portfolio/infra-run/scripts/bash/incident-checks/check_dns_connectivity.sh

#!/usr/bin/env bash
set -o errexit
set -o nounset
set -o pipefail

host_name=""
port=""
count=3
timeout_seconds=3

usage() {
  cat <<'USAGE'
Usage: check_dns_connectivity.sh --host HOST [--port PORT] [--count COUNT] [--timeout SECONDS] [--help]

Check DNS resolution, ping, optional TCP connectivity, and local route hints.
USAGE
}

is_number() {
  [[ "$1" =~ ^[0-9]+$ ]]
}

while (($# > 0)); do
  case "$1" in
    --host) [[ $# -ge 2 ]] || { printf 'CRITICAL: --host requires a value\n'; exit 2; }; host_name="$2"; shift 2 ;;
    --port) [[ $# -ge 2 ]] || { printf 'CRITICAL: --port requires a value\n'; exit 2; }; port="$2"; shift 2 ;;
    --count) [[ $# -ge 2 ]] || { printf 'CRITICAL: --count requires a value\n'; exit 2; }; count="$2"; shift 2 ;;
    --timeout) [[ $# -ge 2 ]] || { printf 'CRITICAL: --timeout requires a value\n'; exit 2; }; timeout_seconds="$2"; shift 2 ;;
    --help|-h) usage; exit 0 ;;
    *) printf 'CRITICAL: unknown option: %s\n' "$1"; usage; exit 2 ;;
  esac
done

if [[ -z "$host_name" ]]; then
  printf 'CRITICAL: --host is required\n'
  usage
  exit 2
fi
for value in "$count" "$timeout_seconds"; do
  if ! is_number "$value"; then
    printf 'CRITICAL: numeric option expected, got: %s\n' "$value"
    exit 2
  fi
done
if [[ -n "$port" ]] && ! is_number "$port"; then
  printf 'CRITICAL: --port must be numeric\n'
  exit 2
fi
if ! command -v getent >/dev/null 2>&1; then
  printf 'CRITICAL: required command not found: getent\n'
  exit 2
fi

dns_ok=0
ping_ok=0
tcp_ok=0
tcp_checked=0
tcp_note=""
ping_output="$(mktemp)"
trap 'rm -f "$ping_output"' EXIT

dns_output="$(getent hosts "$host_name" 2>/dev/null || true)"
if [[ -n "$dns_output" ]]; then
  dns_ok=1
fi

if command -v ping >/dev/null 2>&1; then
  if ping -c "$count" -W "$timeout_seconds" "$host_name" > "$ping_output" 2>&1; then
    ping_ok=1
  fi
else
  printf 'WARNING: ping command not available; ICMP check skipped\n' > "$ping_output"
fi

if [[ -n "$port" ]]; then
  tcp_checked=1
  if command -v timeout >/dev/null 2>&1; then
    if timeout "$timeout_seconds" bash -c ":</dev/tcp/${host_name}/${port}" >/dev/null 2>&1; then
      tcp_ok=1
    fi
  else
    tcp_note="WARNING: timeout command not available; TCP /dev/tcp check used without external timeout"
    if bash -c ":</dev/tcp/${host_name}/${port}" >/dev/null 2>&1; then
      tcp_ok=1
    fi
  fi
fi

status="OK"
exit_code=0
if ((dns_ok == 0)); then
  status="CRITICAL"
  exit_code=3
elif ((tcp_checked == 1 && tcp_ok == 0)); then
  status="CRITICAL"
  exit_code=3
elif command -v ping >/dev/null 2>&1 && ((ping_ok == 0)); then
  status="WARNING"
  exit_code=1
fi

printf '%s: DNS=%s ping=%s' "$status" "$([[ "$dns_ok" == 1 ]] && printf OK || printf FAILED)" "$([[ "$ping_ok" == 1 ]] && printf OK || printf UNKNOWN_OR_FAILED)"
if ((tcp_checked == 1)); then
  printf ' tcp_%s=%s' "$port" "$([[ "$tcp_ok" == 1 ]] && printf OK || printf FAILED)"
fi
printf '\n\n'

printf 'DNS result:\n'
if [[ -n "$dns_output" ]]; then
  printf '%s\n' "$dns_output"
else
  printf 'CRITICAL: getent hosts returned no records for %s\n' "$host_name"
fi
printf '\n'

printf 'Ping result:\n'
if [[ -s "$ping_output" ]]; then
  cat "$ping_output"
else
  printf 'WARNING: ping result unavailable or ping command missing\n'
fi
printf '\n'

if ((tcp_checked == 1)); then
  printf 'TCP port result:\n'
  if ((tcp_ok == 1)); then
    printf 'OK: TCP connection to %s:%s succeeded\n' "$host_name" "$port"
  else
    printf 'CRITICAL: TCP connection to %s:%s failed or timed out\n' "$host_name" "$port"
  fi
  if [[ -n "$tcp_note" ]]; then
    printf '%s\n' "$tcp_note"
  fi
  printf '\n'
fi

printf 'Local network hints:\n'
if command -v ip >/dev/null 2>&1; then
  ip route show default 2>/dev/null || printf 'WARNING: unable to read default route\n'
elif command -v ss >/dev/null 2>&1; then
  ss -tuln 2>/dev/null | head -n 20 || printf 'WARNING: unable to read socket summary\n'
else
  printf 'WARNING: ip and ss are unavailable; local network hints skipped\n'
fi
printf '\n'

printf 'Evidence:\n'
printf 'Host: %s count=%s timeout=%ss port=%s\n' "$host_name" "$count" "$timeout_seconds" "${port:-not checked}"
if [[ -n "$tcp_note" ]]; then
  printf '%s\n' "$tcp_note"
fi
printf '\n'

printf 'Recommended next steps:\n'
printf -- '- Verify the DNS record and resolver path\n'
printf -- '- Check firewall, routing, security group, or proxy policy\n'
printf -- '- Compare results from another host or network segment\n'
printf -- '- Check application endpoint health after network reachability is confirmed\n'
printf -- '- Attach this output to incident ticket\n'

exit "$exit_code"