proxmox/scripts/archive/consolidated/fix/fix-pve2-hook-issues-proper.sh

#!/usr/bin/env bash
# Fix pre-start hook issues by checking what the hook actually needs
# The hook might be failing due to missing volumes or other issues
# Usage: ./scripts/fix-pve2-hook-issues-proper.sh

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


NODE_IP="${PROXMOX_HOST_R630_01}"
NODE_NAME=$(ssh -o ConnectTimeout=5 -o StrictHostKeyChecking=no root@${NODE_IP} "hostname" || echo "unknown")

# 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"; }

echo ""
log_info "Fixing hook issues on $NODE_NAME ($NODE_IP)..."
echo ""

# All containers with hook issues
HOOK_ISSUE_CONTAINERS=(3000 3001 3002 3003 3500 3501 6000 6400 10000 10001 10020 10030 10040 10050 10060 10070 10080 10090 10091 10092 10100 10101 10120 10130 10150 10151 10200 10201 10202 10210 10230 10232)

FIXED=0
FAILED=0

for vmid in "${HOOK_ISSUE_CONTAINERS[@]}"; do
    log_info "Processing CT $vmid..."
    
    # Get container config
    config_file="/etc/pve/nodes/${NODE_NAME}/lxc/${vmid}.conf"
    rootfs=$(ssh -o ConnectTimeout=5 -o StrictHostKeyChecking=no root@${NODE_IP} \
        "pct config $vmid 2>/dev/null | grep '^rootfs:'" || echo "")
    
    if [[ -z "$rootfs" ]]; then
        log_warn "  Config not found, skipping..."
        continue
    fi
    
    # Extract volume name
    volume_name=$(echo "$rootfs" | sed 's/^rootfs: //' | cut -d':' -f2 | cut -d',' -f1)
    storage_pool=$(echo "$rootfs" | sed 's/^rootfs: //' | cut -d':' -f1)
    
    log_info "  Storage: $storage_pool:$volume_name"
    
    # Check if volume exists
    volume_exists=$(ssh -o ConnectTimeout=5 -o StrictHostKeyChecking=no root@${NODE_IP} \
        "lvs 2>/dev/null | grep -q \"^${volume_name}\" && echo 'exists' || echo 'missing'" || echo "error")
    
    if [[ "$volume_exists" == "missing" ]]; then
        log_warn "  Volume missing, attempting to start (will create volume)..."
        # Try to start - Proxmox will create the volume
        if ssh -o ConnectTimeout=5 -o StrictHostKeyChecking=no root@${NODE_IP} \
            "pct start $vmid" 2>&1 >/dev/null; then
            log_success "  ✓ Container started (volume created)"
            ((FIXED++))
            sleep 1
            continue
        else
            error=$(ssh -o ConnectTimeout=5 -o StrictHostKeyChecking=no root@${NODE_IP} \
                "pct start $vmid 2>&1" || true)
            
            # Check if it's still a hook error
            if echo "$error" | grep -q "hook.pre-start"; then
                log_info "  Hook error detected, checking hook script..."
                
                # Check the pre-start hook
                hook_output=$(ssh -o ConnectTimeout=5 -o StrictHostKeyChecking=no root@${NODE_IP} \
                    "bash -x /usr/share/lxc/hooks/lxc-pve-prestart-hook lxc 3000 start 2>&1" || true)
                
                # Try bypassing hook by checking container directly
                log_info "  Attempting direct start with debug..."
                debug_output=$(ssh -o ConnectTimeout=5 -o StrictHostKeyChecking=no root@${NODE_IP} \
                    "lxc-start -n $vmid -F -l DEBUG -o /tmp/lxc-$vmid.log 2>&1" || true)
                
                # For now, just try starting - the hook might work on retry
                log_info "  Retrying start..."
                if ssh -o ConnectTimeout=5 -o StrictHostKeyChecking=no root@${NODE_IP} \
                    "pct start $vmid" 2>&1 >/dev/null; then
                    log_success "  ✓ Container started on retry"
                    ((FIXED++))
                else
                    log_warn "  ⚠️  Still failing - may need manual intervention"
                    ((FAILED++))
                fi
            else
                log_warn "  ⚠️  Other error:"
                echo "$error" | sed 's/^/    /' | head -2
                ((FAILED++))
            fi
        fi
    else
        log_info "  Volume exists, attempting start..."
        if ssh -o ConnectTimeout=5 -o StrictHostKeyChecking=no root@${NODE_IP} \
            "pct start $vmid" 2>&1 >/dev/null; then
            log_success "  ✓ Container started"
            ((FIXED++))
            sleep 1
        else
            error=$(ssh -o ConnectTimeout=5 -o StrictHostKeyChecking=no root@${NODE_IP} \
                "pct start $vmid 2>&1" || true)
            
            if echo "$error" | grep -q "already running"; then
                log_success "  ✓ Container already running"
                ((FIXED++))
            elif echo "$error" | grep -q "hook.pre-start"; then
                log_warn "  ⚠️  Hook error - container may need manual fix"
                ((FAILED++))
            else
                log_warn "  ⚠️  Start failed:"
                echo "$error" | sed 's/^/    /' | head -2
                ((FAILED++))
            fi
        fi
    fi
    
    echo ""
done

echo ""
log_info "Summary: Fixed: $FIXED, Failed: $FAILED"
log_success "Done!"