feat(eval): internal-eval stack, judge fixes, and eval web overhaul

- Merge internal-eval into development.sh (single Celery/infra); internal-eval.sh
  wraps with LIFE_ECHO_WITH_INTERNAL_EVAL; EVAL_ATTACH_ONLY for attaching 8001
  when :8000 is already up; document in api/docs/internal-eval.md.
- Evaluation: transcript_for_judge, judge error surfacing, rubric/schema tweaks,
  execution_service and router updates; tests for judge and composite eval.
- Memory: ingest nested transaction for embedding/enrichment rollback safety.
- Conversation WS: logger.exception for pipeline errors (avoid loguru KeyError).
- app-eval-web: Playground saved replays, dialogue turns helper, hash user_id
  for Memoir; Memoir chapter baseline↔DB row compare with title heuristics;
  Stories page (#memoir-stories); Markdown + copy buttons; toolbar/panel UI;
  react-markdown; development proxy and fixture updates.
This commit is contained in:
Kevin
2026-04-07 17:15:01 +08:00
parent a50b72e7b5
commit 99543d04c6
47 changed files with 4968 additions and 1279 deletions

View File

@@ -1,517 +1,27 @@
#!/usr/bin/env bash
# 仅启动「内部回归评测」栈app/internal_main.py不启动主站 consumer API
#
# 与 development.sh 的区别:
# - development.shmain:app + Celery通常 :8000面向 App/主业务。
# - internal-eval.shinternal_app + Celery:8001仅评测/回放/GLM 打分/门禁。
# 二者共用数据库与 Redis不会拉起第二份 main:app。
#
# 若本机已在跑 ./development.sh只想多开评测 HTTP推荐避免第二套 worker/docker
# SKIP_INFRA=1 SKIP_INSTALL=1 SKIP_CELERY=1 ./internal-eval.sh
# 在 development.sh 全栈之上附加 internal_main默认 :8001与 app-eval-web。
# 只需一条命令,无需再并行跑两份脚本;共用同一份 Postgres/Redis/Celery本脚本只起一个 Worker
#
# 用法cd api && ./internal-eval.sh
# 可选环境变量:
# SKIP_INFRA=1 已起好 Postgres/Redis 时跳过 docker compose
# SKIP_INSTALL=1 跳过 uv sync
# SKIP_CELERY=1 仅起内部 API别处已有 Celery worker 时)
# START_EVAL_WEB=0 不起评测前端(默认会起 app-eval-web需已 npm install
# OPEN_EVAL_WEB=0 起前端但不自动打开浏览器(默认 Vite --open
# EVAL_WEB_PORT 打印提示用,默认 5174与 app-eval-web/vite.config.ts 一致)
# INTERNAL_EVAL_PORT 默认 8001
# CELERY_POOL 默认 solo与 development.sh 一致)
#
# 若主站已在其他终端由 ./development.sh 占用 :8000仅多开评测 HTTP + 前端(不再起第二份 Celery
# SKIP_INFRA=1 SKIP_INSTALL=1 EVAL_ATTACH_ONLY=1 ./internal-eval.sh
#
# 兼容旧环境变量SKIP_CELERY=1 等价于 EVAL_ATTACH_ONLY=1仍要求 :8000 已有监听)。
#
# 其他可选变量与 development.sh 一致,例如:
# SKIP_INFRA=1 SKIP_INSTALL=1 START_EVAL_WEB=0 OPEN_EVAL_WEB=0
# INTERNAL_EVAL_PORT EVAL_WEB_PORT INTERNAL_EVAL_API_KEY
set -euo pipefail
RED='\033[0;31m'
GREEN='\033[0;32m'
YELLOW='\033[1;33m'
BLUE='\033[0;34m'
NC='\033[0m'
ROOT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
