rdev/history/v0.1.0.md

rdev v0.1.0 - Base Case

Date: 2026-01-24 Status: Deployed and verified

Summary

First deployment of rdev - a single claudebox pod running on k3s with Claude Code CLI.

What Was Built

Container Image

• Image: ghcr.io/orchard9/rdev-claudebox:v0.1.0
• Base: Ubuntu 22.04
• Platform: linux/amd64 (k3s nodes are amd64, built from arm64 Mac)
• Contents:
  • Node.js 20 (required for Claude Code CLI)
  • Claude Code CLI v2.1.19
  • Git, vim, build-essential
  • Healthcheck script

Kubernetes Resources

• Namespace: rdev
• StatefulSet: claudebox (1 replica)
• PVCs:
  • claudebox-workspace (20Gi) - for project files
  • claudebox-claude-config (1Gi) - for Claude auth persistence
• Service: claudebox (headless)
• Secret: ghcr-secret (copied from apps namespace)

Resource Limits

requests:
  cpu: "500m"
  memory: "1Gi"
limits:
  cpu: "2"
  memory: "4Gi"

Key Decisions

1. GitHub Container Registry over GCP Artifact Registry

• This is an on-prem k3s cluster, not GKE
• Already have ghcr-secret for other images
• No gcloud dependency

2. PVC for Claude Config instead of Secret

• Claude credentials are too large for K8s secrets (>3MB limit)
• Auth persists across pod restarts
• User authenticates once via kubectl exec

3. Platform-specific Build

• k3s nodes are amd64
• Mac builds arm64 by default
• Added --platform linux/amd64 to docker build

Authentication

Claude Code authenticates via OAuth. To authenticate:

export KUBECONFIG=~/.kube/orchard9-k3sf.yaml
kubectl exec -it -n rdev claudebox-0 -- claude

This opens interactive mode. First run triggers OAuth flow - follow the URL to authenticate. Auth persists in the PVC.

Verification

# Check pod status
kubectl get pods -n rdev
# NAME          READY   STATUS    RESTARTS   AGE
# claudebox-0   1/1     Running   0          58s

# Check Claude version
kubectl exec -n rdev claudebox-0 -- claude --version
# 2.1.19 (Claude Code)

# Test Claude (after auth)
kubectl exec -it -n rdev claudebox-0 -- claude "say hello"

Files Created

rdev/
├── CLAUDE.md                           # Project instructions
├── README.md                           # Setup guide
├── Dockerfile                          # claudebox image
├── .gitignore
├── docs/
│   └── reference.md                    # Original reference guide
├── deployments/
│   └── k8s/
│       └── base/
│           ├── kustomization.yaml
│           ├── namespace.yaml
│           ├── pvc.yaml                # workspace + claude-config PVCs
│           └── claudebox.yaml          # StatefulSet + Service
├── scripts/
│   ├── build-push.sh                   # Build and push to ghcr.io
│   ├── deploy.sh                       # Deploy to k3s
│   ├── verify.sh                       # Verify deployment
│   └── create-credentials-secret.sh    # (deprecated - using PVC now)
└── history/
    └── v0.1.0.md                       # This file

Issues Encountered

1. Image Pull Failed - Wrong Platform

• Error: no match for platform in manifest: not found
• Cause: Built arm64 image on Mac, k3s nodes are amd64
• Fix: Added --platform linux/amd64 to docker build

2. Credentials Secret Too Large

• Error: Request entity too large: limit is 3145728
• Cause: ~/.claude directory is ~30MB (history, debug logs, etc.)
• Fix: Switched to PVC for claude config instead of mounting from secret

3. GCP Artifact Registry Auth Failed

• Error: 403 Forbidden when pushing to us-central1-docker.pkg.dev
• Cause: Wrong approach - this is k3s, not GKE
• Fix: Switched to GitHub Container Registry (ghcr.io)

What's Next (v0.2)

1. Mount real project workspaces (pantheon, aeries repos)
2. Add SSH keys for git push/pull
3. Multiple claudebox pods (one per project)
4. Project-specific environment configurations

Commands Reference

# Set kubeconfig (REQUIRED)
export KUBECONFIG=~/.kube/orchard9-k3sf.yaml

# Deploy
kubectl apply -k deployments/k8s/base

# Check status
kubectl get pods -n rdev

# Interactive Claude session
kubectl exec -it -n rdev claudebox-0 -- claude

# Run Claude with prompt
kubectl exec -it -n rdev claudebox-0 -- claude "your prompt here"

# Shell access
kubectl exec -it -n rdev claudebox-0 -- bash

# View logs
kubectl logs -n rdev claudebox-0

# Restart pod
kubectl delete pod -n rdev claudebox-0