ROMFASTSQL/proxmox/cluster/ups/scripts/ups-shutdown-cluster.sh

#!/bin/bash
#
# Script de shutdown orchestrat pentru cluster Proxmox cand UPS este pe baterie critica
# Trimite notificari email via PVE::Notify pentru fiecare pas
#
# Creat: 2025-10-06
# Actualizat: 2026-01-14 - Fix permisiuni log file, adaugat sudo pentru PVE::Notify

LOGFILE=/var/log/ups-shutdown.log
NODES=("10.0.20.200" "10.0.20.202")  # pve1, pveelite (pvemini va fi ultimul)
NODE_NAMES=("pve1" "pveelite")        # Nume pentru notificari
UPS_NAME="nutdev1"
UPS_USER="admin"
UPS_PASS="parola99"
TEMPLATE_DIR="/etc/pve/notification-templates/default"
HOSTNAME=$(hostname)
FQDN=$(hostname -f 2>/dev/null || hostname)

# Asigura ca fisierul de log exista si are permisiunile corecte
# Scriptul poate rula ca user 'nut', deci fisierul trebuie sa fie writable
ensure_logfile() {
    if [ ! -f "$LOGFILE" ]; then
        touch "$LOGFILE" 2>/dev/null || sudo touch "$LOGFILE"
    fi
    if [ ! -w "$LOGFILE" ]; then
        sudo chown nut:nut "$LOGFILE" 2>/dev/null
        sudo chmod 664 "$LOGFILE" 2>/dev/null
    fi
}

ensure_logfile

log_message() {
    echo "[$(date '+%Y-%m-%d %H:%M:%S')] $1" | tee -a $LOGFILE 2>/dev/null
    logger -t ups-shutdown "$1"
}

# Obtine status UPS
get_ups_info() {
    echo "Status: $(upsc $UPS_NAME ups.status 2>/dev/null || echo 'UNKNOWN')"
    echo "Battery: $(upsc $UPS_NAME battery.charge 2>/dev/null || echo '?')%"
    echo "Input: $(upsc $UPS_NAME input.voltage 2>/dev/null || echo '?')V"
}

# Trimite notificare email via PVE::Notify
send_notification() {
    local EVENT_TYPE="$1"
    local EVENT_TITLE="$2"
    local EVENT_DESC="$3"
    local SEVERITY="$4"

    local UPS_STATUS=$(upsc $UPS_NAME ups.status 2>/dev/null || echo "UNKNOWN")
    local BATTERY_CHARGE=$(upsc $UPS_NAME battery.charge 2>/dev/null || echo "0")
    local INPUT_VOLTAGE=$(upsc $UPS_NAME input.voltage 2>/dev/null || echo "0")
    local EVENT_DATE=$(date '+%Y-%m-%d %H:%M:%S')

    log_message "Sending notification: $EVENT_TITLE"

    /usr/bin/sudo /usr/bin/perl -I/usr/share/perl5 << EOFPERL 2>&1 | tee -a $LOGFILE
use strict;
use warnings;
use PVE::Notify;

my \$template_data = {
    'hostname' => '$FQDN',
    'event_date' => '$EVENT_DATE',
    'event_type' => '$EVENT_TYPE',
    'event_title' => '$EVENT_TITLE',
    'event_description' => '$EVENT_DESC',
    'event_class' => 'shutdown',
    'alert_type' => 'danger',
    'ups_status' => '$UPS_STATUS',
    'battery_charge' => '$BATTERY_CHARGE',
    'input_voltage' => '$INPUT_VOLTAGE',
    'action_taken' => 'Shutdown in curs',
    'next_steps' => ''
};

my \$fields = {
    'hostname' => '$HOSTNAME',
    'type' => 'ups-power-event'
};

eval {
    PVE::Notify::notify('$SEVERITY', 'ups-power-event', \$template_data, \$fields);
    print "Notification sent\\n";
};
if (\$@) {
    print STDERR "Notification failed: \$@\\n";
}
EOFPERL
}

log_message "========================================"
log_message "UPS SHUTDOWN ORCHESTRATION STARTED"
log_message "$(get_ups_info)"
log_message "========================================"

# Verifica daca UPS este intr-adevar pe baterie critica
UPS_STATUS=$(upsc $UPS_NAME ups.status 2>/dev/null)
if [[ ! $UPS_STATUS =~ (OB|LB) ]]; then
    log_message "WARNING: UPS status is $UPS_STATUS - not critical. Aborting shutdown."
    exit 0
fi

# Email: START SHUTDOWN
send_notification \
    "SHUTDOWN_START" \
    "Shutdown cluster PORNIT" \
    "UPS pe baterie critica. Se initiaza oprirea ordonata a cluster-ului Proxmox." \
    "error"

log_message "Step 1: Oprire VM-uri pe toate nodurile..."

