add full debug to intel device

This commit is contained in:
Meliox 2026-06-13 17:56:23 +02:00
parent 30d99079a7
commit bfce5e7f0c
2 changed files with 98 additions and 44 deletions

View File

@ -20,6 +20,23 @@ our @EXPORT_OK = qw(
sub get_intel_gpu_devices {
my @devices = ();
if ($config{debug}{intel_mode} && -f $config{debug}{intel_devices_file}) {
debug(__LINE__, "Debug mode: reading Intel GPU devices from $config{debug}{intel_devices_file}");
my $data = safe_read_json($config{debug}{intel_devices_file});
if ($data && ref $data eq 'ARRAY') {
for my $dev (@$data) {
push @devices, {
card => $dev->{card},
name => $dev->{name},
path => $dev->{path},
drm_path => $dev->{drm_path},
};
debug(__LINE__, "Found Intel device (debug): $dev->{card} -> $dev->{name} ($dev->{path})");
}
} else {
debug(__LINE__, "Failed to parse debug file $config{debug}{intel_devices_file}");
}
} else {
debug(__LINE__, "Getting Intel GPU devices");
if (open my $fh, '-|', 'intel_gpu_top -L') {
while (<$fh>) {
@ -41,6 +58,7 @@ sub get_intel_gpu_devices {
} else {
debug(__LINE__, "Failed to run intel_gpu_top -L: $!");
}
}
return @devices;
}
@ -132,6 +150,41 @@ sub collector_for_intel_device {
if defined $intel_gpu_top_pid && $intel_gpu_top_pid > 0;
});
my $node_name = "node0";
if ($config{debug}{intel_mode} && -f $config{debug}{intel_output_file}) {
debug(__LINE__, "Debug mode: reading Intel GPU stats from $config{debug}{intel_output_file}");
while (!$shutdown) {
if (open my $fh, '<', $config{debug}{intel_output_file}) {
while (my $line = <$fh>) {
chomp $line;
next if $line =~ /MHz|IRQ|RC6|Power|RCS|BCS|VCS|VECS|req\s+act|^\s*$/;
if ($line =~ /^\s*[\d\s\.]+$/) {
my $stats = _parse_intel_gpu_line($line);
if ($stats) {
my $device_data = {
$node_name => {
name => $device->{name},
device_path => $device->{path},
drm_path => $device->{drm_path},
stats => $stats,
}
};
safe_write_json($device_state_file, $device_data);
update_intel_gpu_rrd($device->{card}, $stats);
}
}
}
close $fh;
} else {
debug(__LINE__, "Failed to open debug file $config{debug}{intel_output_file}: $!");
}
sleep $config{intervals}{data_pull} unless $shutdown;
}
} else {
debug(__LINE__, "About to open pipe to intel_gpu_top");
my $intel_pull_interval = $config{intervals}{data_pull} * 1000; # milliseconds
$intel_gpu_top_pid = open(my $fh, '-|',
@ -144,8 +197,6 @@ sub collector_for_intel_device {
debug(__LINE__, "Pipe opened successfully, PID=$intel_gpu_top_pid");
my $node_name = "node0";
while (my $line = <$fh>) {
last if $shutdown;
chomp $line;
@ -172,6 +223,8 @@ sub collector_for_intel_device {
}
close $fh;
}
debug(__LINE__, "Collector for $device->{card} shutting down");
exit 0;
}

View File

@ -42,6 +42,7 @@ our %config = (
lm_sensors_output_file => '/tmp/sensors-output.json',
intel_mode => 0,
intel_devices_file => '/tmp/intel-gpu-devices.json',
intel_output_file => '/tmp/intel-gpu-output.json',
nvidia_mode => 0,
nvidia_devices_file => '/tmp/nvidia-smi-devices.csv',
nvidia_output_file => '/tmp/nvidia-smi-output.csv',