bzzz/infrastructure/migration-scripts/migrate-v1-to-v2.sh

#!/bin/bash
set -euo pipefail

# BZZZ v1 to v2 Migration Script
# This script handles the complete migration from BZZZ v1 (SystemD) to v2 (Docker Swarm)

SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
LOG_FILE="/var/log/bzzz-migration-$(date +%Y%m%d-%H%M%S).log"
BACKUP_DIR="/rust/bzzz-v2/backup/$(date +%Y%m%d-%H%M%S)"
DRY_RUN=${DRY_RUN:-false}

# Colors for output
RED='\033[0;31m'
GREEN='\033[0;32m'
YELLOW='\033[1;33m'
BLUE='\033[0;34m'
NC='\033[0m' # No Color

log() {
    echo -e "${BLUE}[$(date +'%Y-%m-%d %H:%M:%S')]${NC} $1" | tee -a "$LOG_FILE"
}

error() {
    echo -e "${RED}[ERROR]${NC} $1" | tee -a "$LOG_FILE"
    exit 1
}

warn() {
    echo -e "${YELLOW}[WARN]${NC} $1" | tee -a "$LOG_FILE"
}

success() {
    echo -e "${GREEN}[SUCCESS]${NC} $1" | tee -a "$LOG_FILE"
}

check_prerequisites() {
    log "Checking prerequisites..."
    
    # Check if running as root for some operations
    if [[ $EUID -eq 0 ]]; then
        error "This script should not be run as root. Run as tony user with sudo access."
    fi
    
    # Check required commands
    local commands=("docker" "systemctl" "pg_dump" "rsync" "curl")
    for cmd in "${commands[@]}"; do
        if ! command -v "$cmd" &> /dev/null; then
            error "Required command '$cmd' not found"
        fi
    done
    
    # Check Docker Swarm status
    if ! docker info | grep -q "Swarm: active"; then
        error "Docker Swarm is not active. Please initialize swarm first."
    fi
    
    # Check available disk space
    local available=$(df /rust | awk 'NR==2 {print $4}')
    local required=10485760  # 10GB in KB
    if [[ $available -lt $required ]]; then
        error "Insufficient disk space. Need at least 10GB available in /rust"
    fi
    
    success "Prerequisites check passed"
}

backup_v1_data() {
    log "Creating backup of v1 data..."
    
    if [[ "$DRY_RUN" == "true" ]]; then
        log "[DRY RUN] Would create backup at: $BACKUP_DIR"
        return 0
    fi
    
    mkdir -p "$BACKUP_DIR"
    
    # Backup v1 configuration
    if [[ -d "/home/tony/chorus/project-queues/active/BZZZ" ]]; then
        rsync -av "/home/tony/chorus/project-queues/active/BZZZ/" "$BACKUP_DIR/v1-source/"
    fi
    
    # Backup systemd service files
    sudo cp /etc/systemd/system/bzzz.service "$BACKUP_DIR/" 2>/dev/null || true
    
    # Backup hypercore logs (if any)
    if [[ -d "/home/tony/.config/bzzz" ]]; then
        rsync -av "/home/tony/.config/bzzz/" "$BACKUP_DIR/config/"
    fi
    
    # Backup any existing data directories
    for node in walnut ironwood acacia; do
        if [[ -d "/rust/bzzz/$node" ]]; then
            rsync -av "/rust/bzzz/$node/" "$BACKUP_DIR/data/$node/"
        fi
    done
    
    success "Backup completed at: $BACKUP_DIR"
}

stop_v1_services() {
    log "Stopping BZZZ v1 services..."
    
    if [[ "$DRY_RUN" == "true" ]]; then
        log "[DRY RUN] Would stop v1 systemd services"
        return 0
    fi
    
    local nodes=("walnut" "ironwood" "acacia")
    for node in "${nodes[@]}"; do
        if sudo systemctl is-active --quiet "bzzz@$node" 2>/dev/null || sudo systemctl is-active --quiet bzzz 2>/dev/null; then
            log "Stopping BZZZ service on $node..."
            sudo systemctl stop "bzzz@$node" 2>/dev/null || sudo systemctl stop bzzz 2>/dev/null || true
            sudo systemctl disable "bzzz@$node" 2>/dev/null || sudo systemctl disable bzzz 2>/dev/null || true
        fi
    done
    
