proxmox/scripts/create-raid10-4disk-simple.sh.bak

#!/usr/bin/env bash
set -euo pipefail

# Simplified script to create RAID 10 with 4 available disks (sde-sdh)
# This is the safer approach - creates RAID, migrates data, keeps 4-disk RAID

set -u

TARGET_NODE="r630-01"
TARGET_NODE_IP="192.168.11.11"
TARGET_NODE_PASS="password"

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

log_info() { echo -e "${BLUE}[INFO]${NC} $1"; }
log_success() { echo -e "${GREEN}[✓]${NC} $1"; }
log_error() { echo -e "${RED}[✗]${NC} $1"; }
log_warn() { echo -e "${YELLOW}[⚠]${NC} $1"; }

ssh_r630_01() {
    sshpass -p "$TARGET_NODE_PASS" ssh -o StrictHostKeyChecking=no -o ConnectTimeout=10 root@"$TARGET_NODE_IP" "$@" 2>&1
}

install_mdadm() {
    log_info "Installing mdadm..."
    if ssh_r630_01 "which mdadm >/dev/null 2>&1"; then
        log_success "mdadm already installed"
        return 0
    fi
    
    ssh_r630_01 "apt-get update && apt-get install -y mdadm" || {
        log_error "Failed to install mdadm"
        return 1
    }
    log_success "mdadm installed"
    return 0
}

check_disks() {
    log_info "Checking disk availability..."
    
    local available_disks=("sde" "sdf" "sdg" "sdh")
    local missing_disks=()
    
    for disk in "${available_disks[@]}"; do
        if ! ssh_r630_01 "test -b /dev/$disk"; then
            missing_disks+=("$disk")
        fi
    done
    
    if [ ${#missing_disks[@]} -gt 0 ]; then
        log_error "Missing disks: ${missing_disks[*]}"
        return 1
    fi
    
    # Check if disks are in use
    for disk in "${available_disks[@]}"; do
        if ssh_r630_01 "pvs 2>/dev/null | grep -q /dev/$disk || mount | grep -q /dev/$disk"; then
            log_error "Disk /dev/$disk is already in use"
            return 1
        fi
    done
    
    log_success "All disks are available"
    return 0
}

create_raid10() {
    log_info "Creating RAID 10 with 4 disks (sde, sdf, sdg, sdh)..."
    
    ssh_r630_01 "mdadm --create /dev/md0 --level=10 --raid-devices=4 /dev/sde /dev/sdf /dev/sdg /dev/sdh" || {
        log_error "Failed to create RAID 10"
        return 1
    }
    
    log_success "RAID 10 created on /dev/md0"
    
    # Wait for sync
    log_info "Waiting for RAID array to synchronize (this may take 30-60 minutes)..."
    local max_wait=7200  # 2 hours max
    local waited=0
    
    while [ $waited -lt $max_wait ]; do
        local status=$(ssh_r630_01 "cat /proc/mdstat 2>/dev/null | grep -A 2 md0 | tail -1")
        
        if echo "$status" | grep -q "\[UUUU\]"; then
            log_success "RAID array is fully synchronized"
            break
        elif echo "$status" | grep -q "recovery\|resync"; then
            local progress=$(echo "$status" | grep -oP '\d+\.\d+%' || echo "in progress")
            if [ $((waited % 300)) -eq 0 ]; then  # Log every 5 minutes
                log_info "RAID sync progress: $progress (elapsed: $((waited/60)) minutes)"
            fi
            sleep 30
            waited=$((waited + 30))
        else
            sleep 10
            waited=$((waited + 10))
        fi
    done
    
    if [ $waited -ge $max_wait ]; then
        log_warn "RAID sync may still be in progress. Check manually: cat /proc/mdstat"
    fi
    
    # Save configuration
    log_info "Saving RAID configuration..."
    ssh_r630_01 "mdadm --detail --scan >> /etc/mdadm/mdadm.conf" || {
        log_warn "Failed to save to mdadm.conf, but RAID is created"
    }
    
    ssh_r630_01 "update-initramfs -u" || true
    
    return 0
}

add_to_lvm() {
    log_info "Adding RAID to pve volume group..."
    
