#!/usr/bin/env bash
set -euo pipefail

REMOTE_HOST="${REMOTE_HOST:-root@43.110.128.200}"
REMOTE_CONFIG="${REMOTE_CONFIG:-/root/inter-rp-runners/inter-rp.env}"
REMOTE_SCRIPTS_DIR="${REMOTE_SCRIPTS_DIR:-/root/inter-rp-runners/scripts}"
DEFAULT_RETAIN_RUNS="${RETAIN_RUNS:-20}"
DEFAULT_RSS_SAMPLE_MS="${RSS_SAMPLE_MS:-500}"

usage() {
  cat <<'USAGE'
Usage:
  control_remote200_rp.sh status [rpki-client|routinator|all]
  control_remote200_rp.sh start  <rpki-client|routinator>
  control_remote200_rp.sh stop   <rpki-client|routinator|all>
  control_remote200_rp.sh restart <rpki-client|routinator>

Environment overrides:
  REMOTE_HOST=root@43.110.128.200
  REMOTE_CONFIG=/root/inter-rp-runners/inter-rp.env
  REMOTE_SCRIPTS_DIR=/root/inter-rp-runners/scripts
  RETAIN_RUNS=20
  RSS_SAMPLE_MS=500

Notes:
  - start uses the remote run_single_rp_with_rss.sh loop and runs until stopped.
  - stop only kills the selected RP loop branch and its current child processes.
  - rpki-client and routinator are managed independently.
USAGE
}

ACTION="${1:-status}"
RP="${2:-all}"

case "$ACTION" in
  status|start|stop|restart) ;;
  -h|--help|help)
    usage
    exit 0
    ;;
  *)
    echo "unknown action: $ACTION" >&2
    usage >&2
    exit 2
    ;;
esac

case "$RP" in
  rpki-client|routinator|all) ;;
  *)
    echo "unknown RP: $RP" >&2
    usage >&2
    exit 2
    ;;
esac

if [[ "$ACTION" == "start" || "$ACTION" == "restart" ]] && [[ "$RP" == "all" ]]; then
  echo "start/restart requires one RP: rpki-client or routinator" >&2
  exit 2
fi

ssh "$REMOTE_HOST" \
  "REMOTE_CONFIG='$REMOTE_CONFIG' REMOTE_SCRIPTS_DIR='$REMOTE_SCRIPTS_DIR' RETAIN_RUNS='$DEFAULT_RETAIN_RUNS' RSS_SAMPLE_MS='$DEFAULT_RSS_SAMPLE_MS' ACTION='$ACTION' RP='$RP' bash -s" <<'REMOTE'
set -euo pipefail

load_config() {
  if [[ -f "$REMOTE_CONFIG" ]]; then
    # shellcheck disable=SC1090
    source "$REMOTE_CONFIG"
  fi
  INTER_RP_ROOT="${INTER_RP_ROOT:-/var/lib/inter-rp-runners}"
  RETAIN_RUNS="${RETAIN_RUNS:-20}"
  RSS_SAMPLE_MS="${RSS_SAMPLE_MS:-500}"
  ROUTINATOR_RUN_COMMAND="${ROUTINATOR_RUN_COMMAND:-/root/inter-rp-runners/scripts/run_routinator_once.sh}"
  RPKI_CLIENT_RUN_COMMAND="${RPKI_CLIENT_RUN_COMMAND:-/root/inter-rp-runners/scripts/run_rpki_client_official_98_once.sh}"
}

rp_root_name() {
  case "$1" in
    routinator) echo "routinator" ;;
    rpki-client) echo "rpki-client" ;;
  esac
}

rp_pid_file() {
  echo "$INTER_RP_ROOT/$(rp_root_name "$1").loop.pid"
}

rp_log_file() {
  echo "$INTER_RP_ROOT/$(rp_root_name "$1").loop.log"
}

rp_command() {
  case "$1" in
    routinator) echo "$ROUTINATOR_RUN_COMMAND" ;;
    rpki-client) echo "$RPKI_CLIENT_RUN_COMMAND" ;;
  esac
}

rp_binary_pattern() {
  case "$1" in
    routinator) echo "/root/inter-rp-runners/bin/routinator|[[:space:]]routinator[[:space:]]" ;;
    rpki-client) echo "rpki-client-official-9\\.8|[[:space:]]rpki-client[[:space:]]" ;;
  esac
}

loop_pattern() {
  local rp="$1"
  echo "run_single_rp_with_rss\\.sh --rp $rp "
}

is_running_pid() {
  local pid="${1:-}"
  [[ -n "$pid" ]] && kill -0 "$pid" 2>/dev/null
}

current_loop_pids() {
  local rp="$1"
  pgrep -af "$(loop_pattern "$rp")" 2>/dev/null | awk '{print $1}' || true
}

