debug : additional display debug messages\n";
|
||||
print "\nAdvanced information are also available here: http://http://wiki.intranet.1and1.com/bin/view/PO/woauchimmer\n";
|
||||
exit;
|
||||
}
|
||||
|
|
@ -90,18 +90,8 @@ if ( $tty =~ /^\/dev\/pts\// ) {
|
|||
}
|
||||
|
||||
|
||||
### subs
|
||||
sub check_link {
|
||||
my $dir = shift;
|
||||
my $result = readlink $dir;
|
||||
if ( !$result ) {
|
||||
return 0;
|
||||
} else {
|
||||
return $result;
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
#########################################################################################
|
||||
### print color
|
||||
sub print_screen {
|
||||
my $Text = shift;
|
||||
|
|
@ -133,10 +123,24 @@ sub print_screen {
|
|||
}
|
||||
|
||||
|
||||
#########################################################################################
|
||||
### read link
|
||||
sub check_link {
|
||||
my $dir = shift;
|
||||
my $result = readlink $dir;
|
||||
if ( !$result ) {
|
||||
return 0;
|
||||
} else {
|
||||
return $result;
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
#########################################################################################
|
||||
### read links
|
||||
sub convert_link {
|
||||
my $link = shift;
|
||||
$link = check_link "$link";
|
||||
$link = check_link "$link";
|
||||
if (( !$link ) || ( $link eq 0 )) {
|
||||
print_screen "off", 'red';
|
||||
} else {
|
||||
|
|
@ -146,11 +150,11 @@ sub convert_link {
|
|||
}
|
||||
|
||||
|
||||
#########################################################################################
|
||||
### StatusCode
|
||||
sub monitoring {
|
||||
my $Code = shift;
|
||||
my $Text = shift;
|
||||
|
||||
# UpToDate - eishokey
|
||||
# UpDateIng - worker
|
||||
# OutDate - replaying
|
||||
|
|
@ -180,18 +184,18 @@ sub monitoring {
|
|||
#########################################################################################
|
||||
### sub display resource-partner
|
||||
sub display_partner {
|
||||
my %p = @_;
|
||||
my $PRes = $p{ressource};
|
||||
my $PName = $p{nodename};
|
||||
my $PSize = $p{ressource_size};
|
||||
my $ref_ResPartner = $p{res_partner};
|
||||
my $ref_AULogfile = $p{res_AULogfile};
|
||||
|
||||
my $PStatus = check_link "$mars_dir/$PRes/primary";
|
||||
my $PDevice = check_link "$mars_dir/$PRes/device-$PName";
|
||||
my $Ljoined = check_link "$mars_dir/$PRes/device-$himself";
|
||||
my %p = @_;
|
||||
my $PRes = $p{ressource};
|
||||
my $PName = $p{nodename};
|
||||
my $PSize = $p{ressource_size};
|
||||
my $ref_ResPartner = $p{res_partner};
|
||||
my $ref_AULogfile = $p{res_AULogfile};
|
||||
my $PStatus = check_link "$mars_dir/$PRes/primary";
|
||||
my $PDevice = check_link "$mars_dir/$PRes/device-$PName";
|
||||
my $Ljoined = check_link "$mars_dir/$PRes/device-$himself";
|
||||
|
||||
### status
|
||||
##########################################################################
|
||||
### check status
|
||||
if ( $PStatus eq $PName ) {
|
||||
print_screen "Primary",'blue';
|
||||
print_screen "Primary [$PRes on $PName]",'', 'Rmonitor';
|
||||
|
|
@ -201,31 +205,33 @@ sub display_partner {
|
|||
print_screen "not joined",'red';
|
||||
print_screen "not joined\n",'', 'Rmonitor';
|
||||
print_screen " -> Resource is not joined to this node\n", 'red';
|
||||
monitoring "unknown", "not joined";
|
||||
monitoring "unknown", "not joined";
|
||||
return;
|
||||
} else {
|
||||
