6 use Time::HiRes qw(time sleep);
8 my $DEBUG = $ENV{DEBUG} || 0;
9 my $INFLUX = $ENV{INFLUX} || 'http://10.80.3.89:8086/write?consistency=any&db=gnt';
10 my $INTERVAL = $ENV{INTERVAL} || 1;
12 my $RACK = $ENV{RACK};
14 my $node = `hostname -s`;
18 $location .= qq{,dc=$DC} if $DC;
19 $location .= qq{,rack=$RACK} if $RACK;
21 warn $0 = "gnt-monitor\@$node$location $INFLUX $INTERVAL $DEBUG";
24 sub XXX { $DEBUG ? warn "XXX ",Data::Dump::dump( @_ ) : {} };
34 my $t = int( time() * 1000_000_000 );
36 foreach my $instance ( glob '/var/run/ganeti/kvm-hypervisor/pid/*' ) {
38 open(my $fh, '<', $instance);
39 my $pid = <$fh>; chomp $pid;
41 $instance =~ s{^.*/}{};
43 if ( ! -d "/proc/$pid" ) {
49 my $vcpu = $last->{$instance}->{vcpu};
52 foreach my $fd ( glob "/proc/$pid/fd/*" ) {
53 $vcpu++ if -l $fd && readlink($fd) =~ m/kvm-vcpu/;
56 $last->{$instance}->{vcpu} = $vcpu;
59 # https://www.kernel.org/doc/Documentation/filesystems/proc.txt
60 open($fh, '<', "/proc/$pid/stat");
61 my $line = <$fh>; chomp $line;
62 my $gtime = (split(/\s+/,$line))[42]; # guest time of the task in jiffies
64 if ( my $last_gtime = $last->{$instance}->{gtime} ) {
65 my $clock_ticks = POSIX::sysconf( &POSIX::_SC_CLK_TCK ); # clock ticks per second
67 my $cpu = ( ( $gtime - $last_gtime ) * 100 ) / ( $clock_ticks * $vcpu ) / $INTERVAL;
68 $stat->{$instance}->{cpu} = $cpu;
69 $stat->{$instance}->{ticks} = ($gtime - $last_gtime) / $INTERVAL;
72 $last->{$instance}->{gtime} = $gtime;
76 foreach my $glob ( glob '/var/run/ganeti/instance-disks/*' ) {
77 my ( $instance, $disk ) = split(/:/,$glob,2);
78 $instance =~ s{^.*/}{};
80 next unless exists $stat->{$instance};
82 my $dev = readlink $glob;
85 if ( ! -e "/sys/class/block/$dev" ) {
91 open( my $fh, '<', "/sys/class/block/$dev/stat" );
92 my $v = <$fh>; chomp $v; $v =~ s/^\s+//;
93 my @s = split(/\s+/, $v );
94 # https://www.kernel.org/doc/Documentation/block/stat.txt
96 read_io => $s[0] / $INTERVAL,
97 read_bytes => $s[2] * 512 / $INTERVAL,
98 read_wait => $s[3] / $INTERVAL,
99 write_io => $s[4] / $INTERVAL,
100 write_bytes => $s[6] * 512 / $INTERVAL,
101 write_wait => $s[7] / $INTERVAL,
103 if ( my $l = $last->{$instance}->{disk}->[$disk] ) {
105 $delta->{$_} = $d->{$_} - $l->{$_} foreach keys %$d;
106 $stat->{$instance}->{disk}->[$disk] = $delta;
107 $stat->{$instance}->{disk}->[$disk]->{dev} = $dev;
108 $stat->{$instance}->{disk}->[$disk]->{disk} = $disk;
110 $last->{$instance}->{disk}->[$disk] = $d;
114 foreach my $full_instance ( glob '/var/run/ganeti/kvm-hypervisor/nic/*' ) {
115 my $instance = $full_instance;
116 $instance =~ s{^.*/}{};
117 next unless exists $stat->{$instance};
119 foreach my $nic ( glob "$full_instance/*" ) {
120 open(my $fh, '<', $nic);
123 next unless -e "/sys/class/net/$dev";
129 foreach my $f (qw( rx_bytes tx_bytes rx_packets tx_packets )) {
130 open( my $fh, '<', "/sys/class/net/$dev/statistics/$f" );
131 my $v = <$fh>; chomp $v;
132 $d->{$f} = $v / $INTERVAL;
134 if ( my $l = $last->{$instance}->{nic}->[$nic] ) {
135 $stat->{$instance}->{nic}->[$nic]->{$_} = $d->{$_} - $l->{$_} foreach keys %$d;
136 $stat->{$instance}->{nic}->[$nic]->{dev} = $dev;
137 if ( -e "/sys/class/net/$dev/master" ) {
138 my $vlan = readlink "/sys/class/net/$dev/master";
140 $stat->{$instance}->{nic}->[$nic]->{vlan} = $vlan;
143 $last->{$instance}->{nic}->[$nic] = $d;
154 foreach my $k ( keys %$hash ) {
156 my ( $d, $s ) = $v =~ m/^\d+(\.\d+)?$/i ? ( '', '' ) : # float
157 ( '"' ,'' ) ; # string
159 push @v, "$k=$d$v$d$s";
161 my $i = join(',', @v);
165 open(my $fh, '>', '/dev/shm/gnt-monitor.influx');
167 foreach my $instance ( keys %$stat ) {
169 next if $skip->{$instance};
171 print $fh qq{ganeti_cpu,node=$node,instance=$instance$location cpu=$stat->{$instance}->{cpu},ticks=$stat->{$instance}->{ticks} $t\n};
173 foreach my $disk ( @{ $stat->{$instance}->{disk} } ) {
174 print $fh qq{ganeti_disk,node=$node,instance=$instance$location },dump4influx( $disk ), " $t\n";
177 foreach my $nic ( @{ $stat->{$instance}->{nic} } ) {
178 my $vlan = delete $nic->{vlan};
179 $vlan = 0 unless defined $vlan;
180 print $fh qq{ganeti_nic,node=$node,instance=$instance,vlan=${vlan}$location },dump4influx( $nic ), " $t\n";
187 if ( system( 'curl', '-s', '-XPOST', $INFLUX, '--data-binary', '@/dev/shm/gnt-monitor.influx' ) == 0 ) {
189 rename '/dev/shm/gnt-monitor.influx', '/dev/shm/gnt-monitor.influx.bug.' . scalar glob '/dev/shm/gnt-monitor.influx.bug.*';
190 warn "curl failed: $? $!";
193 my $dt = time() - $begin_t;
194 sleep $INTERVAL - $dt if $dt > 0 && $dt < $INTERVAL;