---

name: launch-nemo-rl license: Apache-2.0 description: Playbook for launching, monitoring, stopping, and debugging NeMo-RL recipes on a Kubernetes cluster via the nrl-k8s CLI. Covers ephemeral vs long-lived RayCluster modes, iterating on runs, and debugging hung or failed training jobs. when_to_use:

• "run this recipe on k8s"
• "launch on the cluster"
• "submit a training job"
• "tear down the cluster"
• "resubmit as rayjob"
• "why is the run stuck"
• "how do I get logs for job X"
• "bring the cluster back up" allowed-tools: Bash Read Grep Glob Edit Write

---

launch-nemo-rl — running NeMo-RL recipes on Kubernetes via nrl-k8s

This is the playbook for the nrl-k8s CLI at infra/nrl_k8s/. Follow it when the user asks to launch / iterate / debug a NeMo-RL recipe on a Kubernetes cluster. Verify current state (kubectl, git log, the recipe + infra files) before acting — the cluster is shared and the cost of a wrong action is high.

1. One command, two modes

There is a single top-level submission command: nrl-k8s run. It has two lifecycle modes.

| Mode | Invocation | When to use | Cluster after? | | :----------------- | :---------------- | :---------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | :------------- | | Ephemeral (default) | nrl-k8s run | One-shot. KubeRay applies a RayJob, runs, tears the cluster down. Best for most runs. | No (auto) | | Long-lived | nrl-k8s run --raycluster | Dev loop. Reuses a matching live cluster, applies if absent, warns + reuses on drift (pass --recreate to replace). Then submits daemons and training. First-choice for iteration. | Yes |

Ask: Do I need this cluster after the run? If yes, use --raycluster. Otherwise use the default (ephemeral).

The rest of the CLI is observability / stage-by-stage control:

| Command | Purpose | | :---------------------- | :---------------------------------------------------------------------------------------------- | | nrl-k8s check | Validate a recipe + infra pair; optionally write the fully-resolved manifests (-o). | | nrl-k8s status | Per-role RayCluster state, head pod phase, worker pod phases, daemon job status. | | nrl-k8s cluster up/down/list/dashboard | Manage RayClusters independently of a run (e.g. render a manifest with --dry-run). | | nrl-k8s job list/logs/stop | Observability over Ray Jobs already submitted to a role's cluster. | | nrl-k8s logs | Tail a role's pod / daemon logs without needing a submission id. |

2. Recipe + infra pair

Every launch takes two files. Pass the infra with --infra, not merged inline:

nrl-k8s run infra/nrl_k8s/examples/<recipe>.yaml \
  --infra infra/nrl_k8s/examples/<recipe>.<profile>.infra.yaml

• Recipe (e.g. qwen3_30b_math_8n_4gpu.yaml) — NeMo-RL config: model, GRPO/SFT knobs, cluster.{gpus_per_node,num_nodes}. Uses defaults: to inherit from examples/configs/recipes/llm/....
• Infra (e.g. *.<profile>.infra.yaml) — K8s/Ray shape: namespace, image, service account, RayCluster spec under kuberay:, optional Deployments under deployments:, submit.submitter, launch.{mode,codeSource,codePath,entrypoint}. Pair names follow <recipe>.<profile>[.prod].infra.yaml where <profile> names the hardware target (e.g. gb300).

Example pairs in infra/nrl_k8s/examples/ — read the neighbouring files to see the current conventions for the target profile.

3. Long-lived mode flags

Three independent dimensions. --mode is a macro that picks defaults; individual flags override it.

--mode interactive   → --submitter portForward  --code-source upload  (tails logs)
--mode batch         → --submitter exec         --code-source image   (returns after nohup)

• Submitter: portForward uses kubectl port-forward + Ray Job SDK (gets a submission_id the dashboard tracks). exec uses kubectl exec + nohup on the head pod (no submission_id; driver appears as type=DRIVER in the dashboard).
• Code source: upload stages a working_dir from the laptop (Ray 100 MiB cap). image / lustre expect code on the pod's filesystem — paired with --code-path (typically /opt/nemo-rl), which is a subPath of the shared-filesystem PVC mount in the standard infra examples.
• Wait: --wait tails logs until terminal; --no-wait returns as soon as the driver is running.

Other long-lived-only flags:

• --replace — stop any running training / daemon job before submitting new ones (suffixes daemon submissionIds with a timestamp so Ray accepts the resubmit).
• --recreate — delete + re-apply a RayCluster whose live spec has drifted from the rendered manifest (default is warn + reuse).
• --skip-daemons — bring up all declared clusters but only submit training. Use on disagg recipes where gym/generation are already healthy.

Gotcha: on infra where the entrypoint does cd /opt/nemo-rl (or another in-image / Lustre path) and loads the recipe from there, --code-source upload does NOT override the recipe on the pod — the uploaded working_dir sits in /tmp/ray/... but the entrypoint cds away from it. To actually test a local recipe change, either sync your edits to the shared filesystem mounted into the pods or flip the Hydra overrides in the entrypoint.

4. Ephemeral mode flags (--rayjob)

When --rayjob is set, run branches into the RayJob code path. Relevant flags:

• --rayjob-name NAME — RayJob metadata name (defaults to the training cluster name).
• --shutdown / --no-shutdown — default true: KubeRay deletes the RayCluster once the Ray Job reaches a terminal state.
• --ttl SECONDS — default 3600s: keep the RayJob object around after the run finishes for post-mortem log access.
• --wait / --no-wait — default wait: poll jobDeploymentStatus until Complete/Failed. --no-wait returns as soon as the RayJob is applied.
• --timeout SECONDS — default 86400s (24h): bound the --wait poll.
• --dry-run — render the RayJob manifest and print it; do not apply.