print_screen "Secondary",'blue';
|
||||
print_screen "Secondary [$PRes on $PName]",'', 'Rmonitor';
|
||||
monitoring '', "joined";
|
||||
monitoring "", "joined";
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
### alive
|
||||
##########################################################################
|
||||
### check alive
|
||||
my @PAlive = lstat("$mars_dir/alive-$PName");
|
||||
if ( !$PAlive[9] ) { $PAlive[9]=0 };
|
||||
if ( !$PAlive[9] ) { $PAlive[9] = 0 };
|
||||
my $PAlive = time()- $PAlive[9] - $alife_timeout;
|
||||
print_screen ", System", '';
|
||||
if ( $PAlive > 1 ) {
|
||||
print_screen " unknown (last message before $PAlive sec) !!!\n", 'red';
|
||||
monitoring 'Failed', "not alive"
|
||||
monitoring "Failed", "not alive"
|
||||
} else {
|
||||
print_screen " alive\n",'green';
|
||||
monitoring '', "alive";
|
||||
monitoring "", "alive";
|
||||
}
|
||||
|
||||
|
||||
### device
|
||||
##########################################################################
|
||||
### check device
|
||||
print_screen "\tDevices : Disk-Device ".check_link "$mars_dir/$PRes/data-$PName";
|
||||
print_screen ", used as Mars-Device /dev/mars/$PDevice";
|
||||
my $ASize = check_link "$mars_dir/$PRes/actsize-$PName";
|
||||
|
|
@ -236,7 +242,8 @@ sub display_partner {
|
|||
}
|
||||
|
||||
|
||||
# check mountpint
|
||||
##########################################################################
|
||||
### check mountpint
|
||||
if ( $himself eq $PName ) {
|
||||
my $PUDevice = "/dev/mars/$PDevice";
|
||||
if ( stat( $PUDevice) ) {
|
||||
|
|
@ -257,15 +264,16 @@ sub display_partner {
|
|||
$$ref_ResPartner++;
|
||||
|
||||
|
||||
### sync
|
||||
##########################################################################
|
||||
### check sync
|
||||
### sync - status
|
||||
my $PSyncsize = check_link "$mars_dir/$PRes/syncstatus-$PName";
|
||||
my $SStatus = sprintf("%.2f", ($PSyncsize / $PSize * 100));
|
||||
my $SStatus = sprintf("%.2f", ($PSyncsize / $PSize * 100));
|
||||
print_screen (sprintf "\tSync : %s bytes (%.3fTB) synced = ", $PSyncsize, ( $PSyncsize/1024/1024/1024/1024));
|
||||
|
||||
### sync - speed
|
||||
my $SSpeed = check_link "$mars_dir/$PRes/actual-$PName/sync_rate";
|
||||
$SSpeed = sprintf ("%.2f", $SSpeed / 1024 / 1024);
|
||||
my $SSpeed = check_link "$mars_dir/$PRes/actual-$PName/sync_rate";
|
||||
$SSpeed = sprintf ("%.2f", $SSpeed / 1024 / 1024);
|
||||
if ( $SSpeed eq "0.00" ) {
|
||||
$SSpeed = "%";
|
||||
} else {
|
||||
|
|
@ -276,68 +284,81 @@ sub display_partner {
|
|||
if ( $SStatus < 100) {
|
||||
print_screen "$SStatus$SSpeed\n";
|
||||
print_screen "\t\t---> WORK: Sync in progress = ($SStatus% < 100.00%)\n", 'red';
|
||||
monitoring "InvaliDate", "not in sync ($SStatus%)";
|
||||
monitoring "InvaliDate", "not in sync ($SStatus%)";
|
||||
} else {
|
||||
print_screen "$SStatus$SSpeed\n", 'green';
|
||||
monitoring '', "synced";
|
||||
monitoring "", "synced";
|
||||
}
|
||||
|
||||
|
||||
##########################################################################
|
||||
### TODO: work by resize ...
|
||||
#print "\n$PSize\n$PSyncsize\n$PName";
|
||||
|
||||
|
||||
### logfile
|
||||
##########################################################################
|
||||
### check logfile
|
||||
### logfile - status
|
||||
my @PLogFile = split (',', check_link "$mars_dir/$PRes/replay-$PName" );
|
||||
my @PLogLink = split ("-", $PLogFile[0]);
