ducklm/scripts/duck.sh

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

ROOT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"

ENV_KEYS=(
  DUCK_LLAMA_SERVER_BIN
  DUCK_MAIN_MODEL_PATH
  DUCK_MAIN_PORT
  DUCK_CTX_SIZE
  DUCK_N_GPU_LAYERS
  DUCK_LLAMA_DEVICE
  DUCK_PARALLEL
  DUCK_LLAMA_PID_FILE
  DUCK_LLAMA_LOG_FILE
  DUCK_LLAMA_EXTRA_ARGS
  DUCK_HOST
  DUCK_API_HOST
  DUCK_API_PORT
  DUCK_API_PID_FILE
  DUCK_API_LOG_FILE
  DUCK_API_COMMAND
)
declare -A ENV_OVERRIDES=()
for key in "${ENV_KEYS[@]}"; do
  if [[ -v "${key}" ]]; then
    ENV_OVERRIDES["${key}"]="${!key}"
  fi
done

if [[ -f "${ROOT_DIR}/.env" ]]; then
  set -a
  # shellcheck disable=SC1091
  source "${ROOT_DIR}/.env"
  set +a
fi
for key in "${!ENV_OVERRIDES[@]}"; do
  export "${key}=${ENV_OVERRIDES[${key}]}"
done

ACTION="${1:-start}"
API_PID_FILE="${DUCK_API_PID_FILE:-${ROOT_DIR}/data/duck-api.pid}"
API_LOG_FILE="${DUCK_API_LOG_FILE:-${ROOT_DIR}/data/duck-api.log}"
API_URL="http://${DUCK_API_HOST:-127.0.0.1}:${DUCK_API_PORT:-8000}"
LLAMA_SCRIPT="${ROOT_DIR}/scripts/llama/start_main.sh"

usage() {
  cat <<'EOF'
Usage: scripts/duck.sh <command>

Commands:
  start       Start llama-server and DuckLM API in the background
  stop        Stop DuckLM API and managed llama-server
  restart     Stop and start the whole local DuckLM stack
  status      Print process and HTTP health status
              Use "status --probe" to include live model/vector checks
  logs        Show DuckLM API and llama-server logs; use --follow/-f and --lines N
  help        Show this help

Environment:
  DUCK_API_HOST       API bind host, default 127.0.0.1
  DUCK_API_PORT       API port, default 8000
  DUCK_API_PID_FILE   API PID file path
  DUCK_API_LOG_FILE   API log file path
  DUCK_API_COMMAND    API command override, default ".venv/bin/python -m duck_core.api"

llama-server environment is handled by scripts/llama/start_main.sh.
EOF
}

api_is_running() {
  [[ -f "${API_PID_FILE}" ]] || return 1
  local pid
  pid="$(cat "${API_PID_FILE}")"
  [[ "${pid}" =~ ^[0-9]+$ ]] || return 1
  kill -0 "${pid}" 2>/dev/null
}

api_pid_value() {
  if [[ -f "${API_PID_FILE}" ]]; then
    cat "${API_PID_FILE}"
  fi
}

start_api() {
  if api_is_running; then
    echo "DuckLM API already running: pid=$(api_pid_value)"
    return 0
  fi

  mkdir -p "$(dirname "${API_PID_FILE}")" "$(dirname "${API_LOG_FILE}")"
  rm -f "${API_PID_FILE}"

  local command_string="${DUCK_API_COMMAND:-.venv/bin/python -m duck_core.api}"
  echo "Starting DuckLM API..."
  echo "Command: ${command_string}" >> "${API_LOG_FILE}"
  if command -v setsid >/dev/null 2>&1; then
    (
      cd "${ROOT_DIR}"
      nohup setsid bash -lc "${command_string}" >> "${API_LOG_FILE}" 2>&1 < /dev/null &
      echo "$!" > "${API_PID_FILE}"
    )
  else
    (
      cd "${ROOT_DIR}"
      nohup bash -lc "${command_string}" >> "${API_LOG_FILE}" 2>&1 < /dev/null &
      echo "$!" > "${API_PID_FILE}"
    )
  fi

  sleep 0.2
  if api_is_running; then
    echo "DuckLM API started: pid=$(api_pid_value)"
    echo "WebChat: ${API_URL}/"
    echo "Log: ${API_LOG_FILE}"
    for _ in {1..20}; do
      if command -v curl >/dev/null 2>&1 && curl --noproxy "*" -fsS "${API_URL}/health" >/dev/null 2>&1; then
        echo "DuckLM API health: ok (${API_URL})"
        return 0
      fi
      sleep 0.25
    done
    echo "DuckLM API health: not ready yet (${API_URL})"
    return 0
  fi

  echo "DuckLM API failed to start. See ${API_LOG_FILE}" >&2
  rm -f "${API_PID_FILE}"
  return 1
}

stop_api() {
  if ! api_is_running; then
    rm -f "${API_PID_FILE}"
    echo "DuckLM API not running"
    return 0
  fi

  local pid
  pid="$(api_pid_value)"
  echo "Stopping DuckLM API: pid=${pid}"
  kill "${pid}" 2>/dev/null || true

  for _ in {1..30}; do
    if ! kill -0 "${pid}" 2>/dev/null; then
      rm -f "${API_PID_FILE}"
      echo "DuckLM API stopped"
      return 0
    fi
    sleep 0.2
  done

