DR: full bootstrap script + recovery runbook

bootstrap.sh now covers a complete fresh Ubuntu 24.04 setup:
- NVIDIA driver 580 + container toolkit + k3s GPU config
- k3s install with nvidia as default runtime + device plugin
- Host cloudflared (Gitea SSH tunnel)
- Terraform + Helm
- systemd-resolved disabled, resolv.conf pinned

RUNBOOK.md documents the full 4-step recovery:
bootstrap → terraform apply → Gitea/Postgres restore → verify
Target: under 1 hour on a new machine.
This commit is contained in:
Will Anderson
2026-03-24 00:46:36 -05:00
parent 82b902f7c4
commit f8bc853136
2 changed files with 308 additions and 15 deletions
+176
View File
@@ -0,0 +1,176 @@
# Legion Disaster Recovery Runbook
Complete recovery from total loss of the Legion server.
Target: running cluster in under 2 hours on a fresh machine.
---
## Prerequisites (stored offsite)
| What | Where |
|------|-------|
| `~/Secrets` (encrypted) | R2 bucket `legion-secrets-backup` |
| Terraform state | R2 bucket `legion-terraform-state` |
| Gitea repos + DB dump | R2 bucket `legion-gitea-backup` (restic) |
| age decryption key | Printed copy / separate secure storage |
> **Note:** The age private key (`~/Secrets/age-backup.key`) is inside the encrypted
> archive — which means you need the key to decrypt the archive that contains the key.
> Store the age private key somewhere independent (password manager, printed sheet, etc.).
---
## Step 1 — Recover Secrets (on Mac or any machine with `age` + `aws`)
```bash
# Install age if needed
brew install age
# Configure R2 access (you'll need the R2 access key from your password manager)
export AWS_ACCESS_KEY_ID=<r2-access-key-id>
export AWS_SECRET_ACCESS_KEY=<r2-secret-access-key>
R2_ENDPOINT="https://651161e0a3d321561b4c90b5bcd5f15b.r2.cloudflarestorage.com"
# Download the latest encrypted backup
aws s3 cp s3://legion-secrets-backup/secrets-latest.tar.gz.age /tmp/secrets.tar.gz.age \
--endpoint-url "$R2_ENDPOINT"
# Decrypt and extract (you need the age private key for this)
age --decrypt -i /path/to/age-backup.key /tmp/secrets.tar.gz.age | tar -xz -C ~/
# Verify
ls ~/Secrets/credentials/
source ~/Secrets/credentials/infrastructure.env
```
---
## Step 2 — Provision Fresh Machine
Fresh Ubuntu 24.04 LTS, same LAN IP (`192.168.68.77`).
```bash
# From Mac — copy SSH key and cloudflared credentials
ssh-copy-id will@192.168.68.77
scp ~/Secrets/certs/cloudflared-gitea.json legion:/tmp/cloudflared-gitea.json
# Run bootstrap
ssh legion "bash -s" < ~/Development/infrastructure/servers/legion/bootstrap.sh
# Copy kubeconfig back to Mac
scp legion:/home/will/.kube/config ~/.kube/legion-config
export KUBECONFIG=~/.kube/legion-config
kubectl get nodes # should show legion as Ready
```
**Bootstrap installs:**
- NVIDIA driver 580 + container toolkit
- k3s (latest stable) with NVIDIA as default runtime
- NVIDIA device plugin
- Cloudflared (host-level, Gitea SSH tunnel)
- Terraform + Helm
---
## Step 3 — Apply Terraform
Terraform state lives in R2 — it will re-create all k8s resources from scratch.
```bash
cd ~/Development/infrastructure/servers/legion/
direnv exec . terraform init # pulls state from R2
direnv exec . terraform plan # review what will be created
direnv exec . terraform apply # recreates everything: namespaces, PVCs, secrets, Helm releases, Argo CD
```
Wait for Argo CD to sync (~2 minutes after apply). Check:
```bash
kubectl get pods -A # all pods should reach Running
```
---
## Step 4 — Restore Gitea Data
Gitea repos and the Postgres DB are backed up nightly via restic to `legion-gitea-backup`.