|
||||
### TODO: kein Logfile vorhanden ...
|
||||
### TODO: kein Logfile vorhanden ... mmh.
|
||||
my $PLogName = "$PLogLink[0]-$PLogLink[1]";
|
||||
my $PLogSize = -s "$mars_dir/$PRes/$PLogFile[0]";
|
||||
if ( !$PLogFile[1] ) {
|
||||
$PLogFile[1] = 0; $PLogFile[2] = 0;
|
||||
}
|
||||
$PLogSize = 0.0001 if (( !$PLogSize ) || ( $PLogSize eq 0 ));
|
||||
|
||||
### logfile - status
|
||||
$PLogSize = 0.0001 if (( !$PLogSize ) || ( $PLogSize eq 0 ));
|
||||
my $LogSpeed = check_link "$mars_dir/$PRes/actual-$PName/file_rate";
|
||||
$LogSpeed = sprintf ("%.2f", $LogSpeed / 1024 / 1024);
|
||||
if ( $LogSpeed eq "0.00" ) {
|
||||
$LogSpeed = ""
|
||||
} else {
|
||||
$LogSpeed = ", Logfiles received with $LogSpeed mb/s";
|
||||
}
|
||||
|
||||
$LogSpeed = sprintf ("%.2f", $LogSpeed / 1024 / 1024);
|
||||
|
||||
### logfile - delaytime
|
||||
### TODO: function buggy by mars links ...
|
||||
my $LogDelay = "";
|
||||
my $LogDelay = "0.000000000";
|
||||
my $NewLogDelay = "0.000000000";
|
||||
my @LogDelayLink = lstat ("$mars_dir/$PRes/actual-$PName/timestamp");
|
||||
my @LogDelayTime = split (',', check_link "$mars_dir/$PRes/actual-$PName/timestamp");
|
||||
|
||||
### offset replay
|
||||
if (( $LogDelayTime[4] ) && ( $LogDelayTime[4] ne '0.000000000' )) {
|
||||
# if ( $LogDelayTime[4] ) {
|
||||
### lstat
|
||||
my @LogDelayLink = lstat ("$mars_dir/$PRes/actual-$PName/timestamp");
|
||||
$LogDelay = strftime("%H:%M:%S", gmtime( $LogDelayLink[9] - $LogDelayTime[4]) );
|
||||
my $LogDelayMonitor = $LogDelay;
|
||||
$LogDelay = ", Delay $LogDelay - but not really sure ;)";
|
||||
|
||||
### monitoring
|
||||
my ($h,$m,$s) = split /:/, $LogDelayMonitor;
|
||||
$LogDelayMonitor = (($h*3600) + ($m*60) + $s);
|
||||
if ( $LogDelayMonitor eq 0 ) { # 0
|
||||
monitoring "UpToDate", "Delay $LogDelayMonitor sec";
|
||||
} elsif ( $LogDelayMonitor < 60 ) { # unter 1 min
|
||||
monitoring "UpDateIng", "Delay $LogDelayMonitor sec";
|
||||
} else { # rest
|
||||
monitoring "OutDate", "Delay $LogDelayMonitor sec";
|
||||
}
|
||||
# my $A = strftime ("%H:%M:%S", gmtime($LogDelayTime[4]) );
|
||||
# my $B = strftime ("%H:%M:%S", gmtime($LogDelayLink[9]) );
|
||||
# print "*** $A / $B / $LogDelay / $LogDelayMonitor ***\n";
|
||||
}
|
||||
$LogDelay = $LogDelayLink[9] - $LogDelayTime[4];
|
||||
}
|
||||
|
||||
### offset newer logfile
|
||||
if ( ( $ref_AULogfile ) && !($PLogName eq $ref_AULogfile) ) {
|
||||
my @NewLogDelay = `ls $mars_dir/$PRes/$ref_AULogfile*`;
|
||||
$NewLogDelay[0] =~ s/\n//;
|
||||
@NewLogDelay = stat ("$NewLogDelay[0]");
|
||||
#$NewLogDelay = $NewLogDelay[9];
|
||||
$LogDelay = $LogDelayLink[9] - $NewLogDelay[9];
|
||||
}
|
||||
$LogDelay = strftime("%H:%M:%S", gmtime($LogDelay));
|
||||
|
||||
#print "*ld $LogDelay\n";
|
||||
#print "#dl9 lstat - $LogDelayLink[9] - ".gmtime($LogDelayLink[9])."\n";
