#!/bin/bash

###############################################################################
# NUT UPS Monitoring Script with Advanced Communication Loss Protection
# Version: 2.1
# 
# Features:
# - Prevents shutdown on communication loss with all UPS units
# - Respects MINSUPPLIES parameter from upsmon.conf
# - Differentiates between power loss and communication issues
# - Comprehensive logging and status tracking
# - Supports calibration mode (CAL) as safe state
###############################################################################

### Configuration Section - Adjust these parameters as needed ###
UPSMON_CONF="/etc/nut/upsmon.conf"       # Path to NUT main configuration
LOG_IDENT="NUT-UPS-Monitor"              # Syslog identification tag
SAFE_UPS_STATES="OL CAL"                 # States considered operational (space-separated)
COMM_RETRY_INTERVAL=30                   # Seconds between communication retries
MAX_COMM_FAILURES=3                      # Allowed consecutive communication failures

### Initialization Section ###
# Read MINSUPPLIES from config with fallback to default value 1
MINSUPPLIES=$(grep -oP '^MINSUPPLIES\s+\K\d+' "$UPSMON_CONF" 2>/dev/null || echo 1)

# Get list of all configured UPS devices (filtering out SSL messages)
UPS_LIST=$(upsc -l 2>/dev/null | grep -v '^Init SSL')
TOTAL_UPS=$(echo "$UPS_LIST" | wc -w)

### Status Tracking Variables ###
SAFE_UPS_COUNT=0                         # Count of UPS in safe states
COMM_FAILURE_COUNT=0                     # Count of consecutive communication failures
declare -a UPS_STATUSES                  # Array to store detailed UPS states

###############################################################################
# Function: check_ups_status
# Description: Checks and logs the status of a specific UPS unit
# Arguments:
#   $1 - UPS identifier
###############################################################################
check_ups_status() {
    local ups_id=$1
    local status
    
    # Get UPS status with error handling
    status=$(upsc "$ups_id" 2>/dev/null | grep -oP 'ups.status:\s*\K\w+')
    
    # Handle communication failures
    if [ -z "$status" ]; then
        status="NOCOMM"
        ((COMM_FAILURE_COUNT++))
        logger -t "$LOG_IDENT" -p warn "Communication failed with UPS $ups_id (Attempt $COMM_FAILURE_COUNT/$MAX_COMM_FAILURES)"
    else
        # Reset communication failure counter on successful contact
        COMM_FAILURE_COUNT=0
    fi
    
    # Check against safe states
    for safe_state in $SAFE_UPS_STATES; do
        if [[ "$status" == "$safe_state" ]]; then
            ((SAFE_UPS_COUNT++))
            logger -t "$LOG_IDENT" "UPS $ups_id operational (Status: $status)"
            break
        fi
    done
    
    # Store detailed status for reporting
    UPS_STATUSES+=("$ups_id:$status")
}

###############################################################################
# Main Event Handler
# Processes ups-on-battery events and makes shutdown decisions
###############################################################################
case $1 in
    ups-on-battery)
        ### Check all UPS units
        for ups in $UPS_LIST; do
            check_ups_status "$ups"
        done

        ### Communication Failure Protection ###
        if [ $COMM_FAILURE_COUNT -ge $MAX_COMM_FAILURES ]; then
            logger -t "$LOG_IDENT" -p err \
                "CRITICAL: Persistent communication failures with $COMM_FAILURE_COUNT UPS units"
            logger -t "$LOG_IDENT" -p err \
                "Possible network/SNMP issue. NOT initiating shutdown."
            exit 2
        fi

        ### Power State Evaluation ###
        if [ "$SAFE_UPS_COUNT" -lt "$MINSUPPLIES" ]; then
            # Genuine power failure condition
            logger -t "$LOG_IDENT" -p crit \
                "POWER FAILURE: Only $SAFE_UPS_COUNT/$TOTAL_UPS UPS operational (Required: $MINSUPPLIES)"
            logger -t "$LOG_IDENT" -p crit \
                "Detailed status: ${UPS_STATUSES[*]}"
            logger -t "$LOG_IDENT" "Initiating controlled shutdown..."
            
            # Create powerdown flag file
            touch /etc/nut/killpower 2>/dev/null
            
            # Execute shutdown command
            /usr/sbin/upsmon -c fsd
        else
            # Adequate power protection available
            logger -t "$LOG_IDENT" \
                "Power stable: $SAFE_UPS_COUNT/$TOTAL_UPS UPS operational (Required: $MINSUPPLIES)"
        fi
        ;;
        
    *)
        # Unknown event type handler
        logger -t "$LOG_IDENT" -p warn "Received unconfigured event: $1"
        exit 1
        ;;
esac

exit 0