```bash
# Set up restic env
set -a; source ~/Secrets/credentials/infrastructure.env; set +a
export RESTIC_REPOSITORY="s3:$(secret r2-endpoint)/legion-gitea-backup"
export RESTIC_PASSWORD="gitea-restic-2026"
export AWS_ACCESS_KEY_ID=$(secret r2-access-key)
export AWS_SECRET_ACCESS_KEY=$(secret r2-secret-key)
# List available snapshots
restic snapshots
# Restore Gitea data volume to a local dir
restic restore latest --target /tmp/gitea-restore --include /gitea-data
# Copy into the PVC via kubectl
GITEA_POD=$(kubectl get pod -n git -l app=gitea -o jsonpath='{.items[0].metadata.name}')
kubectl cp /tmp/gitea-restore/gitea-data/. git/$GITEA_POD:/data/gitea/
# Restore Postgres DB dump
restic restore latest --target /tmp/gitea-restore --include /dump/all-databases.sql
kubectl exec -n platform deploy/postgres-postgresql -- \
psql -U postgres < /tmp/gitea-restore/dump/all-databases.sql
# Restart Gitea to pick up restored data
kubectl rollout restart deployment/gitea -n git
```
---
## Step 5 — Verify
```bash
# k8s cluster healthy
kubectl get nodes
kubectl get pods -A | grep -v Running | grep -v Completed
# Argo CD synced
kubectl get applications -n argocd
# DNS working (AdGuard)
dig @192.168.68.77 git.neuralplatform.ai
# Services reachable
curl -sk https://git.neuralplatform.ai/api/v1/version
curl -sk https://argocd.neuralplatform.ai/healthz
# GPU available
kubectl describe node legion | grep nvidia.com/gpu
```
---
## Key Reference
| Secret | Vault path | Field |
|--------|------------|-------|
| R2 access key | `secret/r2` | `access_key_id` |
| R2 secret | `secret/r2` | `secret_access_key` |
| Cloudflare API key | `secret/cloudflare` | `api_key` |
| Gitea admin token | `secret/gitea` | `api_token` |
| Postgres password | `secret/postgres` | `password` |
| Vault root token | `~/Secrets/tokens/vault-root-token` | — |
Tunnel IDs:
- Platform tunnel (k8s pod): `54bc9b05-3953-47a2-9c3e-adecdcc53d51`
- Gitea SSH tunnel (host cloudflared): `b1bf80bf-5448-4c33-8667-d6fce8a82c7a`
---
## Estimated Recovery Time
| Step | Time |
|------|------|
| Recover secrets | 5 min |
| Provision OS + bootstrap | 20-30 min (driver install) |
| Terraform apply | 10-15 min |
| Gitea data restore | 10 min |
| Verify | 5 min |
| **Total** | **~1 hour** |
+132 -15
View File
@@ -1,18 +1,70 @@
#!/usr/bin/env bash
# Legion host bootstrap — run once on a fresh install before Terraform
# Usage: ssh legion "bash -s" < bootstrap.sh
# Legion host bootstrap — run ONCE on a fresh Ubuntu 24.04 install, before Terraform.
#
# Usage (from Mac after restoring ~/Secrets):
# ssh-copy-id legion
# scp ~/Secrets/cloudflared-gitea.json legion:/tmp/cloudflared-gitea.json
# ssh legion "bash -s" < bootstrap.sh
#
# After this script: run `direnv exec . terraform apply` from ~/Development/infrastructure/servers/legion/
set -euo pipefail
LEGION_IP="${1:-192.168.68.77}"
NVIDIA_DRIVER="nvidia-driver-580"
echo "==> Disabling systemd-resolved (conflicts with AdGuard on port 53)"
sudo systemctl disable --now systemd-resolved || true
sudo rm -f /etc/resolv.conf
echo "nameserver 1.1.1.1" | sudo tee /etc/resolv.conf
log() { echo ""; echo "==> $*"; }
echo "==> Configuring k3s tls-san for LAN IP ${LEGION_IP}"
sudo mkdir -p /etc/rancher/k3s
sudo tee /etc/rancher/k3s/config.yaml > /dev/null << EOF
# ── 0. Sanity check ──────────────────────────────────────────────────────────
if [ "$(id -u)" != "0" ]; then
echo "Run as root or via sudo. Re-running with sudo..."
exec sudo -E bash "$0" "$@"
fi
log "Starting Legion bootstrap (IP: ${LEGION_IP})"
# ── 1. Base packages ──────────────────────────────────────────────────────────
log "Installing base prerequisites"
apt-get update -qq
apt-get install -y --no-install-recommends \
curl wget gnupg ca-certificates apt-transport-https \
nfs-common open-iscsi \
jq unzip git
# Enable open-iscsi (required for some k3s storage backends)
systemctl enable --now iscsid || true
# ── 2. Disable systemd-resolved (conflicts with AdGuard on port 53) ───────────
log "Disabling systemd-resolved"
systemctl disable --now systemd-resolved || true
rm -f /etc/resolv.conf
echo "nameserver 1.1.1.1" > /etc/resolv.conf
echo "nameserver 8.8.8.8" >> /etc/resolv.conf
chattr +i /etc/resolv.conf # prevent NetworkManager from overwriting it
# ── 3. NVIDIA drivers ─────────────────────────────────────────────────────────
log "Installing NVIDIA drivers (${NVIDIA_DRIVER})"
add-apt-repository -y ppa:graphics-drivers/ppa
apt-get update -qq
apt-get install -y "${NVIDIA_DRIVER}"
# NVIDIA container toolkit (required for k3s GPU support)
log "Installing NVIDIA container toolkit"
curl -fsSL https://nvidia.github.io/libnvidia-container/gpgkey \
| gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg
curl -s -L https://nvidia.github.io/libnvidia-container/stable/deb/nvidia-container-toolkit.list \
| sed 's#deb https://#deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://#g' \
> /etc/apt/sources.list.d/nvidia-container-toolkit.list
apt-get update -qq
apt-get install -y nvidia-container-toolkit
# Configure NVIDIA runtime for containerd (k3s uses containerd)
nvidia-ctk runtime configure --runtime=containerd
# ── 4. k3s ────────────────────────────────────────────────────────────────────
log "Configuring k3s"
mkdir -p /etc/rancher/k3s
cat > /etc/rancher/k3s/config.yaml << EOF
write-kubeconfig-mode: "644"
default-runtime: nvidia
tls-san:
@@ -20,12 +72,77 @@ tls-san:
- 127.0.0.1
EOF
echo "==> Rotating k3s certificates"
sudo k3s certificate rotate
sudo systemctl restart k3s
log "Installing k3s"
curl -sfL https://get.k3s.io | sh -
echo "==> Waiting for k3s to come back up..."
