Oke Multi Home Pods

Workflow

1. Confirm the target cluster explicitly. Do not run discovery against an implicit or default cluster.

2. If the user wants discovery or does not know cluster metadata, run:

bash scripts/scaffold_workspace.sh
python3 scripts/discover_customer_inputs.py --cluster <cluster-name-or-ocid>

This creates customer-input-template.yaml in the current working directory if it is missing, then writes customer-input.discovered.yaml with discovered cluster metadata and cached subnet/NSG choices.

3. Check whether Multus is already deployed on the target cluster context:

bash scripts/check_multus.sh --context <kubectl-context>

If multus_installed is false, stop and ask the user whether they want to install Multus before continuing.

• Do not assume Multus is present because the guide requires it.
• Do not proceed to final multi-home pod manifests until the user either confirms Multus is installed or asks you to install it.

4. Use the x_codex_prompting block in the customer input YAML to ask for missing values.

• Ask in small batches.
• Use encoded options where they exist.
• Use cached subnet and NSG discovery data to present numbered choices.
• Do not guess interface names; inspect them on a real worker node and then ask the user to confirm.
• Respect deployment.mode. If it is existing-gva-node-pool, do not require create-only node pool fields and do not render nodepool.json. If it is create-gva-node-pool, require the node pool creation inputs and render nodepool.json.

5. When required values are complete, render the artifacts:

python3 scripts/render_artifacts.py --input customer-input.discovered.yaml

If discovery was not used:

python3 scripts/render_artifacts.py --input customer-input-template.yaml

6. Before handoff, run strict validation:

python3 scripts/render_artifacts.py --input <input-yaml> --strict

Expected Outputs

The renderer writes rendered/ in the current working directory:

• nodepool.json
• cloud-init.sh
• oci-vcn-native-network-nad.yaml
• ipvlan-network.yaml
• sleep-forever-pod.yaml
• verification.md

Operating Rules

• Never invent OCIDs, subnet IDs, NSG IDs, interface names, or image IDs.
• Never auto-select a cluster without explicit user input in the current turn.
• Use discovery to reduce prompting, not to bypass customer confirmation.
• Always check Multus on the explicit target cluster context before assuming the cluster is ready for multi-home pods.
• If Multus is missing, ask the user whether they want to install it before continuing.
• Keep eth0 pinned to the intended default interface.
• Keep net1 pinned to the intended secondary interface.
• Do not silently choose subnets or NSGs from discovery results.
• If the cluster CNI type is not OCI_VCN_IP_NATIVE, stop and explain the mismatch.
• If GVA is not enabled or planned on the node pool, stop and explain that multi-homed pods require the needed secondary VNICs.

Interface Inspection

When interface names are still unknown, prefer kubectl debug node/... against a worker in the target node pool.

Primary command:

kubectl --context <kubectl-context> debug node/<node-name> -it --image=nicolaka/netshoot --profile=sysadmin -- chroot /host ip -br addr

Fallback command:

kubectl --context <kubectl-context> debug node/<node-name> -it --image=nicolaka/netshoot --profile=sysadmin -- chroot /host ifconfig -a

Useful follow-up:

kubectl --context <kubectl-context> debug node/<node-name> -it --image=nicolaka/netshoot --profile=sysadmin -- chroot /host ip route

Use these results to identify:

• the interface carrying the default pod network subnet
• the interface carrying the secondary net1 subnet
• the exact interface names to place into interfaces.default_network.interface_name and interfaces.secondary_network.interface_name

When To Use Which Script

• Use scripts/scaffold_workspace.sh when the working directory does not yet have the template files.
• Use scripts/check_multus.sh immediately after the cluster is identified and before generating final multi-home pod artifacts.
• Use scripts/discover_customer_inputs.py when the user has provided an explicit cluster name, context, or OCID and wants you to auto-populate cluster metadata.
• Use scripts/render_artifacts.py once the customer input YAML is complete enough to render outputs.

Response Shape

Keep operational responses short and ordered like this:

1. Status
2. Missing Inputs or Generated Files
3. Commands to Run
4. Validation
5. Risks