OpenTelemetry GPU Collector 0.0.8
Features
- Add environment variable OTEL_GPU_FS_TYPES_EXCLUDE to exclude specified filesystem types from system metrics, defaulting to skip image-based and optical filesystems. (#1395)
- Rename GPU metric names (e.g., hw.gpu.up to hw.status), add collector modes (all/light/serving/training/deep) to enable/disable feature sets, and improve resilience (missing libs/caps soft-skip, config faults stop startup). (#1486)
Fixes
- Fix CUDA eBPF launch argument decoding, add CUDA 13 kernel handle resolution, and use stable 'unknown' for unresolved kernel names to prevent metric cardinality issues. (#1423)
- Added PT_REGS_PARM6 macro for x86-64 to handle 6th argument in BPF programs when libbpf lacks definition. (#1489)
Contributors
@amanagarwal042, @cbirkhold, @cherrycove