yuan-lab-llm--clawmanager/deployments/k3s/single-node

ClawManager K3s Single-Node Manifest

This directory contains the all-in-one K3s manifest for a single-node ClawManager deployment:

• clawmanager.yaml: ClawManager workloads plus static HostPath PVs.
• No Longhorn dependency.
• No dynamic StorageClass dependency.

Use this profile only when ClawManager workloads and user runtimes can run on one storage node. For multi-node clusters, use deployments/k3s/cluster instead.

Storage Model

The manifest uses static PV/PVC binding:

• StorageClass name: manual
• System HostPath root: /data/clawmanager/system
• Runtime HostPath root: /data/clawreef
• Storage node label: clawmanager.io/storage-node=true

The manual value is only a matching name between bundled PVs and PVCs. It does not require a Kubernetes StorageClass object or dynamic provisioner.

The bundled PVs use persistentVolumeReclaimPolicy: Retain, so deleting the manifest does not erase HostPath data.

Install

1. Enter This Directory

cd deployments/k3s/single-node

Check that kubectl points to the target cluster:

kubectl config current-context
kubectl get nodes -o wide

Expected result: the target node is Ready.

2. Choose And Label One Storage Node

Label exactly one node before installing:

kubectl label node <node> clawmanager.io/storage-node=true --overwrite
kubectl get nodes -l clawmanager.io/storage-node=true

Expected result: only one node is listed.

If more than one node has this label, remove the label from the extra nodes:

kubectl label node <node> clawmanager.io/storage-node-

3. Check For Old Installation State

kubectl get ns clawmanager-system clawmanager-user-1 --ignore-not-found
kubectl get pv | grep clawmanager || true
kubectl get pvc -A | grep clawmanager || true

Expected result for a fresh install: no old ClawManager namespace, PV, or PVC remains.

If old resources exist, run the uninstall flow in this README before installing again.

4. Check HostPath Directories

The kubelet can create the HostPath directories automatically, but checking the target path first makes permission and disk issues easier to diagnose:

df -h /data || true
ls -ld /data || true

If /data is not the right disk for this node, edit these paths in clawmanager.yaml before applying it:

• /data/clawmanager/system/mysql
• /data/clawmanager/system/minio
• /data/clawmanager/system/redis
• /data/clawmanager/system/workspaces
• K8S_PV_HOST_PATH_PREFIX=/data/clawreef

5. Check Image Availability

The manifest uses the images listed in clawmanager.yaml. Make sure the node can pull from the referenced registry before installing:

grep -n 'image:' clawmanager.yaml | head -n 40

If this is an offline or private-registry deployment, load or push the images to the registry used by the manifest before continuing.

6. Apply The Manifest

kubectl apply -f clawmanager.yaml

7. Check ClawManager

kubectl -n clawmanager-system get pvc
kubectl -n clawmanager-system rollout status deployment/clawmanager-app --timeout=15m
kubectl -n clawmanager-system rollout status deployment/openclaw-runtime --timeout=15m
kubectl -n clawmanager-system rollout status deployment/hermes-runtime --timeout=15m
kubectl -n clawmanager-system get deploy,pod,pvc -o wide

Expected result:

• mysql-data, redis-data, minio-data, and clawmanager-workspaces are Bound.
• clawmanager-app, openclaw-runtime, and hermes-runtime are available.
• MySQL, Redis, MinIO, and skill scanner pods are running.

If PVCs stay Pending, check the static PV binding:

kubectl get pv | grep clawmanager
kubectl -n clawmanager-system describe pvc <pvc-name>

If pods stay Pending with node affinity errors, check the storage node label:

kubectl get nodes -l clawmanager.io/storage-node=true
kubectl describe pod <pod-name> -n clawmanager-system

If pods stay ContainerCreating, check recent events:

kubectl -n clawmanager-system get events --sort-by=.lastTimestamp | tail -n 80

Uninstall

Use this order for a disposable single-node test deployment.

1. Delete ClawManager Resources

kubectl delete -f clawmanager.yaml --ignore-not-found
kubectl delete namespace clawmanager-user-1 --ignore-not-found
kubectl delete namespace clawmanager-system --ignore-not-found

Check:

kubectl get ns clawmanager-system clawmanager-user-1 --ignore-not-found
kubectl get pv | grep clawmanager || true
kubectl get pvc -A | grep clawmanager || true

The PVs may remain because they use Retain. This is expected.

2. Remove Retained PVs When Data Is No Longer Needed

Only run this after confirming the data can be discarded:

kubectl delete pv \
  clawmanager-mysql-pv \
  clawmanager-minio-pv \
  clawmanager-redis-pv \
  clawmanager-workspaces-pv \
  --ignore-not-found

3. Remove HostPath Data When Data Is No Longer Needed

Run this on the labeled storage node only after confirming the data can be discarded:

rm -rf /data/clawmanager/system /data/clawreef

Common Issues

Symptom	Likely Cause	Check Or Fix
PVC stays Pending	Static PV and PVC did not bind	kubectl get pv, then kubectl describe pvc -n clawmanager-system <pvc-name>
Pod stays Pending	Missing or wrong clawmanager.io/storage-node=true label	kubectl get nodes -l clawmanager.io/storage-node=true
Pod is ImagePullBackOff	Node cannot pull an image	kubectl describe pod <pod-name> -n <namespace>
Pod is ContainerCreating	HostPath mount or permission issue	kubectl -n clawmanager-system get events --sort-by=.lastTimestamp
Runtime creation fails	Runtime HostPath root is unavailable	Check K8S_PV_HOST_PATH_PREFIX and disk space on the labeled node