|
||||
#print "#dl @LogDelayLink\n";
|
||||
#print "#dt4 link - $LogDelayTime[4] - ".gmtime($LogDelayTime[4])."\n";
|
||||
#print "#dt @LogDelayTime\n";
|
||||
#print "#nl $NewLogDelay\n";
|
||||
|
||||
### log delay monitoring
|
||||
my $LogDelayMonitor = $LogDelay;
|
||||
my ($h,$m,$s) = split /:/, $LogDelayMonitor;
|
||||
$LogDelayMonitor = (($h*3600) + ($m*60) + $s);
|
||||
if ( $LogDelayMonitor eq 0 ) { # 0
|
||||
monitoring "UpToDate", "Delay $LogDelayMonitor sec";
|
||||
} elsif ( $LogDelayMonitor < 60 ) { # unter 1 min
|
||||
monitoring "UpDateIng", "Delay $LogDelayMonitor sec";
|
||||
} else { # rest
|
||||
monitoring "OutDate", "Delay $LogDelayMonitor sec";
|
||||
}
|
||||
|
||||
|
||||
### logfile - results
|
||||
print_screen (sprintf "\tLogfile : %s bytes (%.3fGB) in %s active %s %s\n", $PLogSize, ( $PLogSize/1024/1024/1024 ), $PLogName, $LogSpeed, $LogDelay);
|
||||
print_screen (sprintf "\tLogfile : %s bytes (%.3fGB) in ", $PLogSize, ( $PLogSize/1024/1024/1024 ));
|
||||
print_screen "$PLogName", 'green';
|
||||
print_screen " active";
|
||||
print_screen ", received with $LogSpeed mb/s" if ( $LogSpeed ne "0.00" );
|
||||
print_screen ", Delay roughly $LogDelay - but not really sure ;)" if ( $LogDelay ne "00:00:00" );
|
||||
print_screen "\n";
|
||||
if ( $Ljoined eq "0" || $PLogSize eq "0.0001" ) {
|
||||
print_screen "\t\t---> WORK: Logfile empty = (Size: $PLogSize)\n", 'red';
|
||||
}
|
||||
|
|
@ -346,17 +367,18 @@ sub display_partner {
|
|||
}
|
||||
|
||||
|
||||
### replay ...
|
||||
##########################################################################
|
||||
### check replay ...
|
||||
### replay - status
|
||||
my $RStatus = sprintf("%.2f", ( $PLogFile[1] / $PLogSize * 100));
|
||||
$RStatus = 0 if ( $Ljoined eq "0" || $PLogSize eq "1" );
|
||||
$RStatus = 99.99 if (( $PLogFile[1] ne $PLogSize ) && ( $RStatus eq "100.00" ));
|
||||
$RStatus = 0 if ( $Ljoined eq "0" || $PLogSize eq "1" );
|
||||
$RStatus = 99.99 if (( $PLogFile[1] ne $PLogSize ) && ( $RStatus eq "100.00" ));
|
||||
print_screen sprintf ("\tReplayed: %s bytes (%.3fGB) now replayed, Todo %d (%.3fGB) = ",
|
||||
$PLogFile[1], ( $PLogFile[1]/1024/1024/1024 ), $PLogFile[2], ( $PLogFile[2]/1024/1024/1024 ));
|
||||
|
||||
### replay - speed
|
||||
my $RSpeed = check_link "$mars_dir/$PRes/actual-$PName/replay_rate";
|
||||
$RSpeed = sprintf ("%.2f", $RSpeed / 1024 / 1024);
|
||||
$RSpeed = sprintf ("%.2f", $RSpeed / 1024 / 1024);
|
||||
if ( $RSpeed eq "0.00" ) {
|
||||
$RSpeed = "%";
|
||||
} else {
|
||||
|
|
@ -402,7 +424,8 @@ sub display_partner {
|
|||
}
|
||||
|
||||
|
||||
### actual
|
||||
##########################################################################
|
||||
### check actual
|
||||
my $ActStatus = check_link "$mars_dir/$PRes/actual-$PName/is-primary";
|
||||
if ( $ActStatus eq 1 ) {
