wild-cloud/bin/wild-node-detect

#!/bin/bash

# Node registration script for Talos cluster setup
# This script discovers hardware configuration from a node in maintenance mode
# and updates config.yaml with per-node hardware settings

set -euo pipefail

# Usage function
usage() {
    echo "Usage: wild-node-detect <node-ip>"
    echo ""
    echo "Detect hardware configuration of a Talos node in maintenance mode."
    echo "Returns JSON with discovered hardware information."
    echo ""
    echo "Arguments:"
    echo "  node-ip       Current IP of the node in maintenance mode"
    echo ""
    echo "Examples:"
    echo "  wild-node-detect 192.168.8.168"
    echo "  NODE_INFO=\$(wild-node-detect 192.168.8.169)"
    echo ""
    echo "This script will:"
    echo "  - Query the node for available network interfaces"
    echo "  - Query the node for available disks"  
    echo "  - Return JSON with hardware information"
    echo ""
    echo "Output JSON format:"
    echo '  {"interface": "eth0", "disks": ["/dev/sda", "/dev/nvme0n1"], "selected_disk": "/dev/sda"}'
}

# Parse arguments
if [ $# -ne 1 ]; then
    usage
    exit 1
fi

NODE_IP="$1"

echo "Detecting hardware for Talos node at $NODE_IP..." >&2

# Test connectivity
echo "Testing connectivity to node..." >&2

# Try insecure first (maintenance mode)
if talosctl -n "$NODE_IP" get links --insecure >/dev/null 2>&1; then
    TALOS_MODE="insecure"
    echo "✅ Node is accessible (maintenance mode)" >&2
# Try with configured credentials (regular mode)
elif talosctl -n "$NODE_IP" get links >/dev/null 2>&1; then
    TALOS_MODE="secure" 
    echo "✅ Node is accessible (configured mode)" >&2
else
    echo "Error: Cannot connect to Talos node at $NODE_IP" >&2
    echo "Make sure the node is running Talos and accessible." >&2
    exit 1
fi

# Discover network interfaces
echo "Discovering network interfaces..." >&2

# First, try to find the interface that's actually carrying traffic (has the default route)
if [ "$TALOS_MODE" = "insecure" ]; then
    CONNECTED_INTERFACE=$(talosctl -n "$NODE_IP" get routes --insecure -o json 2>/dev/null | \
        jq -s -r '.[] | select(.spec.destination == "0.0.0.0/0" and .spec.gateway != null) | .spec.outLinkName' | \
        head -1)
else
    CONNECTED_INTERFACE=$(talosctl -n "$NODE_IP" get routes -o json 2>/dev/null | \
        jq -s -r '.[] | select(.spec.destination == "0.0.0.0/0" and .spec.gateway != null) | .spec.outLinkName' | \
        head -1)
fi

if [ -n "$CONNECTED_INTERFACE" ]; then
    ACTIVE_INTERFACE="$CONNECTED_INTERFACE"
    echo "✅ Discovered connected interface (with default route): $ACTIVE_INTERFACE" >&2
else
    # Fallback: find physical ethernet interface (prefer eth*, en*, avoid virtual interfaces)
    echo "No default route found, checking for physical ethernet interfaces..." >&2
    if [ "$TALOS_MODE" = "insecure" ]; then
        ACTIVE_INTERFACE=$(talosctl -n "$NODE_IP" get links --insecure -o json 2>/dev/null | \
            jq -s -r '.[] | select(.spec.operationalState == "up" and .spec.type == "ether" and .metadata.id != "lo" and (.metadata.id | test("^(eth|en|eno|ens|enp)")) and (.metadata.id | test("(cni|flannel|docker|br-|veth)") | not)) | .metadata.id' | \
            head -1)
    else
        ACTIVE_INTERFACE=$(talosctl -n "$NODE_IP" get links -o json 2>/dev/null | \
            jq -s -r '.[] | select(.spec.operationalState == "up" and .spec.type == "ether" and .metadata.id != "lo" and (.metadata.id | test("^(eth|en|eno|ens|enp)")) and (.metadata.id | test("(cni|flannel|docker|br-|veth)") | not)) | .metadata.id' | \
            head -1)
    fi
    
    # If no physical interface found, fall back to any ethernet interface
    if [ -z "$ACTIVE_INTERFACE" ]; then
        echo "No physical ethernet interface found, checking any ethernet interface..." >&2
        if [ "$TALOS_MODE" = "insecure" ]; then
            ACTIVE_INTERFACE=$(talosctl -n "$NODE_IP" get links --insecure -o json 2>/dev/null | \
                jq -s -r '.[] | select(.spec.operationalState == "up" and .spec.type == "ether" and .metadata.id != "lo") | .metadata.id' | \
                head -1)
        else
            ACTIVE_INTERFACE=$(talosctl -n "$NODE_IP" get links -o json 2>/dev/null | \
                jq -s -r '.[] | select(.spec.operationalState == "up" and .spec.type == "ether" and .metadata.id != "lo") | .metadata.id' | \
                head -1)
        fi
    fi
    
    if [ -z "$ACTIVE_INTERFACE" ]; then
        echo "Error: No active ethernet interface found" >&2
        echo "Available interfaces:" >&2
        if [ "$TALOS_MODE" = "insecure" ]; then
            talosctl -n "$NODE_IP" get links --insecure >&2
        else
            talosctl -n "$NODE_IP" get links >&2
        fi
        echo "" >&2
        echo "Available routes:" >&2
        if [ "$TALOS_MODE" = "insecure" ]; then
            talosctl -n "$NODE_IP" get routes --insecure >&2
        else
            talosctl -n "$NODE_IP" get routes >&2
        fi
        exit 1
    fi
    
    echo "✅ Discovered active interface: $ACTIVE_INTERFACE" >&2
fi

# Discover available disks
echo "Discovering available disks..." >&2
if [ "$TALOS_MODE" = "insecure" ]; then
    DISKS_JSON=$(talosctl -n "$NODE_IP" get disks --insecure -o json 2>/dev/null | \
        jq -s '[.[] | select(.spec.size > 10000000000) | {path: ("/dev/" + .metadata.id), size: .spec.size}]')
else
    DISKS_JSON=$(talosctl -n "$NODE_IP" get disks -o json 2>/dev/null | \
        jq -s '[.[] | select(.spec.size > 10000000000) | {path: ("/dev/" + .metadata.id), size: .spec.size}]')
fi

if [ "$(echo "$DISKS_JSON" | jq 'length')" -eq 0 ]; then
    echo "Error: No suitable disks found (must be >10GB)" >&2
    echo "Available disks:" >&2
    if [ "$TALOS_MODE" = "insecure" ]; then
        talosctl -n "$NODE_IP" get disks --insecure >&2
    else
        talosctl -n "$NODE_IP" get disks >&2
    fi
    exit 1
fi

# Use the disks with size info directly
AVAILABLE_DISKS="$DISKS_JSON"

# Select the first disk as default
SELECTED_DISK=$(echo "$AVAILABLE_DISKS" | jq -r '.[0].path')

echo "✅ Discovered $(echo "$AVAILABLE_DISKS" | jq -r 'length') suitable disks" >&2
echo "✅ Selected disk: $SELECTED_DISK" >&2

# Output JSON to stdout
jq -n \
  --arg interface "$ACTIVE_INTERFACE" \
  --argjson disks "$AVAILABLE_DISKS" \
  --arg selected_disk "$SELECTED_DISK" \
  '{
    interface: $interface,
    disks: $disks,
    selected_disk: $selected_disk
  }'