--replace / --recreate / --skip-daemons are silently ignored in --rayjob mode (KubeRay owns lifecycle).

5. Iterating on a config without touching the shared filesystem

When the recipe on the pod filesystem has the wrong value for your experiment, use Hydra overrides on the entrypoint instead of forking the recipe. Pattern:

entrypoint: |
  set -eu
  cd /opt/nemo-rl
  RUN_ID="\${RAY_JOB_SUBMISSION_ID:-\${NRL_K8S_RUN_ID:-$(date -u +%Y%m%d-%H%M%S)}}"
  python -u examples/run_grpo.py \
    --config infra/nrl_k8s/examples/<recipe>.yaml \
    logger.wandb_enabled=true \
    logger.wandb.project=<project> \
    "logger.wandb.name=<run-name>-\${RUN_ID}"

Escape ${…} with a backslash. OmegaConf otherwise interprets it as interpolation and errors on shell-style ${VAR:-default}. RUN_ID resolves to RAY_JOB_SUBMISSION_ID (injected by KubeRay in rayjob mode) → NRL_K8S_RUN_ID (injected by the CLI in long-lived mode) → local timestamp — so the name is unique across either path.

6. Per-profile concerns (hardware + scheduler + DRA)

Every infra YAML encodes a hardware/scheduler profile. The concrete examples in infra/nrl_k8s/examples/ are authoritative for the profiles they target — read the neighbouring infra file before writing a new one. Things that commonly vary:

• Per-node GPUs (e.g. 4 vs 8) — must match cluster.gpus_per_node in the recipe, otherwise workers stay Pending.
• Node selectors — head pods usually land on a CPU-only node pool; GPU workers match on nvidia.com/gpu.product or a node-group label.
• Scheduler — KAI (schedulerName: kai-scheduler + kai.scheduler/queue label) with topology annotations (kai.scheduler/topology, kai.scheduler/topology-required-placement) gang-schedules workers into one clique. Without it, pods may land on different racks and NVLink/RoCE won't span them.
• DRA claims — ComputeDomain + RoCE are attached via resourceClaims referencing ResourceClaimTemplates. The CLI auto-creates/deletes these when the worker pod spec contains DRA claim references — no manual setup needed.
• Secrets — always via secretKeyRef (wandb-api-key, image pull secret). Never embed.
• Shared filesystem mounts — typically a Lustre PVC mounted twice: once at the code path (e.g. /opt/nemo-rl with a user-scoped subPath) and once at a workspace root (e.g. /mnt/rl-workspace) for datasets, HF cache, and checkpoints.

Before applying an infra, verify prereqs exist in the target namespace:

kubectl get pvc <workspace-pvc>
kubectl get secret <wandb-secret> <image-pull-secret>
kubectl get sa <service-account>

7. End-to-end workflows

7a. Fresh one-shot run (rayjob)

# From the NeMo-RL repo root:
nrl-k8s check <recipe> --infra <infra>                               # validate first
nrl-k8s run <recipe> --infra <infra> --rayjob --dry-run              # render RayJob manifest
nrl-k8s run <recipe> --infra <infra> --rayjob --no-wait              # apply, returns fast

Watch status + teardown (works even after your laptop disconnects because KubeRay owns the lifecycle):

kubectl get rayjob -n default <name> -w
kubectl get raycluster -n default                                    # empty = teardown succeeded

7b. Dev loop (long-lived)

nrl-k8s run <recipe> --infra <infra> --run-id $(date +%Y%m%d-%H%M%S)
# Edits in the recipe? Just re-run — reuses the live cluster.
# Pod spec changed? Add --recreate to delete + re-apply.
# Disagg recipe with gym/gen already healthy? --skip-daemons.

7c. First-time disaggregated bring-up

nrl-k8s run <recipe> --infra <disagg-infra> --mode batch --code-source image

7d. Cluster-only lifecycle

nrl-k8s cluster up   <recipe> --infra <infra> --target kuberay.training --wait
nrl-k8s cluster up   <recipe> --infra <infra> --target kuberay.training --dry-run   # render manifest
nrl-k8s cluster down <recipe> --infra <infra> --target kuberay.training --wait
nrl-k8s cluster down <recipe> --infra <infra>                                       # tear down all
nrl-k8s cluster list -n default
nrl-k8s cluster dashboard <cluster-name>                                  # port-forward + browser

7e. Deployments (e.g. nemo-skills sandbox)

# Bring up just the deployment
nrl-k8s cluster up <recipe> --infra <infra> --target deployments.nemo_skills
# Tear down just the deployment
nrl-k8s cluster down <recipe> --infra <infra> --target deployments.nemo_skills
# Tear down everything (RayClusters + Deployments)
nrl-k8s cluster down <recipe> --infra <infra>

The deployments: section in infra YAML declares Kubernetes Deployments managed alongside RayClusters. The CLI patches image, imagePullSecrets, and serviceAccountName from the top-level infra keys (same as RayClusters). Deployments start in parallel with cluster bring-up — no ordering dependency.

8. Monitoring a run

# Status
nrl-k8s status <recipe> --infra <infra>
kubectl get rayjob,raycluster -n default

# Follow the driver
nrl-k8s job list <recipe> --infra <infra> --role training
nrl-k8s job logs <run-id> <recipe> --infra <infra> --role training -f

When the nrl-k8s job logs -f subprocess dies (kubectl port-forward i/o timeout after ~15 min idle), just re-run it. The training job keeps going.

To fetch driver logs fo