oar-server 2.5.7-3 / usr / lib / oar / oarmonitor

#!/usr/bin/perl
# $Id: oarmonitor 598 2007-07-05 08:13:30Z neyron $
# Feed the monitoring tables

use strict;
use warnings;
use Data::Dumper;
use OAR::Conf qw(init_conf dump_conf get_conf is_conf);
use Getopt::Long;
use OAR::Version;
use OAR::Tools;
use OAR::IO;
use Fcntl;
use Storable qw(dclone);
use Sys::Hostname;

my $Old_umask = sprintf("%lo",umask());
umask(oct("022"));

init_conf($ENV{OARCONFFILE});
my $Monitor_file = get_conf("OARMONITOR_SENSOR_FILE");
$Monitor_file = OAR::Tools::get_default_monitor_sensor_file() if (!defined($Monitor_file));
$Monitor_file = "$ENV{OARDIR}/$Monitor_file" if ($Monitor_file !~ /^\//);

my $Taktuk_cmd = get_conf("TAKTUK_CMD");
if (!defined($Taktuk_cmd)){
    warn("ERROR: TAKTUK_CMD must be defined in $ENV{OARCONFFILE}.\n");
    exit(1);
}

my $Openssh_cmd = get_conf("OPENSSH_CMD");
$Openssh_cmd = OAR::Tools::get_default_openssh_cmd() if (!defined($Openssh_cmd));

my $Cpuset_path = get_conf("CPUSET_PATH");
if (!defined($Cpuset_path) or (! is_conf("JOB_RESOURCE_MANAGER_PROPERTY_DB_FIELD"))){
    warn("The CPUSET feature must be configured to use this command\n");
    exit(6);
}

Getopt::Long::Configure("gnu_getopt");
my $Version;
my $Job_id;
my $Frequency = 60;
my $Default_frequency = $Frequency;
my $Frequency_min = 1;
my $Timeout = 120;
my $sos;
my $Verbose;
GetOptions ("version|V" => \$Version,
            "job_id|j=i" => \$Job_id,
            "frequency|f=i" => \$Frequency,
            "help|h" => \$sos,
            "verbose|v" => \$Verbose
           ) or exit(1);

# Display command help
sub usage {
    print <<EOS;
Usage: $0 [-h] [-V] [-f seconds] -j jobid
Feed the table monitoring from the database with data retrived on each nodes
of the job.
Options:
  -j, --job_id              job id to monitor
  -f, --frequency           number of seconds between each data collect
                            (default is $Default_frequency s)
  -h, --help                show this help screen
  -V, --version             print OAR version number
EOS
}

$Frequency = $Frequency_min if ($Frequency < $Frequency_min);

if (defined($sos)){
    usage();
    exit(0);
}

if (defined($Version)){
    print("OAR version : ".OAR::Version::get_version()."\n");
    exit(0);
}

if (!defined($Job_id)){
    usage();

exit(2);
}

my $base = OAR::IO::connect();

my $cpuset_name = OAR::IO::get_job_cpuset_name($base, $Job_id);
my @hosts = OAR::IO::get_job_current_hostnames($base,$Job_id);
if ($#hosts < 0){
    warn("The job $Job_id does not have currently assigned hostnames\n");
    exit(4);
}

#############################
# TAKTUK process management #
###############################################################################
pipe(tak_node_read,tak_node_write);
pipe(tak_stdin_read,tak_stdin_write);
pipe(tak_stdout_read,tak_stdout_write);
my $pid = fork;
if($pid == 0){
    #CHILD
    undef($base);
    $SIG{CHLD} = 'DEFAULT';
    $SIG{TERM} = 'DEFAULT';
    $SIG{INT}  = 'DEFAULT';
    $SIG{QUIT} = 'DEFAULT';
    $SIG{USR1} = 'DEFAULT';
    $SIG{USR2} = 'DEFAULT';
    my $cmd = "$Taktuk_cmd -c '$Openssh_cmd' ".'-o error=\'"INFO on $host: $line\n"\' -o output=\'"OUTPUT $host $line\n"\''." -f '<&=".fileno(tak_node_read)."' broadcast exec [ perl - $Job_id $Cpuset_path/$cpuset_name ], broadcast input file [ - ], broadcast input close";
    fcntl(tak_node_read, F_SETFD, 0);
    close(tak_node_write);
    close(tak_stdout_read);
    close(STDOUT);
    # Redirect taktuk output into the pipe
    open(STDOUT, ">& tak_stdout_write");

