#!/usr/bin/env bash
# koopa-firecuda-probe — external ICMP+SSH last-alive for firecuda crashes/hangs.
# Runs as hernani on koopa (~15s user timer). Log is hernani-readable on koopa.
# SSH uses dedicated ed25519 → max only (forced command: cat heartbeat.last).
#
# Usage:
#   koopa-firecuda-probe           # one probe cycle (default; timer path)
#   koopa-firecuda-probe once      # same
#   koopa-firecuda-probe show      # tail last lines
#   koopa-firecuda-probe status    # last line + timer hint
#
# Env:
#   FIRECUDA_PROBE_HOST   default 192.168.100.84
#   FIRECUDA_PROBE_USER   default max
#   FIRECUDA_PROBE_KEY    default ~/.ssh/id_ed25519_firecuda_probe
#   FIRECUDA_PROBE_STATE  default ~/.local/state/firecuda-probe
set +e
set +u

CMD="${1:-once}"
HOST="${FIRECUDA_PROBE_HOST:-192.168.100.84}"
USER_R="${FIRECUDA_PROBE_USER:-max}"
KEY="${FIRECUDA_PROBE_KEY:-${HOME}/.ssh/id_ed25519_firecuda_probe}"
STATE="${FIRECUDA_PROBE_STATE:-${XDG_STATE_HOME:-${HOME}/.local/state}/firecuda-probe}"
LOG="${STATE}/firecuda-probe.log"
LAST="${STATE}/firecuda-probe.last"
META="${STATE}/firecuda-probe.meta"
SSH_TO="${FIRECUDA_PROBE_SSH_TIMEOUT:-5}"
PING_TO="${FIRECUDA_PROBE_PING_TIMEOUT:-2}"

mkdir -p "${STATE}" 2>/dev/null || true

do_show() {
  if [ ! -f "${LOG}" ]; then
    echo "firecuda-probe: no log yet (${LOG})"
    echo "  start: systemctl --user start koopa-firecuda-probe.service"
    return 0
  fi
  tail -n "${1:-20}" "${LOG}"
}

do_status() {
  if [ -f "${LAST}" ]; then
    echo "last: $(cat "${LAST}")"
  else
    echo "last: (none)"
  fi
  if [ -f "${META}" ]; then
    cat "${META}"
  fi
  systemctl --user is-active koopa-firecuda-probe.timer 2>/dev/null \
    | awk '{print "timer=" $0}'
}

do_once() {
  local ts icmp ssh hb gap prev_ts prev_epoch now_epoch
  ts="$(date -Iseconds)"
  now_epoch="$(date +%s)"
  icmp=fail
  ssh=fail
  hb="-"

  if ping -c1 -W "${PING_TO}" "${HOST}" >/dev/null 2>&1; then
    icmp=ok
  fi

  if [ ! -f "${KEY}" ]; then
    ssh=nokey
  else
    # Forced command on firecuda returns heartbeat.last; ignore remote argv.
    hb="$(
      ssh -i "${KEY}" -o IdentitiesOnly=yes -o BatchMode=yes \
        -o StrictHostKeyChecking=accept-new \
        -o ConnectTimeout="${SSH_TO}" -o ConnectionAttempts=1 \
        -o PreferredAuthentications=publickey \
        "${USER_R}@${HOST}" true 2>/dev/null \
        | tr '\n' ' ' | tr -s ' ' | head -c 400
    )"
    if [ -n "${hb}" ]; then
      ssh=ok
    else
      if [ "${icmp}" = "ok" ]; then
        ssh=fail
      else
        ssh=down
      fi
      hb="-"
    fi
  fi

  gap="?"
  if [ -f "${LAST}" ]; then
    prev_ts="$(awk '{print $1}' "${LAST}" 2>/dev/null)"
    if prev_epoch="$(date -d "${prev_ts}" +%s 2>/dev/null)"; then
      gap=$((now_epoch - prev_epoch))
    fi
  fi

  local line
  line="$(printf '%s icmp=%s ssh=%s gap_s=%s host=%s user=%s hb=%s' \
    "${ts}" "${icmp}" "${ssh}" "${gap}" "${HOST}" "${USER_R}" "${hb}")"

  printf '%s\n' "${line}" >>"${LOG}"
  printf '%s\n' "${line}" >"${LAST}"
  {
    echo "updated=${ts}"
    echo "log=${LOG}"
    echo "icmp=${icmp}"
    echo "ssh=${ssh}"
    echo "gap_s=${gap}"
  } >"${META}"

  # Keep log bounded (~14d at 15s ≈ 80k lines; rotate soft at 200k lines)
  if [ -f "${LOG}" ]; then
    local lines
    lines="$(wc -l <"${LOG}" 2>/dev/null | tr -d ' ')"
    if [ "${lines:-0}" -gt 200000 ] 2>/dev/null; then
      tail -n 100000 "${LOG}" >"${LOG}.tmp" && mv "${LOG}.tmp" "${LOG}"
    fi
  fi

  return 0
}

case "${CMD}" in
  once|probe|"") do_once ;;
  show) do_show "${2:-20}" ;;
  status) do_status ;;
  -h|--help)
    sed -n '2,20p' "$0" | sed 's/^# \?//'
    ;;
  *)
    echo "unknown: ${CMD} (once|show|status)" >&2
    exit 2
    ;;
esac