|
||||
print_screen "\tActual : Status=Primary, used Device=";
|
||||
|
|
@ -416,60 +439,62 @@ sub display_partner {
|
|||
print_screen "\n";
|
||||
}
|
||||
|
||||
|
||||
### switches
|
||||
##########################################################################
|
||||
### check switches
|
||||
my $SWStatus;
|
||||
print_screen "\tSwitches: Attach=";
|
||||
if ( readlink "$mars_dir/$PRes/todo-$PName/attach" eq 1 ) {
|
||||
print_screen "on", 'green';
|
||||
monitoring '', "attached";
|
||||
monitoring "", "attached";
|
||||
} else {
|
||||
print_screen "off", 'red';
|
||||
monitoring "SwitchOff", "attach off";
|
||||
monitoring "SwitchOff", "attach off";
|
||||
}
|
||||
print_screen " [masked:" if ( $ActStatus eq 1 );
|
||||
print_screen " Connect=";
|
||||
if ( readlink "$mars_dir/$PRes/todo-$PName/connect" eq 1 ) {
|
||||
print_screen "on", 'green';
|
||||
monitoring '', "connected";
|
||||
monitoring "", "connected";
|
||||
} else {
|
||||
print_screen "off", 'red';
|
||||
monitoring "SwitchOff", "connect off";
|
||||
monitoring "SwitchOff", "connect off";
|
||||
}
|
||||
print_screen " Sync=";
|
||||
if ( readlink "$mars_dir/$PRes/todo-$PName/sync" eq 1 ) {
|
||||
print_screen "on", 'green';
|
||||
monitoring '', "synced";
|
||||
monitoring "", "synced";
|
||||
} else {
|
||||
print_screen "off", 'red';
|
||||
monitoring "SwitchOff", "sync off";
|
||||
monitoring "SwitchOff", "sync off";
|
||||
}
|
||||
print_screen " AllowReplay=" ;
|
||||
if ( readlink "$mars_dir/$PRes/todo-$PName/allow-replay" eq 1 ) {
|
||||
print_screen "on", 'green';
|
||||
monitoring '', "replayed";
|
||||
monitoring "", "replayed";
|
||||
} else {
|
||||
print_screen "off", 'red';
|
||||
monitoring "SwitchOff", "replay off";
|
||||
monitoring "SwitchOff", "replay off";
|
||||
}
|
||||
print_screen "]" if ( $ActStatus eq 1 );
|
||||
print_screen "\n";
|
||||
|
||||
|
||||
##########################################################################
|
||||
### node status
|
||||
print_screen "\tStatus : $NodeStatusCode = ", '';
|
||||
my $NodeStatusText = '';
|
||||
foreach (@NodeStatusText) {
|
||||
$NodeStatusText = "$NodeStatusText($_)";
|
||||
}
|
||||
print_screen "$NodeStatusText\n", '';
|
||||
### normal-modus
|
||||
print_screen "\tStatus : $NodeStatusCode = $NodeStatusText\n", '';
|
||||
### monitor-modus
|
||||
print_screen "$NodeStatusCode [$NodeStatusText]", '', 'Dmonitor';
|
||||
|
||||
if ( $NodeStatusCode eq 'SwitchOff' ) {
|
||||
print_screen "Disconnect [$PRes on $PName]", '', 'Cmonitor';
|
||||
} else {
|
||||
print_screen "Connect [$PRes on $PName]", '', 'Cmonitor';
|
||||
}
|
||||
### reset values
|
||||
$NodeStatusCode = 'UpToDate';
|
||||
@NodeStatusText = ();
|
||||
|
||||
|
|
@ -487,19 +512,20 @@ sub check_ressource {
|
|||
next;
|
||||
}
|
||||
|
||||
### read resources
|
||||