log "Waiting for k3s to start..."
sleep 20
sudo k3s kubectl get nodes
k3s kubectl get nodes
echo "==> Bootstrap complete"
# Copy kubeconfig to user home (for 'will' user)
mkdir -p /home/will/.kube
cp /etc/rancher/k3s/k3s.yaml /home/will/.kube/config
sed -i "s/127.0.0.1/${LEGION_IP}/g" /home/will/.kube/config
chown will:will /home/will/.kube/config
# ── 5. NVIDIA device plugin for k3s ──────────────────────────────────────────
log "Installing NVIDIA k3s device plugin"
k3s kubectl create -f https://raw.githubusercontent.com/NVIDIA/k8s-device-plugin/v0.14.5/nvidia-device-plugin.yml \
|| k3s kubectl apply -f https://raw.githubusercontent.com/NVIDIA/k8s-device-plugin/v0.14.5/nvidia-device-plugin.yml
# ── 6. Cloudflared (host-level — Gitea SSH tunnel only) ───────────────────────
# The platform tunnel (neuralplatform.ai) runs as a k8s pod managed by Terraform.
# This host-level cloudflared serves only the Gitea SSH NodePort (git.neuralplatform.ai → :30022).
log "Installing cloudflared"
curl -fsSL https://pkg.cloudflare.com/cloudflare-main.gpg \
| gpg --dearmor -o /usr/share/keyrings/cloudflare-main.gpg
echo 'deb [signed-by=/usr/share/keyrings/cloudflare-main.gpg] https://pkg.cloudflare.com/cloudflared any main' \
> /etc/apt/sources.list.d/cloudflared.list
apt-get update -qq
apt-get install -y cloudflared
# Restore tunnel credentials from Mac-side secret (must exist at /tmp/cloudflared-gitea.json)
TUNNEL_ID="b1bf80bf-5448-4c33-8667-d6fce8a82c7a"
if [ -f "/tmp/cloudflared-gitea.json" ]; then
mkdir -p /etc/cloudflared
cp /tmp/cloudflared-gitea.json "/etc/cloudflared/${TUNNEL_ID}.json"
chmod 600 "/etc/cloudflared/${TUNNEL_ID}.json"
cat > /etc/cloudflared/config.yml << EOF
tunnel: ${TUNNEL_ID}
credentials-file: /etc/cloudflared/${TUNNEL_ID}.json
ingress:
- hostname: git.neuralplatform.ai
service: http://localhost:30322
- service: http_status:404
EOF
cloudflared service install
systemctl enable --now cloudflared
log "Cloudflared installed and running"
else
log "WARNING: /tmp/cloudflared-gitea.json not found — skipping cloudflared setup"
log " To fix: scp ~/Secrets/certs/cloudflared-gitea.json legion:/tmp/cloudflared-gitea.json"
log " Then re-run: cloudflared service install && systemctl enable --now cloudflared"
fi
# ── 7. CLI tools (Terraform, Helm) ────────────────────────────────────────────
log "Installing Terraform"
wget -qO- https://apt.releases.hashicorp.com/gpg \
| gpg --dearmor -o /usr/share/keyrings/hashicorp-archive-keyring.gpg
echo "deb [signed-by=/usr/share/keyrings/hashicorp-archive-keyring.gpg] https://apt.releases.hashicorp.com $(lsb_release -cs) main" \
> /etc/apt/sources.list.d/hashicorp.list
apt-get update -qq && apt-get install -y terraform
log "Installing Helm"
curl -fsSL https://raw.githubusercontent.com/helm/helm/main/scripts/get-helm-3 | bash
# ── 8. Done ───────────────────────────────────────────────────────────────────
log "Bootstrap complete!"
echo ""
echo "Next steps:"
echo " 1. Copy kubeconfig to Mac: scp legion:/home/will/.kube/config ~/.kube/legion-config"
echo " 2. Apply Terraform: cd ~/Development/infrastructure/servers/legion && direnv exec . terraform apply"
echo " 3. Restore Gitea data: see RUNBOOK.md § Data Restore"
echo ""
echo "Verify GPU access:"
echo " ssh legion kubectl describe node | grep nvidia.com/gpu"