microdao-daarion/scripts/node-bootstrap.sh

#!/usr/bin/env bash
#
# DAARION Node Bootstrap Script
# Виконує самореєстрацію ноди при старті.
#
# Використання:
#   ./scripts/node-bootstrap.sh
#
# Environment variables:
#   CITY_SERVICE_URL - URL city-service (default: http://localhost:7001)
#   NODE_ID - Ідентифікатор ноди (required)
#   NODE_NAME - Назва ноди (required)
#   NODE_ENVIRONMENT - production|development|staging (default: development)
#   NODE_HOSTNAME - Hostname (optional)
#   NODE_ROLES - Ролі через кому: gpu,ai_runtime,storage (default: gpu,ai_runtime)
#   NODE_DESCRIPTION - Опис ноди (optional)

set -euo pipefail

# Colors
RED='\033[0;31m'
GREEN='\033[0;32m'
YELLOW='\033[1;33m'
BLUE='\033[0;34m'
NC='\033[0m'

log_info() {
    echo -e "${BLUE}[NODE-BOOTSTRAP]${NC} $1"
}

log_success() {
    echo -e "${GREEN}[NODE-BOOTSTRAP]${NC} ✅ $1"
}

log_error() {
    echo -e "${RED}[NODE-BOOTSTRAP]${NC} ❌ $1"
}

log_warning() {
    echo -e "${YELLOW}[NODE-BOOTSTRAP]${NC} ⚠️  $1"
}

# Configuration
CITY_SERVICE_URL="${CITY_SERVICE_URL:-http://localhost:7001}"
NODE_ID="${NODE_ID:-}"
NODE_NAME="${NODE_NAME:-}"
NODE_ENVIRONMENT="${NODE_ENVIRONMENT:-development}"
NODE_HOSTNAME="${NODE_HOSTNAME:-$(hostname 2>/dev/null || echo '')}"
NODE_ROLES="${NODE_ROLES:-gpu,ai_runtime}"
NODE_DESCRIPTION="${NODE_DESCRIPTION:-}"

# Retry settings
MAX_RETRIES=5
RETRY_DELAY=5

# Validate required params
if [ -z "$NODE_ID" ]; then
    log_error "NODE_ID is required"
    exit 1
fi

if [ -z "$NODE_NAME" ]; then
    log_error "NODE_NAME is required"
    exit 1
fi

# Convert roles to JSON array
roles_json=""
IFS=',' read -ra ROLE_ARRAY <<< "$NODE_ROLES"
for i in "${!ROLE_ARRAY[@]}"; do
    if [ $i -eq 0 ]; then
        roles_json="\"${ROLE_ARRAY[$i]}\""
    else
        roles_json="$roles_json, \"${ROLE_ARRAY[$i]}\""
    fi
done
roles_json="[$roles_json]"

# Build payload
payload=$(cat <<EOF
{
    "id": "$NODE_ID",
    "name": "$NODE_NAME",
    "hostname": "$NODE_HOSTNAME",
    "environment": "$NODE_ENVIRONMENT",
    "roles": $roles_json,
    "description": "$NODE_DESCRIPTION"
}
EOF
)

log_info "Starting node self-registration..."
log_info "  Node ID:      $NODE_ID"
log_info "  Node Name:    $NODE_NAME"
log_info "  Environment:  $NODE_ENVIRONMENT"
log_info "  Hostname:     $NODE_HOSTNAME"
log_info "  Roles:        $NODE_ROLES"
log_info "  City Service: $CITY_SERVICE_URL"

# Self-registration with retries
attempt=1
while [ $attempt -le $MAX_RETRIES ]; do
    log_info "Registration attempt $attempt/$MAX_RETRIES..."
    
    response=$(curl -s -w "\n%{http_code}" \
        -X POST "${CITY_SERVICE_URL}/internal/nodes/register-or-update" \
        -H "Content-Type: application/json" \
        -d "$payload" \
        --max-time 10 \
        2>/dev/null || echo -e "\n000")
    
    http_code=$(echo "$response" | tail -n1)
    body=$(echo "$response" | sed '$d')
    
    if [ "$http_code" = "200" ]; then
        success=$(echo "$body" | grep -o '"success":\s*true' || true)
        
        if [ -n "$success" ]; then
            is_new=$(echo "$body" | grep -o '"is_new":\s*true' || true)
            
            if [ -n "$is_new" ]; then
                log_success "Node registered successfully (new registration)"
            else
                log_success "Node updated successfully"
            fi
            
            # Optional: run initial heartbeat
            log_info "Sending initial heartbeat..."
            curl -s -X POST "${CITY_SERVICE_URL}/internal/node/${NODE_ID}/heartbeat" \
                -H "Content-Type: application/json" \
                -d '{"metrics": {}}' \
                --max-time 5 > /dev/null 2>&1 || true
            
            log_success "Node bootstrap completed"
            exit 0
        fi
    fi
    
    log_warning "Registration failed (HTTP $http_code)"
    
    if [ $attempt -lt $MAX_RETRIES ]; then
        log_info "Retrying in ${RETRY_DELAY}s..."
        sleep $RETRY_DELAY
    fi
    
    attempt=$((attempt + 1))
done

log_error "Node registration failed after $MAX_RETRIES attempts"
log_error "Please check:"
log_error "  1. City service is running at $CITY_SERVICE_URL"
log_error "  2. Migration 039_node_registry_self_healing.sql is applied"
log_error "  3. Network connectivity to city service"
exit 1