    # Create physical volume
    ssh_r630_01 "pvcreate /dev/md0" || {
        log_error "Failed to create PV on RAID"
        return 1
    }
    
    # Extend VG
    ssh_r630_01 "vgextend pve /dev/md0" || {
        log_error "Failed to extend pve VG"
        return 1
    }
    
    log_success "RAID added to pve VG"
    
    # Show status
    log_info "Updated VG status:"
    ssh_r630_01 "vgs pve"
    ssh_r630_01 "pvs | grep pve"
    
    return 0
}

migrate_data() {
    log_info "Migrating data from sdc and sdd to RAID..."
    log_warn "This will take 1-3 hours depending on data size (~408GB)"
    
    # Migrate from sdc
    log_info "Migrating data from sdc to RAID..."
    ssh_r630_01 "pvmove /dev/sdc /dev/md0" || {
        log_error "Failed to migrate data from sdc"
        return 1
    }
    log_success "Data migrated from sdc"
    
    # Migrate from sdd
    log_info "Migrating data from sdd to RAID..."
    ssh_r630_01 "pvmove /dev/sdd /dev/md0" || {
        log_error "Failed to migrate data from sdd"
        return 1
    }
    log_success "Data migrated from sdd"
    
    return 0
}

remove_old_pvs() {
    log_info "Removing sdc and sdd from pve VG..."
    
    # Remove from VG
    ssh_r630_01 "vgreduce pve /dev/sdc /dev/sdd" || {
        log_error "Failed to remove PVs from VG"
        return 1
    }
    
    # Remove PV labels
    ssh_r630_01 "pvremove /dev/sdc /dev/sdd" || {
        log_error "Failed to remove PV labels"
        return 1
    }
    
    log_success "sdc and sdd removed from pve VG"
    
    # Verify
    log_info "Final VG status:"
    ssh_r630_01 "vgs pve"
    ssh_r630_01 "pvs | grep pve"
    
    return 0
}

show_status() {
    log_info "=== RAID Status ==="
    ssh_r630_01 "cat /proc/mdstat"
    echo ""
    ssh_r630_01 "mdadm --detail /dev/md0"
    echo ""
    log_info "=== LVM Status ==="
    ssh_r630_01 "vgs pve"
    ssh_r630_01 "pvs | grep pve"
}

main() {
    echo ""
    log_info "=== RAID 10 Setup for R630-01 ==="
    log_info "Creating RAID 10 with 4 disks: sde, sdf, sdg, sdh"
    log_info "Then migrating data from sdc/sdd to RAID"
    echo ""
    
    # Install mdadm
    if ! install_mdadm; then
        exit 1
    fi
    
    # Check disks
    if ! check_disks; then
        exit 1
    fi
    
    # Show current status
    log_info "Current storage status:"
    ssh_r630_01 "vgs pve"
    ssh_r630_01 "pvs | grep pve"
    echo ""
    
    log_warn "WARNING: This will:"
    log_warn "1. Create RAID 10 with 4 disks (sde-sdh)"
    log_warn "2. Migrate all data from sdc/sdd to RAID (~408GB)"
    log_warn "3. Remove sdc/sdd from pve VG"
    log_warn ""
    log_warn "Estimated time: 2-4 hours"
    log_warn "Some containers may need downtime during migration"
    echo ""
    read -p "Continue? (yes/no): " confirm
    if [ "$confirm" != "yes" ]; then
        log_info "Operation cancelled"
        exit 0
    fi
    
    # Create RAID
    if ! create_raid10; then
        exit 1
    fi
    
    # Add to LVM
    if ! add_to_lvm; then
        exit 1
    fi
    
    # Migrate data
    if ! migrate_data; then
        log_error "Data migration failed"
        log_warn "RAID is created and added to VG, but migration incomplete"
        exit 1
    fi
    
    # Remove old PVs
    if ! remove_old_pvs; then
        log_error "Failed to remove old PVs"
        exit 1
    fi
    
    # Show final status
    show_status
    
    log_success "RAID 10 setup completed successfully!"
    log_info ""
    log_info "RAID Device: /dev/md0"
    log_info "Capacity: ~466GB (RAID 10 with 4 disks)"
    log_info "Performance: Excellent (4x read, 2x write)"
    log_info "Redundancy: Can survive 1-2 disk failures"
    log_info ""
    log_info "sdc and sdd are now free and can be used for other purposes"
    log_info "or added to the RAID later if needed"
}

main "$@"