foreach my $res (@resources) {
|
||||
my $ResPartner = 0;
|
||||
my $res_name = $res;
|
||||
$res_name =~ s/^resource-//;
|
||||
my $ResPartner = 0;
|
||||
my $res_name = $res;
|
||||
$res_name =~ s/^resource-//;
|
||||
if ( $params->{'resource'} ) {
|
||||
if (!( $params->{'resource'} eq $res_name)) {
|
||||
next;
|
||||
}
|
||||
}
|
||||
my $res_size = check_link "$mars_dir/$res/size";
|
||||
my $res_size = check_link "$mars_dir/$res/size";
|
||||
if ( $res_size eq 0 ) { $res_size = 1 };
|
||||
my $res_tbsize = ( $res_size) / 1024 / 1024 /1024 / 1024;
|
||||
my $res_master = check_link "$mars_dir/$res/primary";
|
||||
my $res_tbsize = ( $res_size) / 1024 / 1024 /1024 / 1024;
|
||||
my $res_master = check_link "$mars_dir/$res/primary";
|
||||
if ( $res_master eq 0 ) { $res_master = "unknown" };
|
||||
#print_screen sprintf("-> check resource %s, with %d bytes (%.3fTB), Primary Node is %s\n", $res_name, $res_size, $res_tbsize, $res_master), 'bold';
|
||||
print_screen sprintf("-> check resource %s, with %.3fTB, Primary Node is %s\n", $res_name, $res_tbsize, $res_master), 'bold';
|
||||
|
|
@ -555,8 +581,12 @@ sub check_ressource {
|
|||
|
||||
### check resources debug
|
||||
if ($params->{'debug'}) {
|
||||
my $debug_res = check_debugfile("$res", $params->{'debug'});
|
||||
print_screen " -> Debug:\n", 'red'; print_screen "$debug_res" if ( $debug_res );
|
||||
print_screen " -> $res-Debug:\n", 'red';
|
||||
my $debug_res;
|
||||
### TODO: small hack, read 3 files ...
|
||||
$debug_res = check_debugfile("$res", "2.warn"); print_screen "$debug_res" if ( $debug_res );
|
||||
$debug_res = check_debugfile("$res", "3.error"); print_screen "$debug_res" if ( $debug_res );
|
||||
$debug_res = check_debugfile("$res", "4.fatal"); print_screen "$debug_res" if ( $debug_res );
|
||||
}
|
||||
|
||||
} # end foreach
|
||||
|
|
@ -572,14 +602,13 @@ sub check_logfile {
|
|||
my $LogCountSum = 0;
|
||||
my $LogCountNow = 1;
|
||||
my $LogFailed = 0;
|
||||
print_screen " -> History Replay/Status\n",'blue';
|
||||
|
||||
my @logfile = <$mars_dir/$LResource/log*>;
|
||||
### mal fix zaehlen ...
|
||||
foreach (@logfile) {
|
||||
$LogCountSum++;
|
||||
}
|
||||
|
||||
print_screen " -> History Replay/Status\n",'blue';
|
||||
foreach my $logfile (@logfile) {
|
||||
my $LVersion = $logfile;
|
||||
$LVersion =~ s/^.*log-([0-9]+)-.*$/$1/;
|
||||
|
|
@ -710,6 +739,7 @@ sub check_disk_is_full {
|
|||
#########################################################################################
|
||||
### check debug-files
|
||||
sub check_debugfile {
|
||||
### TODO: Fix Level
|
||||
### 0.debug,1.info,2.warn,3.error,4.fatal,5.total
|
||||
my $debug_dir = shift;
|
||||
my $debug_level = shift;
|
||||
|
|
@ -793,11 +823,12 @@ sub check_jammed {
|
|||
#########################################################################################
