問題文
A platform team must expose accelerators to containers in a container orchestration cluster, keep the drivers consistent across nodes, and label nodes with the accelerator model automatically. They want to avoid doing these steps by hand on each node. Which approach fits?
選択肢
- Install the driver manually on each node once, and let the scheduler discover the devices on its own and apply the labels from what it finds.
- Use the cluster workload manager instead, which already knows about the accelerators.
- Run every workload with elevated host privileges so that containers can reach the devices directly.
- Deploy an operator that automates the driver, the device plugin, the container toolkit, node labeling, and monitoring together.