#!/bin/bash

PATH="/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin"

THRESHOLD_DEFAULT=200
CONFIG_FILE="/etc/exim-limits-alert"
EMAIL_RECIPIENT="sysadmin@sitioshispanos.com"
LOCK_FILE="/var/lock/check-exim-alerts.lock"
LOG_FILE="/var/log/check-exim-alerts.log"

# Hosts that will not report.
# Examples:
# EXCLUDE="dns"
# EXCLUDE="linux{58,{91..95},97} dns"
EXCLUDE="dns"
DOMAIN_SUFFIX="webhosting-network-services.com"

# Do not include the entire queue in the email if it is huge.
QUEUE_DETAILS_LIMIT=300

log() {
    echo "$(date '+%F %T') $*" >> "$LOG_FILE"
}

command_exists() {
    command -v "$1" >/dev/null 2>&1
}

count_lines() {
    grep -c . 2>/dev/null || true
}

send_alert() {
    local subject="$1"
    local body="$2"

    if command_exists mail; then
        printf "%s\n" "$body" | mail -s "$subject" "$EMAIL_RECIPIENT"
    elif command_exists mailx; then
        printf "%s\n" "$body" | mailx -s "$subject" "$EMAIL_RECIPIENT"
    else
        log "ERROR: mail/mailx not found. Could not send alert."
        return 1
    fi
}

# Prevent overlapping runs.
exec 9>"$LOCK_FILE"
if ! flock -n 9; then
    log "Another instance is already running. Exiting."
    exit 0
fi

HOSTNAME_SHORT="$(hostname -s 2>/dev/null || hostname)"
HOSTNAME_FQDN="$(hostname -f 2>/dev/null || hostname)"

THRESHOLD="$THRESHOLD_DEFAULT"

if [ -f "$CONFIG_FILE" ]; then
    CONFIG_THRESHOLD="$(grep -E '^[[:space:]]*[0-9]+[[:space:]]*$' "$CONFIG_FILE" | head -n 1 | tr -d '[:space:]')"
    if [ -n "$CONFIG_THRESHOLD" ]; then
        THRESHOLD="$CONFIG_THRESHOLD"
    else
        log "Invalid threshold in $CONFIG_FILE. Using default: $THRESHOLD_DEFAULT"
    fi
fi

if ! [[ "$THRESHOLD" =~ ^[0-9]+$ ]]; then
    log "Invalid threshold value '$THRESHOLD'. Using default: $THRESHOLD_DEFAULT"
    THRESHOLD="$THRESHOLD_DEFAULT"
fi

# Exclude matching short hostname or FQDN.
for excluded in $(eval echo "$EXCLUDE"); do
    excluded_short="$excluded"
    excluded_fqdn="$excluded.$DOMAIN_SUFFIX"

    if [ "$HOSTNAME_SHORT" = "$excluded_short" ] || [ "$HOSTNAME_FQDN" = "$excluded_fqdn" ] || [ "$HOSTNAME_FQDN" = "$excluded_short" ]; then
        log "Host excluded: short=$HOSTNAME_SHORT fqdn=$HOSTNAME_FQDN"
        exit 0
    fi
done

if ! command_exists exim; then
    log "ERROR: exim command not found."
    exit 1
fi

if ! command_exists exiqgrep; then
    REPORT="ERROR: exiqgrep command not found on $HOSTNAME_FQDN.

This script requires exiqgrep to safely detect frozen and old Exim messages."

    send_alert "Alert: Exim check failed on $HOSTNAME_SHORT" "$REPORT"
    exit 1
fi

QUEUE_SIZE="$(exim -bpc 2>/dev/null | tr -d '[:space:]')"

if ! [[ "$QUEUE_SIZE" =~ ^[0-9]+$ ]]; then
    log "ERROR: invalid queue size returned by exim -bpc: '$QUEUE_SIZE'"
    REPORT="ERROR: Could not get a valid Exim queue size on $HOSTNAME_FQDN.

exim -bpc returned: $QUEUE_SIZE"

    send_alert "Alert: Exim queue check failed on $HOSTNAME_SHORT" "$REPORT"
    exit 1
fi

log "Queue size on $HOSTNAME_FQDN: $QUEUE_SIZE. Threshold: $THRESHOLD."

if [ "$QUEUE_SIZE" -le "$THRESHOLD" ]; then
    exit 0
fi

QUEUE_DETAILS="$(exim -bp 2>/dev/null | head -n "$QUEUE_DETAILS_LIMIT")"
TOTAL_QUEUE_LINES="$(exim -bp 2>/dev/null | wc -l | tr -d '[:space:]')"

FROZEN_IDS="$(exiqgrep -z -i 2>/dev/null | sort -u)"
OLD_IDS="$(exiqgrep -o 86400 -i 2>/dev/null | sort -u)"

FROZEN_COUNT="$(printf "%s\n" "$FROZEN_IDS" | count_lines)"
OLD_COUNT="$(printf "%s\n" "$OLD_IDS" | count_lines)"

REMOVED_FROZEN_COUNT=0
REMOVED_OLD_COUNT=0

if [ "$FROZEN_COUNT" -gt 0 ]; then
    if printf "%s\n" "$FROZEN_IDS" | xargs -r -n 100 exim -Mrm >/dev/null 2>&1; then
        REMOVED_FROZEN_COUNT="$FROZEN_COUNT"
        FROZEN_REMOVAL_SUMMARY="$REMOVED_FROZEN_COUNT frozen emails were removed."
        log "$REMOVED_FROZEN_COUNT frozen emails removed."
    else
        FROZEN_REMOVAL_SUMMARY="Failed while removing frozen emails. Check $LOG_FILE."
        log "ERROR: failed removing frozen emails."
    fi
else
    FROZEN_REMOVAL_SUMMARY="No frozen emails were found."
fi

# Recalculate old IDs after frozen removals to avoid trying to delete already removed IDs.
OLD_IDS="$(exiqgrep -o 86400 -i 2>/dev/null | sort -u)"
OLD_COUNT="$(printf "%s\n" "$OLD_IDS" | count_lines)"

if [ "$OLD_COUNT" -gt 0 ]; then
    if printf "%s\n" "$OLD_IDS" | xargs -r -n 100 exim -Mrm >/dev/null 2>&1; then
        REMOVED_OLD_COUNT="$OLD_COUNT"
        OLD_REMOVAL_SUMMARY="$REMOVED_OLD_COUNT emails older than 24 hours were removed."
        log "$REMOVED_OLD_COUNT old emails removed."
    else
        OLD_REMOVAL_SUMMARY="Failed while removing emails older than 24 hours. Check $LOG_FILE."
        log "ERROR: failed removing old emails."
    fi
else
    OLD_REMOVAL_SUMMARY="No emails older than 24 hours were found."
fi

QUEUE_SIZE_AFTER="$(exim -bpc 2>/dev/null | tr -d '[:space:]')"

REPORT="ALERT: The Exim queue on $HOSTNAME_FQDN has $QUEUE_SIZE emails, exceeding the threshold of $THRESHOLD.

Queue after cleanup: $QUEUE_SIZE_AFTER

--- Removed Emails Summary ---
$FROZEN_REMOVAL_SUMMARY
$OLD_REMOVAL_SUMMARY

--- Mail Queue Details ---
Showing first $QUEUE_DETAILS_LIMIT lines out of approximately $TOTAL_QUEUE_LINES queue output lines.

$QUEUE_DETAILS
"

send_alert "Alert: Exim Queue on $HOSTNAME_SHORT" "$REPORT"

exit 0