|
||||
### limit's auslesen ...
|
||||
sub check_limit {
|
||||
my $LimitText = shift; # sol-text
|
||||
my $LimitSolVar = shift; # sol-filename
|
||||
my $LimitSolEin = shift; # sol-einheit
|
||||
my $LimitIstVar = shift; # ist-filename
|
||||
my $LimitIstEin = shift; # ist-einheit
|
||||
my $LimitText = shift; # sol-text
|
||||
my $LimitSolVar = shift; # sol-filename
|
||||
my $LimitSolEin = shift; # sol-einheit
|
||||
my $LimitIstVar = shift; # ist-filename
|
||||
my $LimitIstEin = shift; # ist-einheit
|
||||
### for better ...
|
||||
$LimitSolVar = "" if (!$LimitSolVar);
|
||||
$LimitIstVar = "" if (!$LimitIstVar);
|
||||
$LimitSolEin = "" if (!$LimitSolEin);
|
||||
|
|
@ -826,7 +857,7 @@ sub check_limit {
|
|||
### presently results
|
||||
print_screen "-> $LimitText: ", 'bold';
|
||||
if ( ($LimitSolVar) && !($LimitIstVar) ) {
|
||||
# only sol & lamport_clock
|
||||
### only sol & lamport_clock
|
||||
if ( $LimitSolVar eq "lamport_clock" ) {
|
||||
my $C_Time = $mars_limit_sol;
|
||||
$C_Time =~ s/CURRENT_TIME=//;
|
||||
|
|
@ -835,7 +866,7 @@ sub check_limit {
|
|||
$L_Time =~ s/.*lamport_now=//;
|
||||
$mars_limit_sol = sprintf("%.2f", $C_Time - $L_Time);
|
||||
print_screen "$mars_limit_sol $LimitSolEin\n";
|
||||
# only sol
|
||||
### only sol
|
||||
} elsif ( $mars_limit_sol < 1 ) {
|
||||
print_screen "is now unsed\n";
|
||||
} else {
|
||||
|
|
@ -843,7 +874,7 @@ sub check_limit {
|
|||
print_screen "$mars_limit_sol $LimitSolEin\n", 'red';
|
||||
}
|
||||
} elsif ( !($LimitSolVar) && ($LimitIstVar) ) {
|
||||
# only ist
|
||||
### only ist
|
||||
if ( $mars_limit_ist < 1 ) {
|
||||
print_screen "is actualy null\n";
|
||||
} else {
|
||||
|
|
@ -851,10 +882,10 @@ sub check_limit {
|
|||
print_screen "$mars_limit_ist $LimitIstEin\n", 'red';
|
||||
}
|
||||
} elsif ( ($LimitSolVar) && ($LimitIstVar) && ($mars_limit_sol < 1) ) {
|
||||
# sol & ist = 0
|
||||
### sol & ist = 0
|
||||
print_screen "is actualy unused\n";
|
||||
} else {
|
||||
# sol & ist
|
||||
### sol & ist / rest ...
|
||||
print_screen "is set to ";
|
||||
print_screen "$mars_limit_sol $LimitSolEin", 'red';
|
||||
print_screen ", actualy used ";
|
||||
|
|
@ -863,7 +894,7 @@ sub check_limit {
|
|||
|
||||
}
|
||||
|
||||
#########################################################################################
|
||||
##############################################################################
|
||||
### main loop ...
|
||||
while(1) {
|
||||
my $dateFormat = Date::Language->new('English');
|
||||
|
|
@ -874,74 +905,79 @@ while(1) {
|
|||
exit 0;
|
||||
}
|
||||
|
||||
#########################################################################################
|
||||
##########################################################################
|
||||
### main run
|
||||
|
||||
print $clearscreen;
|
||||
### check !
|
||||
# print "\nNOTE !!!\n********\nThe author does not guarantee this development-test-alpha-pre-beta-version, it is untested and certainly not fully functional. Use at your own risk ;)\n\n";
|
||||
|
||||
### check and set monitoring default's
|
||||
|
||||
##########################################################################
|
||||
### check and set monitor
|
||||
|
||||
### big monitor
|
||||
if ( $params->{'monitor'} || $params->{'cstate'} || $params->{'dstate'} || $params->{'role'} ) {
|
||||
$params->{'system'} = 1;
|
||||
$params->{'history'} = 1;
