扩展 kubelet 的 Pod 资源监控 gRPC API 端点 List 和 Get, 以包括通过资源分配报告在 ResourceClaim 中分配的资源。
以下示例展示了 NVIDIA dcgm-exporter 如何使用此 API 来收集 NVIDIA DRA driver 分配的每个 Pod 的 GPU 指标:
DCGM_FI_PROF_PCIE_RX_BYTES{gpu="0",UUID="GPU-a4f34abc-7715-3560-dcea-7238b9611a45",pci_bus_id="00000009:01:00.0",device="nvidia0",modelName="NVIDIA GH200 96GB HBM3",Hostname="sc-starwars-xxxx",container="ctr",dra_claim_name="single-gpu",dra_claim_namespace="gpu-test3",dra_device_name="gpu-0",dra_driver_name="gpu.nvidia.com",dra_pool_name="sc-starwars-xxxx",namespace="gpu-test3",pod="pod1"} 23792
DCGM_FI_PROF_PCIE_RX_BYTES{gpu="0",UUID="GPU-a4f34abc-7715-3560-dcea-7238b9611a45",pci_bus_id="00000009:01:00.0",device="nvidia0",modelName="NVIDIA GH200 96GB HBM3",Hostname="sc-starwars-xxxx",container="ctr",dra_claim_name="single-gpu",dra_claim_namespace="gpu-test3",dra_device_name="gpu-0",dra_driver_name="gpu.nvidia.com",dra_pool_name="sc-starwars-xxxx",namespace="gpu-test3",pod="pod2"} 23792
包含 Pod DRA 信息:
container="ctr",
dra_claim_name="single-gpu",
dra_claim_namespace="gpu-test3",
dra_device_name="gpu-0",dra_driver_name="gpu.nvidia.com",
dra_pool_name="sc-starwars-xxxx",
namespace="gpu-test3",
pod="pod1"
来源:Kubernetes 文档贡献者 · 作者与贡献记录 · CC BY 4.0。离线整理了版式、链接和媒体显示。