    # Wait for services to fully stop
    sleep 10
    
    success "v1 services stopped"
}

setup_v2_infrastructure() {
    log "Setting up v2 infrastructure..."
    
    if [[ "$DRY_RUN" == "true" ]]; then
        log "[DRY RUN] Would create v2 directory structure"
        return 0
    fi
    
    # Create directory structure
    mkdir -p /rust/bzzz-v2/{config,data,logs}
    mkdir -p /rust/bzzz-v2/data/{blobs,conversations,dht,postgres,redis}
    mkdir -p /rust/bzzz-v2/data/blobs/{data,index,temp}
    mkdir -p /rust/bzzz-v2/data/dht/{walnut,ironwood,acacia}
    mkdir -p /rust/bzzz-v2/config/{swarm,systemd,secrets}
    mkdir -p /rust/bzzz-v2/logs/{application,p2p,monitoring}
    
    # Set permissions
    sudo chown -R tony:tony /rust/bzzz-v2
    chmod -R 755 /rust/bzzz-v2
    
    # Create placeholder configuration files
    cat > /rust/bzzz-v2/config/bzzz-config.yaml << 'EOF'
agent:
  id: ""
  specialization: "advanced_reasoning"
  capabilities: ["code_generation", "debugging", "analysis"]
  models: ["llama3.2:70b", "qwen2.5:72b"]
  max_tasks: 3

whoosh_api:
  base_url: "http://whoosh.deepblack.cloud"
  api_key: ""

dht:
  bootstrap_nodes:
    - "walnut:9101"
    - "ironwood:9102" 
    - "acacia:9103"

content_store:
  path: "/app/data/blobs"
  replication_factor: 3
  shard_depth: 2

openai:
  rate_limit_rpm: 1000
  rate_limit_tpm: 100000
  cost_tracking: true
EOF
    
    success "v2 infrastructure setup completed"
}

migrate_conversation_data() {
    log "Migrating conversation data..."
    
    if [[ "$DRY_RUN" == "true" ]]; then
        log "[DRY RUN] Would migrate hypercore logs to content-addressed storage"
        return 0
    fi
    
    # Check if there are any hypercore logs to migrate
    local log_files=()
    for node in walnut ironwood acacia; do
        if [[ -f "/home/tony/.config/bzzz/hypercore-$node.log" ]]; then
            log_files+=("/home/tony/.config/bzzz/hypercore-$node.log")
        fi
    done
    
    if [[ ${#log_files[@]} -eq 0 ]]; then
        warn "No hypercore logs found for migration"
        return 0
    fi
    
    # Process each log file and create content-addressed blobs
    local migration_script="$SCRIPT_DIR/convert-hypercore-to-cas.py"
    if [[ -f "$migration_script" ]]; then
        python3 "$migration_script" "${log_files[@]}" --output-dir "/rust/bzzz-v2/data/blobs/data"
        success "Conversation data migrated to content-addressed storage"
    else
        warn "Migration script not found, skipping conversation data migration"
    fi
}

setup_docker_secrets() {
    log "Setting up Docker secrets..."
    
    if [[ "$DRY_RUN" == "true" ]]; then
        log "[DRY RUN] Would create Docker secrets"
        return 0
    fi
    
    # Create PostgreSQL password secret
    if [[ -f "/home/tony/chorus/business/secrets/postgres-bzzz-password" ]]; then
        docker secret create bzzz_postgres_password /home/tony/chorus/business/secrets/postgres-bzzz-password 2>/dev/null || true
    else
        # Generate random password
        openssl rand -base64 32 | docker secret create bzzz_postgres_password - 2>/dev/null || true
    fi
    
    # Create OpenAI API key secret
    if [[ -f "/home/tony/chorus/business/secrets/openai-api-key" ]]; then
        docker secret create bzzz_openai_api_key /home/tony/chorus/business/secrets/openai-api-key 2>/dev/null || true
    else
        warn "OpenAI API key not found in secrets directory"
    fi
    
    success "Docker secrets configured"
}

setup_docker_configs() {
    log "Setting up Docker configs..."
    
    if [[ "$DRY_RUN" == "true" ]]; then
        log "[DRY RUN] Would create Docker configs"
        return 0
    fi
    
    # Create main BZZZ config
    docker config create bzzz_v2_config /rust/bzzz-v2/config/bzzz-config.yaml 2>/dev/null || true
    
