An emergency-management Kubernetes cluster will launch fabric-attached training jobs. What should operators sequence first?
Select an answer to reveal the explanation.
Short Explanation
GPU Operator alone or a bridge CNI won't give those jobs the fabric attachments they need. Roll out the NVIDIA Network Operator first, so you don't watch RDMA-dependent training fail at its first launch.
Full Explanation
An emergency-management Kubernetes cluster preparing fabric-attached training jobs should roll out the NVIDIA Network Operator before launching RDMA-dependent civic training jobs. Sequencing the operator first ensures device plugins, RDMA networking, and secondary attachments exist when Jobs start. Emergency-management model training that depends on InfiniBand collectives will otherwise fail at first run, consuming scarce incident-response compute windows.
Deploying only the NVIDIA GPU Operator to prepare nodes for RDMA-dependent civic training jobs omits the fabric networking layer the Network Operator provides, so Jobs still lack RDMA attachments and fail at first run. Installing Multus with standard bridge CNI plugins for RDMA-dependent civic training jobs fails conceptually: standard bridge CNI does not provide RDMA-capable fabric attachments or the device management the Network Operator exposes for InfiniBand training. Enabling the Node Feature Discovery Operator to label fabric adapters for civic training jobs only advertises hardware; it does not install the RDMA device plugins, secondary network attachments, or fabric networking stack the Network Operator provides, so Jobs still cannot use the fabric.
Exam caveat: Before launching means operator health and RDMA exposure are verified, not merely that a Helm install command was typed. A CrashLooping operator does not satisfy the sequencing requirement.
Operational check: In the emergency-management cluster, deploy and verify the Network Operator, confirm RDMA resources appear on GPU nodes, then submit a small fabric-attached training Job. Document the sequence in the runbook so future civic training campaigns do not start Jobs ahead of operator rollout.