#[cfg(feature = "gpu")] use crate::GlobalError; #[cfg(feature = "gpu")] use crate::system::attributes::ProcessAttributes; #[cfg(feature = "gpu")] use crate::system::metrics::Metrics; #[cfg(feature = "gpu")] use nvml_wrapper::Nvml; #[cfg(feature = "gpu")] use nvml_wrapper::enums::device::UsedGpuMemory; #[cfg(feature = "gpu")] use sysinfo::Pid; #[cfg(feature = "gpu")] use tracing::warn; /// `GpuCollector` is responsible for collecting GPU memory usage metrics. #[cfg(feature = "gpu")] pub struct GpuCollector { nvml: Nvml, pid: Pid, } #[cfg(feature = "gpu")] impl GpuCollector { pub fn new(pid: Pid) -> Result { let nvml = Nvml::init().map_err(|e| GlobalError::GpuInitError(e.to_string()))?; Ok(GpuCollector { nvml, pid }) } pub fn collect(&self, metrics: &Metrics, attributes: &ProcessAttributes) -> Result<(), GlobalError> { if let Ok(device) = self.nvml.device_by_index(0) { if let Ok(gpu_stats) = device.running_compute_processes() { for stat in gpu_stats.iter() { if stat.pid == self.pid.as_u32() { let memory_used = match stat.used_gpu_memory { UsedGpuMemory::Used(bytes) => bytes, UsedGpuMemory::Unavailable => 0, }; metrics.gpu_memory_usage.record(memory_used, &attributes.attributes); return Ok(()); } } } else { warn!("Could not get GPU stats, recording 0 for GPU memory usage"); } } else { return Err(GlobalError::GpuDeviceError("No GPU device found".to_string())); } metrics.gpu_memory_usage.record(0, &attributes.attributes); Ok(()) } } #[cfg(not(feature = "gpu"))] pub struct GpuCollector; #[cfg(not(feature = "gpu"))] impl GpuCollector { pub fn new(_pid: sysinfo::Pid) -> Result { Ok(GpuCollector) } pub fn collect( &self, _metrics: &crate::system::metrics::Metrics, _attributes: &crate::system::attributes::ProcessAttributes, ) -> Result<(), crate::GlobalError> { Ok(()) } }