    # Create MCP server config
    cat > /tmp/mcp-config.yaml << 'EOF'
server:
  port: 3001
  max_connections: 1000
  timeout_seconds: 30

tools:
  enabled: true
  max_execution_time: 300

logging:
  level: info
  format: json
EOF
    docker config create bzzz_mcp_config /tmp/mcp-config.yaml 2>/dev/null || true
    rm /tmp/mcp-config.yaml
    
    # Create proxy config
    cat > /tmp/proxy-config.yaml << 'EOF'
openai:
  rate_limit:
    requests_per_minute: 1000
    tokens_per_minute: 100000
  cost_tracking:
    enabled: true
    log_requests: true
  models:
    - "gpt-4"
    - "gpt-4-turbo"
    - "gpt-3.5-turbo"

server:
  port: 3002
  timeout: 30s
EOF
    docker config create bzzz_proxy_config /tmp/proxy-config.yaml 2>/dev/null || true
    rm /tmp/proxy-config.yaml
    
    # Create Redis config
    cat > /tmp/redis.conf << 'EOF'
bind 0.0.0.0
port 6379
timeout 0
keepalive 300
maxclients 10000
maxmemory 1gb
maxmemory-policy allkeys-lru
save 900 1
save 300 10
save 60 10000
EOF
    docker config create bzzz_redis_config /tmp/redis.conf 2>/dev/null || true
    rm /tmp/redis.conf
    
    success "Docker configs created"
}

deploy_v2_stack() {
    log "Deploying BZZZ v2 Docker stack..."
    
    if [[ "$DRY_RUN" == "true" ]]; then
        log "[DRY RUN] Would deploy Docker stack with: docker stack deploy -c docker-compose.swarm.yml bzzz-v2"
        return 0
    fi
    
    cd "$SCRIPT_DIR/.."
    
    # Verify compose file
    if ! docker-compose -f infrastructure/docker-compose.swarm.yml config > /dev/null; then
        error "Docker compose file validation failed"
    fi
    
    # Deploy the stack
    docker stack deploy -c infrastructure/docker-compose.swarm.yml bzzz-v2
    
    # Wait for services to start
    log "Waiting for services to become ready..."
    local max_wait=300  # 5 minutes
    local wait_time=0
    
    while [[ $wait_time -lt $max_wait ]]; do
        local ready_services=$(docker service ls --filter label=com.docker.stack.namespace=bzzz-v2 --format "table {{.Name}}\t{{.Replicas}}" | grep -v "0/" | wc -l)
        local total_services=$(docker service ls --filter label=com.docker.stack.namespace=bzzz-v2 --format "table {{.Name}}" | wc -l)
        
        if [[ $ready_services -eq $total_services ]]; then
            success "All services are ready"
            break
        fi
        
        log "Waiting for services... ($ready_services/$total_services ready)"
        sleep 10
        wait_time=$((wait_time + 10))
    done
    
    if [[ $wait_time -ge $max_wait ]]; then
        error "Timeout waiting for services to become ready"
    fi
}

verify_v2_deployment() {
    log "Verifying v2 deployment..."
    
    # Check service health
    local services=("bzzz-v2_bzzz-agent" "bzzz-v2_postgres" "bzzz-v2_redis" "bzzz-v2_mcp-server")
    for service in "${services[@]}"; do
        if ! docker service ps "$service" | grep -q "Running"; then
            error "Service $service is not running properly"
        fi
    done
    
    # Test DHT connectivity
    log "Testing DHT connectivity..."
    if ! timeout 30 docker exec "$(docker ps -q -f label=com.docker.swarm.service.name=bzzz-v2_dht-bootstrap-walnut)" \
        curl -f http://localhost:9101/health > /dev/null 2>&1; then
        warn "DHT bootstrap node (walnut) health check failed"
    fi
    
    # Test MCP server
    log "Testing MCP server..."
    if ! timeout 10 curl -f http://localhost:3001/health > /dev/null 2>&1; then
        warn "MCP server health check failed"
    fi
    
    # Test content resolver
    log "Testing content resolver..."
    if ! timeout 10 curl -f http://localhost:3003/health > /dev/null 2>&1; then
        warn "Content resolver health check failed"
    fi
    
    success "v2 deployment verification completed"
}

update_node_labels() {
    log "Updating Docker node labels for service placement..."
    
    if [[ "$DRY_RUN" == "true" ]]; then
        log "[DRY RUN] Would update node labels"
        return 0
    fi
    
