Expand description
airun-collector: Production Rust Real-Time Data Plane DaemonSet for NVIDIA DCGM & Kubernetes.
Provides high-throughput, low-latency node telemetry collection, in-kernel eBPF network fabric tracing, OpenTelemetry span correlation, tri-state circuit breaking, and distributed event broadcasting for AI infrastructure.
Re-exports§
pub use circuit_breaker::BreakerConfig;pub use circuit_breaker::BreakerState;pub use circuit_breaker::CircuitBreaker;pub use dcgm::Dcgmsample;pub use dcgm::DcgmScraper;pub use dcgm::HardwareMode;pub use dcgm::TelemetryBatch;pub use dcgm::TelemetryRingBuffer;pub use ebpf::EbpfFabricProbe;pub use ebpf::EbpfMode;pub use ebpf::FabricDropSample;pub use ebpf::FabricHealthReport;pub use otlp::OtlpSpan;pub use otlp::SpanCorrelator;pub use prometheus::format_prometheus_metrics;pub use pubsub::AirunEventEnvelope;pub use pubsub::AirunEventType;pub use pubsub::GpuAlertPayload;pub use pubsub::PubSubPublisher;pub use pubsub::WorkloadLifecyclePayload;
Modules§
- circuit_
breaker - High-throughput Tri-State Circuit Breaker State Machine Runtime.
- dcgm
- NVIDIA DCGM & Hardware Telemetry Engine with Ring Buffer & Graceful Degradation.
- ebpf
- In-Kernel eBPF Fabric Tracing & InfiniBand/RoCE Network Telemetry.
- otlp
- OTLP Trace Ingestion and Time-Window Span Correlation in Rust Data Plane.
- prometheus
- Prometheus / OpenMetrics metrics exporter for the airun Rust Data Plane.
- pubsub
- Pub/Sub streaming publisher and event envelope for the airun distributed event backbone.