  echo "DuckLM API did not stop after SIGTERM; sending SIGKILL"
  kill -9 "${pid}" 2>/dev/null || true
  rm -f "${API_PID_FILE}"
  echo "DuckLM API stopped"
}

start_stack() {
  "${LLAMA_SCRIPT}" start
  start_api
  echo
  echo "Status:"
  status_stack
}

stop_stack() {
  stop_api
  "${LLAMA_SCRIPT}" stop
}

status_stack() {
  local probe=0
  while [[ $# -gt 0 ]]; do
    case "$1" in
      --probe)
        probe=1
        shift
        ;;
      *)
        echo "Unknown status argument: $1" >&2
        return 2
        ;;
    esac
  done

  local rc=0
  if api_is_running; then
    local pid
    pid="$(api_pid_value)"
    echo "DuckLM API running: pid=${pid}"
    if command -v curl >/dev/null 2>&1 && curl --noproxy "*" -fsS "${API_URL}/health" >/dev/null 2>&1; then
      echo "DuckLM API health: ok (${API_URL})"
      print_runtime_status "${probe}"
    else
      echo "DuckLM API health: not ready (${API_URL})"
    fi
  else
    if [[ -f "${API_PID_FILE}" ]]; then
      echo "DuckLM API not running; removing stale pid file ${API_PID_FILE}"
      rm -f "${API_PID_FILE}"
    else
      echo "DuckLM API not running"
    fi
    rc=3
  fi

  local llama_rc=0
  "${LLAMA_SCRIPT}" status || llama_rc=$?
  if [[ "${llama_rc}" != "0" && "${rc}" == "0" ]]; then
    rc="${llama_rc}"
  fi
  return "${rc}"
}

print_runtime_status() {
  local probe="${1:-0}"
  local status_url="${API_URL}/v1/status"
  if [[ "${probe}" == "1" ]]; then
    status_url="${status_url}?probe=true"
  fi
  local payload
  if ! payload="$(curl --noproxy "*" -fsS "${status_url}" 2>/dev/null)"; then
    echo "DuckLM runtime status: unavailable (${status_url})"
    return 0
  fi

  local python_bin="${DUCK_PYTHON_BIN:-${ROOT_DIR}/.venv/bin/python}"
  if [[ ! -x "${python_bin}" ]]; then
    python_bin="python3"
  fi
  if ! command -v "${python_bin}" >/dev/null 2>&1 && [[ ! -x "${python_bin}" ]]; then
    echo "DuckLM runtime status: ${status_url}"
    return 0
  fi

  printf '%s' "${payload}" | "${python_bin}" -c '
import json
import sys

data = json.load(sys.stdin)
models = data.get("models", {})
roles = sorted((models.get("roles") or {}).keys())
services = data.get("services", {})
llama = services.get("llama", {})
vector = services.get("vector_memory", {})

def service_line(name, service):
    if not service.get("probed"):
        return f"{name}: not probed"
    ok = "ok" if service.get("ok") else "failed"
    error = service.get("error")
    return f"{name}: {ok}" + (f" ({error})" if error else "")

print("DuckLM runtime:")
print("  workspace: {}".format(data.get("workspace")))
print("  db: {}".format(data.get("db_path")))
print("  model endpoints: {}".format(len(models.get("endpoints") or [])))
print("  roles: {}".format(", ".join(roles)))
print("  {}".format(service_line("llama", llama)))
print("  {}".format(service_line("vector memory", vector)))
'
}

logs_stack() {
  local follow=0
  local lines=100
  shift || true
  while [[ $# -gt 0 ]]; do
    case "$1" in
      -f|--follow)
        follow=1
        shift
        ;;
      --lines)
        lines="${2:?--lines requires a value}"
        shift 2
        ;;
      *)
        echo "Unknown logs argument: $1" >&2
        return 2
        ;;
    esac
  done

  mkdir -p "$(dirname "${API_LOG_FILE}")"
  touch "${API_LOG_FILE}"
  if [[ "${follow}" == "1" ]]; then
    local llama_log_file="${DUCK_LLAMA_LOG_FILE:-${ROOT_DIR}/data/llama-main.log}"
    mkdir -p "$(dirname "${llama_log_file}")"
    touch "${llama_log_file}"
    tail -n "${lines}" -f "${API_LOG_FILE}" "${llama_log_file}"
  else
    echo "==> DuckLM API log: ${API_LOG_FILE} <=="
    tail -n "${lines}" "${API_LOG_FILE}"
    echo
    echo "==> llama-server log: ${DUCK_LLAMA_LOG_FILE:-${ROOT_DIR}/data/llama-main.log} <=="
    "${LLAMA_SCRIPT}" logs --lines "${lines}"
  fi
}

case "${ACTION}" in
  start)
    start_stack
    ;;
  stop)
    stop_stack
    ;;
  restart)
    stop_stack
    start_stack
    ;;
  status)
    shift || true
    status_stack "$@"
    ;;
  logs)
    logs_stack "$@"
    ;;
  help|-h|--help)
    usage
    ;;
  *)
    echo "Unknown command: ${ACTION}" >&2
    usage >&2
    exit 2
    ;;
esac