|
||||
### check!
|
||||
### TODO: check!
|
||||
$params->{'debug'} = 0;
|
||||
$params->{'monitor'} = 1;
|
||||
}
|
||||
### small-monitor
|
||||
if (( $params->{'cstate'} || $params->{'dstate'} || $params->{'role'} ) && ( !$params->{'resource'})) {
|
||||
display_help "Syntax-Error: Option resource is missing by --cstate / --dstate / --role!";
|
||||
}
|
||||
|
||||
|
||||
##########################################################################
|
||||
### read mars infos
|
||||
# read mars info
|
||||
info_version;
|
||||
|
||||
##########################################################################
|
||||
### check system limits
|
||||
if ( $params->{'system'} ) {
|
||||
# check system limits
|
||||
# text / sol-file / sol-einheit / ist-file / ist-einheit
|
||||
### text / sol-file / sol-einheit / ist-file / ist-einheit
|
||||
my $mars_disk_space = `df '$mars_dir' | grep '$mars_dir'| awk '{print \$2}'`;
|
||||
$mars_disk_space = sprintf("%01.2f", $mars_disk_space / 1024);
|
||||
|
||||
check_limit "AVG-Limit", "loadavg_limit", "loadavg";
|
||||
check_limit "Memory-Limit", "mem_limit_percent", "%", "mem_used_raw_kb", "kb";
|
||||
check_limit "Network-IO-Timeout", "network_io_timeout", "sec";
|
||||
|
||||
check_limit "Traffic Limit", "tuning/traffic_limit_kb", "kb/s", "tuning/traffic_rate_kb", "kb/s";
|
||||
check_limit "Server-IO Limit", "tuning/server_io_limit_kb", "kb/s", "tuning/server_io_rate_kb", "kb/s";
|
||||
|
||||
check_limit "Delay say Overflow", "", "", "delay_say_on_overflow", "(on/off)";
|
||||
check_limit "Statusfile Rollover", "", "", "statusfiles_rollover_sec", "sec";
|
||||
check_limit "Flying IO Count", "", "", "io_flying_count";
|
||||
|
||||
check_limit "LoggerMemory", "", "", "logger_mem_used_kb", "kb";
|
||||
check_limit "FreeSpaceLimit on /mars", "free_space_mb", "mb", "", "$mars_disk_space";
|
||||
check_limit "FreeSpaceLimit LogDelete", "logdel_auto_gb", "gb";
|
||||
check_limit "FreeSpaceLimit LogRotate", "logrot_auto_gb", "gb";
|
||||
|
||||
check_limit "LamportClockDifferenz", "lamport_clock", "sec";
|
||||
# /proc/sys/mars/syslog*
|
||||
# /proc/sys/mars/tuning/aio*
|
||||
# /proc/sys/mars/tuning/bio*
|
||||
# /proc/sys/mars/tuning/wirteback*
|
||||
|
||||
# check system params
|
||||
### check system params
|
||||
check_jammed;
|
||||
check_disk_is_full;
|
||||
}
|
||||
|
||||
|
||||
##########################################################################
|
||||
### check resources
|
||||
check_ressource;
|
||||
|
||||
|
||||
##########################################################################
|
||||
### check global debug
|
||||
if ($params->{'debug'}) {
|
||||
my $debug_res = check_debugfile("", $params->{'debug'});
|
||||
print_screen "-> Main-Debug:\n", 'red'; print_screen "$debug_res" if ( $debug_res );
|
||||
print_screen "-> Main-Debug:\n", 'red';
|
||||
my $debug_res;
|
||||
$debug_res = check_debugfile("", "2.warn"); print_screen "$debug_res" if ( $debug_res );
|
||||
$debug_res = check_debugfile("", "3.error"); print_screen "$debug_res" if ( $debug_res );
|
||||
$debug_res = check_debugfile("", "4.fatal"); print_screen "$debug_res" if ( $debug_res );
|
||||
}
|
||||
|
||||
|
||||
##########################################################################
|
||||
### end, exit for monitor
|
||||
if ( $params->{'monitor'} ) {
|
||||
if (( $StatusCode eq 'InvaliDate' || $StatusCode eq 'Failed' || $StatusCode eq 'OutDate' || $StatusCode eq 'SwitchOff' )) {
|
||||
|
|
@ -952,6 +988,7 @@ while(1) {
|
|||
}
|
||||
|
||||
|
||||
##########################################################################
|
||||
### end, next loop
|
||||
print color 'reset';
|
||||
exit if (not $params->{'interval'});
|
||||
|
|
|
|||
Loading…
Reference in New Issue