    # Use the child STDIN to send the user command
    close(tak_stdin_write);
    close(STDIN);
    open(STDIN, "<& tak_stdin_read");

    exec($cmd);
    warn("[ERROR] Cannot execute $cmd\n");
    exit(-1);
}
close(tak_node_read);
close(tak_stdin_read);
close(tak_stdout_write);

# Send node list
my $hosts_hash;
foreach my $n (@hosts){
    $hosts_hash->{$n} = 1;
    print(tak_node_write "$n\n");
}
close(tak_node_write);

my $oldfh;
$oldfh = select(tak_stdin_write); $| = 1; select($oldfh);
$oldfh = select(tak_stdout_read); $| = 1; select($oldfh);

###############################################################################

# Send sensor perl script to all nodes
if (open(SENSOR, $Monitor_file)){
    while(<SENSOR>){
        print(tak_stdin_write $_);
    }
    close(SENSOR);
    print(tak_stdin_write "\n__END__\n");
}else{
    warn("Cannot read the monitor sensor file $Monitor_file\n");
    exit(3);
}

# Signals management
sub exit_monitor(){
    print(tak_stdin_write "STOP\n");
}

$SIG{TERM} = \&exit_monitor;
$SIG{INT} = \&exit_monitor;
$SIG{QUIT} = \&exit_monitor;
$SIG{USR2} = sub {return};

# Node outputs parsing
my $monitor_process_id = $Job_id.'_'.hostname().'_'.time().'_'.$$;
my $tmp_hosts_hash;
my $nb_end_nodes = $#hosts + 1;
my $stop = 0;
my $tic_time = 0;
my $tic_time_prev = $tic_time;
while (($stop == 0) and ($nb_end_nodes > $#hosts)){
    my $sleep_time = $Frequency - (OAR::IO::get_date($base) - $tic_time);
    if ($sleep_time > 0){
        OAR::IO::disconnect($base);
        sleep($sleep_time);
        $base = OAR::IO::connect();
    }
    $tic_time_prev = $tic_time;
    $tic_time = OAR::IO::get_date($base); $tic_time = $tic_time_prev if ($tic_time < $tic_time_prev);
    eval{
        $SIG{ALRM} = sub { die "alarm\n" };
        alarm($Frequency + $Timeout);
        
        # Ask every nodes to print their monitoring values
        print(tak_stdin_write "monitor_process_id=$monitor_process_id window_start=$tic_time_prev window_stop=$tic_time\n");

        # Check the taktuk STDOUT
        $tmp_hosts_hash = dclone($hosts_hash);
        $nb_end_nodes = 0;
        while (($nb_end_nodes <= $#hosts) and ($stop == 0) and ($_ = <tak_stdout_read>)){
            chop;
            if ($_ =~ /^OUTPUT\s+([\w\.\-\d]+)\s+END$/){
                delete($tmp_hosts_hash->{$1}) if (defined($tmp_hosts_hash->{$1}));
                $nb_end_nodes++;
            }elsif ($_ =~ /^OUTPUT\s+([\w\.\-\d]+)\s+(\w+)\s*(.*)$/){
                if (($2 eq "STOP") or ($2 eq "STOP_REQUESTED") or ($2 eq "ERROR")){
                    warn("    $1 $2\n");
                    $stop++;
                }else{
                    # Sensor gives the DB table name ($2) and the values for
                    # each fields separated by spaces ($3)
                    if ($tic_time_prev > 0){
                        my @fields;
                        my @values;
                        my @tmp = split(" ",$3);
                        foreach my $i (@tmp){
                            my ($field,$value) = split('=',$i);
                            if ((defined($field)) and (defined($value))){
                                push(@fields, $field);
                                push(@values, $value);
                            }
                        }
                        if (defined($Verbose)){
                            print("@fields\n");
                            print("@values\n");
                        }
                        if (OAR::IO::register_monitoring_values($base,$2,\@fields,\@values) > 0){
                            warn("[ERROR] Cannot register in the database: @fields, @values\n");
                            $stop++;
                        }
                    }
                }
            }else{
                warn("[TAKTUK] $_\n");
            }
        }
        alarm(0);
    };
}
close(tak_stdin_write);
close(tak_stdout_read);

if ($stop > 0){
    warn("At least one sensor stopped to retrieve data.\n");
    exit(0);
}

my @bads = keys(%{$tmp_hosts_hash});
if ($#bads >= 0){
    warn("Some nodes timeouted: @bads\n");
    exit(5);
}