# Opreste VM-uri pe toate nodurile (inclusiv local)
for node in ${NODES[@]} localhost; do
    if [ "$node" == "localhost" ]; then
        NODE_NAME="pvemini (local)"
    else
        NODE_NAME=$node
    fi

    log_message "  - Oprire VM-uri pe $NODE_NAME..."

    if [ "$node" == "localhost" ]; then
        for vmid in $(qm list | awk 'NR>1 {print $1}'); do
            vm_status=$(qm status $vmid | awk '{print $2}')
            if [ "$vm_status" == "running" ]; then
                log_message "    * Oprire VM $vmid pe pvemini..."
                qm shutdown $vmid --timeout 60 &
            fi
        done
    else
        ssh -o ConnectTimeout=5 root@$node "
            for vmid in \$(qm list | awk 'NR>1 {print \$1}'); do
                vm_status=\$(qm status \$vmid | awk '{print \$2}')
                if [ \"\$vm_status\" == \"running\" ]; then
                    echo '    * Oprire VM '\$vmid' pe $node...'
                    qm shutdown \$vmid --timeout 60 &
                fi
            done
        " 2>&1 | tee -a $LOGFILE
    fi
done

log_message "Step 2: Oprire containere LXC pe toate nodurile..."

# Opreste containere LXC pe toate nodurile
for node in ${NODES[@]} localhost; do
    if [ "$node" == "localhost" ]; then
        NODE_NAME="pvemini (local)"
    else
        NODE_NAME=$node
    fi

    log_message "  - Oprire LXC pe $NODE_NAME..."

    if [ "$node" == "localhost" ]; then
        pct list 2>/dev/null | awk 'NR>1 && $2=="running" {print $1}' | while read ctid; do
            log_message "    * Oprire container $ctid pe pvemini..."
            pct shutdown $ctid --timeout 60 &
        done
    else
        ssh -o ConnectTimeout=5 root@$node "
            pct list 2>/dev/null | awk 'NR>1 && \$2==\"running\" {print \$1}' | while read ctid; do
                echo '    * Oprire container '\$ctid' pe $node...'
                pct shutdown \$ctid --timeout 60 &
            done
        " 2>&1 | tee -a $LOGFILE
    fi
done

log_message "Step 3: Asteptare 90 secunde pentru oprirea VM-urilor si LXC..."
sleep 90

log_message "Step 4: Oprire noduri secundare..."

# Opreste nodurile secundare si trimite notificare pentru fiecare
for i in "${!NODES[@]}"; do
    node="${NODES[$i]}"
    node_name="${NODE_NAMES[$i]}"

    log_message "  - Shutdown nod $node_name ($node)..."

    # Email: SHUTDOWN NOD SECUNDAR
    send_notification \
        "SHUTDOWN_NODE" \
        "Shutdown $node_name trimis" \
        "Comanda shutdown a fost trimisa catre nodul $node_name ($node)." \
        "error"

    ssh -o ConnectTimeout=5 root@$node "shutdown -h +1 'UPS battery critical - shutting down'" 2>&1 | tee -a $LOGFILE &
done

log_message "Step 5: Asteptare 60 secunde pentru shutdown noduri secundare..."
sleep 60

log_message "Step 6: Oprire nod local (pvemini - primary)..."

# Email: SHUTDOWN NOD PRIMARY (ultimul email inainte de shutdown)
send_notification \
    "SHUTDOWN_PRIMARY" \
    "Shutdown pvemini (ULTIMUL NOD)" \
    "Se opreste nodul primary pvemini. Acesta este ultimul nod din cluster. UPS-ul se va opri dupa shutdown." \
    "error"

log_message "Step 7: Oprire UPS dupa shutdown..."

# Comanda UPS sa se opreasca dupa un delay (permite shutdown-ul sa se finalizeze)
# Verifica daca comanda este disponibila
if upscmd -l $UPS_NAME 2>/dev/null | grep -q "shutdown.stayoff"; then
    log_message "  - Comanda UPS shutdown.stayoff (oprire completa)..."
    upscmd -u $UPS_USER -p $UPS_PASS $UPS_NAME shutdown.stayoff 2>&1 | tee -a $LOGFILE
elif upscmd -l $UPS_NAME 2>/dev/null | grep -q "shutdown.return"; then
    log_message "  - Comanda UPS shutdown.return (oprire cu restart la revenire curent)..."
    upscmd -u $UPS_USER -p $UPS_PASS $UPS_NAME shutdown.return 2>&1 | tee -a $LOGFILE
else
    log_message "  - WARNING: Nu s-a gasit comanda UPS shutdown disponibila"
fi

log_message "========================================"
log_message "UPS SHUTDOWN ORCHESTRATION COMPLETED"
log_message "$(get_ups_info)"
log_message "Local node will shutdown in 1 minute"
log_message "========================================"

# Opreste nodul local (ultimul)
shutdown -h +1 "UPS battery critical - primary node shutting down"

exit 0