    # Set node labels for service placement
    docker node update --label-add bzzz.role=agent walnut 2>/dev/null || true
    docker node update --label-add bzzz.role=agent ironwood 2>/dev/null || true
    docker node update --label-add bzzz.role=agent acacia 2>/dev/null || true
    
    success "Node labels updated"
}

cleanup_v1_artifacts() {
    log "Cleaning up v1 artifacts..."
    
    if [[ "$DRY_RUN" == "true" ]]; then
        log "[DRY RUN] Would clean up v1 systemd files and binaries"
        return 0
    fi
    
    # Remove systemd service files (but keep backup)
    sudo rm -f /etc/systemd/system/bzzz.service
    sudo rm -f /etc/systemd/system/bzzz@.service
    sudo systemctl daemon-reload
    
    # Move v1 binaries to backup location
    if [[ -f "/home/tony/chorus/project-queues/active/BZZZ/bzzz" ]]; then
        mv "/home/tony/chorus/project-queues/active/BZZZ/bzzz" "$BACKUP_DIR/bzzz-v1-binary"
    fi
    
    success "v1 cleanup completed"
}

print_migration_summary() {
    log "Migration Summary:"
    log "=================="
    log "✅ v1 services stopped and disabled"
    log "✅ v2 infrastructure deployed to Docker Swarm"
    log "✅ Data migrated to content-addressed storage"
    log "✅ DHT network established across 3 nodes"
    log "✅ MCP server and OpenAI proxy deployed"
    log "✅ Monitoring and health checks configured"
    log ""
    log "Access Points:"
    log "- BZZZ Agent API: https://bzzz.deepblack.cloud"
    log "- MCP Server: https://mcp.deepblack.cloud" 
    log "- Content Resolver: https://resolve.deepblack.cloud"
    log "- OpenAI Proxy: https://openai.deepblack.cloud"
    log ""
    log "Monitoring:"
    log "- docker service ls --filter label=com.docker.stack.namespace=bzzz-v2"
    log "- docker stack ps bzzz-v2"
    log "- docker service logs bzzz-v2_bzzz-agent"
    log ""
    log "Backup Location: $BACKUP_DIR"
    log "Migration Log: $LOG_FILE"
}

rollback_to_v1() {
    log "Rolling back to v1..."
    
    # Stop v2 services
    docker stack rm bzzz-v2 2>/dev/null || true
    sleep 30
    
    # Restore v1 systemd service
    if [[ -f "$BACKUP_DIR/bzzz.service" ]]; then
        sudo cp "$BACKUP_DIR/bzzz.service" /etc/systemd/system/
        sudo systemctl daemon-reload
        sudo systemctl enable bzzz
        sudo systemctl start bzzz
    fi
    
    # Restore v1 binary
    if [[ -f "$BACKUP_DIR/bzzz-v1-binary" ]]; then
        cp "$BACKUP_DIR/bzzz-v1-binary" "/home/tony/chorus/project-queues/active/BZZZ/bzzz"
        chmod +x "/home/tony/chorus/project-queues/active/BZZZ/bzzz"
    fi
    
    success "Rollback to v1 completed"
}

main() {
    log "Starting BZZZ v1 to v2 migration..."
    log "DRY_RUN mode: $DRY_RUN"
    
    # Handle rollback if requested
    if [[ "${1:-}" == "--rollback" ]]; then
        rollback_to_v1
        return 0
    fi
    
    # Trap to handle errors
    trap 'error "Migration failed at line $LINENO"' ERR
    
    check_prerequisites
    backup_v1_data
    stop_v1_services
    setup_v2_infrastructure
    migrate_conversation_data
    setup_docker_secrets
    setup_docker_configs
    update_node_labels
    deploy_v2_stack
    verify_v2_deployment
    cleanup_v1_artifacts
    print_migration_summary
    
    success "BZZZ v2 migration completed successfully!"
    log "Run with --rollback to revert to v1 if needed"
}

# Handle script arguments
case "${1:-}" in
    --dry-run)
        DRY_RUN=true
        main
        ;;
    --rollback)
        main --rollback
        ;;
    --help|-h)
        echo "Usage: $0 [--dry-run|--rollback|--help]"
        echo ""
        echo "Options:"
        echo "  --dry-run   Preview migration steps without making changes"
        echo "  --rollback  Rollback to v1 (emergency use only)"
        echo "  --help      Show this help message"
        exit 0
        ;;
    *)
        main
        ;;
esac