proxmox/scripts/repair-thin-storage.sh

#!/usr/bin/env bash
# Repair and redeploy thin1-thin3 storage pools on pve and pve2
# This script removes broken storage configurations and recreates them properly

set -euo pipefail

# Load IP configuration
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
PROJECT_ROOT="$(cd "$SCRIPT_DIR/.." && pwd)"
source "${PROJECT_ROOT}/config/ip-addresses.conf" 2>/dev/null || true


PROXMOX_HOST_PVE="${PROXMOX_HOST_R630_01}"
PROXMOX_HOST_PVE2="${PROXMOX_HOST_R630_02}"
PVE_PASS="password"
STORAGE_POOLS=("thin1" "thin2" "thin3")

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

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

# Check what's using a storage pool
check_storage_usage() {
    local host=$1
    local storage=$2
    
    log_info "Checking if $storage is in use on $host..."
    
    local usage=$(sshpass -p "$PVE_PASS" ssh -o StrictHostKeyChecking=accept-new root@${host} \
        "pvesm list $storage 2>/dev/null | wc -l" || echo "0")
    
    if [ "$usage" -gt 1 ]; then
        log_warn "Storage $storage has $((usage - 1)) items (header excluded)"
        return 1
    else
        log_info "Storage $storage appears to be unused"
        return 0
    fi
}

# Remove storage from Proxmox
remove_storage() {
    local host=$1
    local storage=$2
    
    log_info "Removing storage $storage from Proxmox on $host..."
    
    # Check if storage exists
    if ! sshpass -p "$PVE_PASS" ssh -o StrictHostKeyChecking=accept-new root@${host} \
        "pvesm status 2>/dev/null | grep -q '$storage'" 2>/dev/null; then
        log_info "Storage $storage does not exist, skipping removal"
        return 0
    fi
    
    # Remove from Proxmox config
    sshpass -p "$PVE_PASS" ssh -o StrictHostKeyChecking=accept-new root@${host} \
        "pvesm remove $storage 2>&1" || {
        log_warn "Failed to remove via pvesm, trying to edit config directly..."
        # Remove from storage.cfg
        sshpass -p "$PVE_PASS" ssh -o StrictHostKeyChecking=accept-new root@${host} \
            "sed -i '/^lvmthin: $storage$/,/^$/d' /etc/pve/storage.cfg 2>/dev/null" || true
    }
    
    log_success "Storage $storage removed from Proxmox configuration"
}

# Get volume group name for storage
get_vg_for_storage() {
    local host=$1
    local storage=$2
    
    sshpass -p "$PVE_PASS" ssh -o StrictHostKeyChecking=accept-new root@${host} \
        "cat /etc/pve/storage.cfg 2>/dev/null | grep -A 5 'lvmthin: $storage' | grep 'vgname' | awk '{print \$2}'" 2>/dev/null || echo ""
}

# Get thin pool name for storage
get_thinpool_for_storage() {
    local host=$1
    local storage=$2
    
    local pool=$(sshpass -p "$PVE_PASS" ssh -o StrictHostKeyChecking=accept-new root@${host} \
        "cat /etc/pve/storage.cfg 2>/dev/null | grep -A 5 'lvmthin: $storage' | grep 'thinpool' | awk '{print \$2}'" 2>/dev/null || echo "")
    
    # If pool is in vg/pool format, extract just the pool name
    if [[ "$pool" == *"/"* ]]; then
        pool=$(echo "$pool" | cut -d'/' -f2)
    fi
    
    echo "$pool"
}

# Check if VG exists
check_vg_exists() {
    local host=$1
    local vg_name=$2
    
    sshpass -p "$PVE_PASS" ssh -o StrictHostKeyChecking=accept-new root@${host} \
        "vgs $vg_name 2>/dev/null | grep -q '$vg_name'" 2>/dev/null
}

# Check if thin pool exists in VG
check_thinpool_exists() {
    local host=$1
    local vg_name=$2
    local pool_name=$3
    
    sshpass -p "$PVE_PASS" ssh -o StrictHostKeyChecking=accept-new root@${host} \
        "lvs $vg_name/$pool_name 2>/dev/null | grep -q '$pool_name'" 2>/dev/null
}

# Create thin pool if it doesn't exist
create_thin_pool() {
    local host=$1
    local vg_name=$2
    local pool_name=$3
    
    log_info "Creating thin pool $vg_name/$pool_name on $host..."
    
    # Check if VG exists
    if ! check_vg_exists "$host" "$vg_name"; then
        log_error "Volume group $vg_name does not exist on $host"
        return 1
    fi
    
    # Check if pool already exists
    if check_thinpool_exists "$host" "$vg_name" "$pool_name"; then
        log_info "Thin pool $vg_name/$pool_name already exists"
        return 0
    fi
    
    # Get available space
    local vg_free=$(sshpass -p "$PVE_PASS" ssh -o StrictHostKeyChecking=accept-new root@${host} \
        "vgs -o vg_free --noheadings --units g $vg_name 2>/dev/null | awk '{print int(\$1)}'" || echo "0")
    
    if [ "$vg_free" -lt 10 ]; then
        log_error "Not enough free space in $vg_name (${vg_free}G available)"
        return 1
    fi
    
    # Use 80% of available space
    local pool_size=$((vg_free * 80 / 100))
    log_info "Creating thin pool with ${pool_size}G (80% of ${vg_free}G free)"
    
    sshpass -p "$PVE_PASS" ssh -o StrictHostKeyChecking=accept-new root@${host} <<EOF
lvcreate -L ${pool_size}G -n ${pool_name} ${vg_name} 2>&1
lvconvert --type thin-pool ${vg_name}/${pool_name} 2>&1
EOF
    
    if [ $? -eq 0 ]; then
        log_success "Thin pool $vg_name/$pool_name created"
        return 0
    else
        log_error "Failed to create thin pool"
        return 1
    fi
}

# Recreate storage in Proxmox
recreate_storage() {
    local host=$1
    local storage=$2
    local vg_name=$3
    local pool_name=$4
    local node_name=$5
    
    log_info "Recreating storage $storage on $host..."
    