status_one() {
  local rp="$1"
  local pid_file
  pid_file="$(rp_pid_file "$rp")"
  local pid=""
  [[ -f "$pid_file" ]] && pid="$(cat "$pid_file" 2>/dev/null || true)"
  echo "== $rp =="
  if is_running_pid "$pid"; then
    echo "loop: running pid=$pid"
  else
    local detected
    detected="$(current_loop_pids "$rp" | paste -sd, -)"
    if [[ -n "$detected" ]]; then
      echo "loop: running detected_pids=$detected"
    else
      echo "loop: stopped"
    fi
  fi
  ps -eo pid,ppid,stat,etime,%cpu,%mem,rss,cmd --sort=-%cpu \
    | grep -E "$(loop_pattern "$rp")|$(rp_binary_pattern "$rp")" \
    | grep -v grep \
    | head -20 || true
  local latest="$INTER_RP_ROOT/$(rp_root_name "$rp")/latest/run-meta.json"
  if [[ -f "$latest" ]]; then
    python3 - "$latest" <<'PY'
import json, sys
path = sys.argv[1]
with open(path, "r", encoding="utf-8") as handle:
    meta = json.load(handle)
print("latest:", "runSeq=", meta.get("runSeq"), "success=", meta.get("success"), "wallMs=", meta.get("wallMs"), "vrps=", meta.get("counts", {}).get("vrps"), "vaps=", meta.get("counts", {}).get("vaps"), "rssKb=", meta.get("maxRssKb", {}).get("aggregatePeak"))
PY
  else
    echo "latest: none"
  fi
}

start_one() {
  local rp="$1"
  local pid_file log_file root_name run_command
  pid_file="$(rp_pid_file "$rp")"
  log_file="$(rp_log_file "$rp")"
  root_name="$(rp_root_name "$rp")"
  run_command="$(rp_command "$rp")"
  mkdir -p "$INTER_RP_ROOT/$root_name" "$INTER_RP_ROOT"
  local existing=""
  if [[ -f "$pid_file" ]]; then
    existing="$(cat "$pid_file" 2>/dev/null || true)"
  fi
  if is_running_pid "$existing"; then
    echo "$rp already running pid=$existing"
    return 0
  fi
  local detected
  detected="$(current_loop_pids "$rp" | head -1)"
  if [[ -n "$detected" ]] && is_running_pid "$detected"; then
    echo "$detected" >"$pid_file"
    echo "$rp already running detected pid=$detected"
    return 0
  fi
  nohup bash -lc '
    set -euo pipefail
    while true; do
      "'"$REMOTE_SCRIPTS_DIR"'/run_single_rp_with_rss.sh" \
        --rp "'"$rp"'" \
        --root "'"$INTER_RP_ROOT/$root_name"'" \
        --command "'"$run_command"'" \
        --retain-runs "'"$RETAIN_RUNS"'" \
        --sample-ms "'"$RSS_SAMPLE_MS"'" || true
    done
  ' >"$log_file" 2>&1 &
  local pid="$!"
  echo "$pid" >"$pid_file"
  echo "$rp started pid=$pid log=$log_file"
}

stop_one() {
  local rp="$1"
  local candidates=()
  local pid_file pid
  pid_file="$(rp_pid_file "$rp")"
  if [[ -f "$pid_file" ]]; then
    pid="$(cat "$pid_file" 2>/dev/null || true)"
    [[ -n "$pid" ]] && candidates+=("$pid")
  fi
  while read -r pid; do
    [[ -n "$pid" ]] && candidates+=("$pid")
  done < <(current_loop_pids "$rp")
  while read -r pid; do
    [[ -n "$pid" ]] && candidates+=("$pid")
  done < <(pgrep -af "$(rp_binary_pattern "$rp")" 2>/dev/null | awk '{print $1}' || true)

  if ((${#candidates[@]} == 0)); then
    echo "$rp already stopped"
    rm -f "$pid_file"
    return 0
  fi

  local unique_pids
  unique_pids="$(printf '%s\n' "${candidates[@]}" | awk '!seen[$0]++' | tr '\n' ' ')"
  echo "stopping $rp pids=$unique_pids"
  for pid in $unique_pids; do
    kill -TERM "$pid" 2>/dev/null || true
  done
  sleep 3
  for pid in $unique_pids; do
    kill -KILL "$pid" 2>/dev/null || true
  done
  rm -f "$pid_file"
  echo "$rp stopped"
}

load_config

case "$ACTION:$RP" in
  status:all)
    date -Is
    uptime
    status_one routinator
    status_one rpki-client
    ;;
  status:*)
    date -Is
    uptime
    status_one "$RP"
    ;;
  start:*)
    start_one "$RP"
    status_one "$RP"
    ;;
  stop:all)
    stop_one routinator
    stop_one rpki-client
    ;;
  stop:*)
    stop_one "$RP"
    status_one "$RP"
    ;;
  restart:*)
    stop_one "$RP"
    start_one "$RP"
    status_one "$RP"
    ;;
esac
REMOTE