    # Create thin pool if needed
    if ! check_thinpool_exists "$host" "$vg_name" "$pool_name"; then
        if ! create_thin_pool "$host" "$vg_name" "$pool_name"; then
            log_error "Cannot recreate storage without thin pool"
            return 1
        fi
    fi
    
    # Add storage to Proxmox
    # Note: thinpool parameter should be just the pool name, not vg/pool
    sshpass -p "$PVE_PASS" ssh -o StrictHostKeyChecking=accept-new root@${host} <<EOF
pvesm add lvmthin $storage \
    --thinpool ${pool_name} \
    --vgname ${vg_name} \
    --content images,rootdir \
    --nodes ${node_name} 2>&1
EOF
    
    if [ $? -eq 0 ]; then
        log_success "Storage $storage recreated successfully"
        return 0
    else
        log_warn "Storage add command returned non-zero, checking if it exists..."
        # Check if it was created
        if sshpass -p "$PVE_PASS" ssh -o StrictHostKeyChecking=accept-new root@${host} \
            "pvesm status 2>/dev/null | grep -q '$storage'" 2>/dev/null; then
            log_success "Storage $storage exists (may have already been configured)"
            return 0
        else
            log_error "Failed to recreate storage"
            return 1
        fi
    fi
}

# Process a single node
process_node() {
    local host=$1
    local node_name=$2
    
    log_info "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
    log_info "Processing node: $node_name ($host)"
    log_info "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
    echo ""
    
    # Get current VG status
    log_info "Current volume groups:"
    sshpass -p "$PVE_PASS" ssh -o StrictHostKeyChecking=accept-new root@${host} "vgs 2>/dev/null" || true
    echo ""
    
    # Process each storage pool
    for storage in "${STORAGE_POOLS[@]}"; do
        log_info "Processing storage: $storage"
        
        # Get VG and pool names from config (before removal)
        local vg_name=$(get_vg_for_storage "$host" "$storage")
        local pool_name=$(get_thinpool_for_storage "$host" "$storage")
        
        if [ -z "$vg_name" ]; then
            # Use storage name as VG name (common pattern)
            vg_name="$storage"
            pool_name="$storage"
            log_info "Using default: VG=$vg_name, Pool=$pool_name"
        else
            log_info "Found config: VG=$vg_name, Pool=$pool_name"
        fi
        
        # Check if storage is in use
        if ! check_storage_usage "$host" "$storage"; then
            log_warn "Storage $storage may be in use. Proceeding with caution..."
        fi
        
        # Remove storage
        remove_storage "$host" "$storage"
        
        # Recreate storage
        if recreate_storage "$host" "$storage" "$vg_name" "$pool_name" "$node_name"; then
            log_success "Storage $storage repaired on $node_name"
        else
            log_error "Failed to repair storage $storage on $node_name"
        fi
        
        echo ""
    done
    
    # Show final status
    log_info "Final storage status:"
    sshpass -p "$PVE_PASS" ssh -o StrictHostKeyChecking=accept-new root@${host} \
        "pvesm status | grep -E '(thin1|thin2|thin3)'" || true
    echo ""
}

# Main execution
main() {
    echo ""
    log_info "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
    log_info "Repair and Redeploy Thin Storage Pools"
    log_info "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
    echo ""
    
    log_warn "This script will:"
    log_warn "  1. Remove thin1, thin2, thin3 storage configurations"
    log_warn "  2. Recreate them properly"
    log_warn "  3. Ensure they are properly configured"
    echo ""
    
    # Check for non-interactive mode
    if [[ "${NON_INTERACTIVE:-}" == "1" ]] || [[ ! -t 0 ]]; then
        log_info "Non-interactive mode: proceeding automatically"
    else
        read -p "Continue? (y/N): " -n 1 -r
        echo ""
        if [[ ! $REPLY =~ ^[Yy]$ ]]; then
            log_info "Operation cancelled"
            exit 0
        fi
    fi
    
    # Process pve
    if sshpass -p "$PVE_PASS" ssh -o ConnectTimeout=5 -o StrictHostKeyChecking=accept-new root@${PROXMOX_HOST_PVE} "echo 'connected'" 2>/dev/null; then
        process_node "$PROXMOX_HOST_PVE" "pve"
    else
        log_error "Cannot connect to pve"
    fi
    
    # Process pve2
    if sshpass -p "$PVE_PASS" ssh -o ConnectTimeout=5 -o StrictHostKeyChecking=accept-new root@${PROXMOX_HOST_PVE2} "echo 'connected'" 2>/dev/null; then
        process_node "$PROXMOX_HOST_PVE2" "pve2"
    else
        log_warn "Cannot connect to pve2, skipping..."
    fi
    
    echo ""
    log_info "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
    log_info "Repair Complete"
    log_info "━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━"
    echo ""
    log_info "Storage pools have been repaired and redeployed."
    log_info "Verify status with: ssh root@<node> 'pvesm status'"
    echo ""
}

main "$@"