REPO_ROOT="$(cd "${ROOT_DIR}/.." && pwd)"
EVAL_WEB_DIR="${REPO_ROOT}/app-eval-web"
VENV_DIR="${ROOT_DIR}/.venv"
UVICORN_BIN="${VENV_DIR}/bin/uvicorn"
CELERY_BIN="${VENV_DIR}/bin/celery"
export LIFE_ECHO_WITH_INTERNAL_EVAL=1
INTERNAL_EVAL_HOST="${INTERNAL_EVAL_HOST:-0.0.0.0}"
INTERNAL_EVAL_PORT="${INTERNAL_EVAL_PORT:-8001}"
CELERY_POOL="${CELERY_POOL:-solo}"
SKIP_INSTALL="${SKIP_INSTALL:-0}"
SKIP_INFRA="${SKIP_INFRA:-0}"
SKIP_CELERY="${SKIP_CELERY:-0}"
START_EVAL_WEB="${START_EVAL_WEB:-1}"
OPEN_EVAL_WEB="${OPEN_EVAL_WEB:-1}"
EVAL_WEB_PORT="${EVAL_WEB_PORT:-5174}"
SHUTDOWN_TIMEOUT="${SHUTDOWN_TIMEOUT:-12}"
if [[ "${SKIP_CELERY:-}" == "1" ]]; then
export EVAL_ATTACH_ONLY=1
echo -e "\033[1;33m⚠ SKIP_CELERY=1 已映射为 EVAL_ATTACH_ONLY=1请确保 :8000 主站与 Celery 已由其他进程提供\033[0m" >&2
fi
API_PID=""
CELERY_PID=""
EVAL_WEB_PID=""
CLEANED_UP=0
INFRA_STARTED=0
print_header() {
echo -e "\n${BLUE}========================================${NC}"
echo -e "${BLUE}$1${NC}"
echo -e "${BLUE}========================================${NC}"
}
print_ok() {
echo -e "${GREEN}$1${NC}"
}
print_warn() {
echo -e "${YELLOW}$1${NC}"
}
print_err() {
echo -e "${RED}$1${NC}"
}
is_pid_alive() {
local pid="$1"
[[ -n "${pid}" ]] && kill -0 "${pid}" 2>/dev/null
}
wait_pid_exit() {
local pid="$1"
local timeout="$2"
local waited=0
while is_pid_alive "${pid}"; do
if (( waited >= timeout )); then
return 1
fi
sleep 1
waited=$((waited + 1))
done
return 0
}
kill_children_term() {
local pid="$1"
local children
children="$(pgrep -P "${pid}" 2>/dev/null || true)"
if [[ -n "${children}" ]]; then
while IFS= read -r child_pid; do
[[ -z "${child_pid}" ]] && continue
kill_children_term "${child_pid}"
kill -TERM "${child_pid}" 2>/dev/null || true
done <<< "${children}"
fi
}
stop_process_gracefully() {
local name="$1"
local pid="$2"
local timeout="${3:-10}"
if ! is_pid_alive "${pid}"; then
print_ok "${name} 已退出"
return 0
fi
print_warn "正在停止 ${name}PID: ${pid}..."
kill_children_term "${pid}"
kill -TERM "${pid}" 2>/dev/null || true
if wait_pid_exit "${pid}" "${timeout}"; then
print_ok "${name} 已停止"
return 0
fi
print_warn "${name}${timeout}s 内未退出,准备强制结束"
kill -KILL "${pid}" 2>/dev/null || true
wait_pid_exit "${pid}" 3 || true
print_ok "${name} 已强制结束"
}
cleanup() {
if [[ "${CLEANED_UP}" == "1" ]]; then
return 0
fi
CLEANED_UP=1
print_header "正在关闭内部评测环境"
if is_pid_alive "${EVAL_WEB_PID}"; then
stop_process_gracefully "eval-web (Vite)" "${EVAL_WEB_PID}" "${SHUTDOWN_TIMEOUT}"
fi
if is_pid_alive "${API_PID}"; then
stop_process_gracefully "Internal Eval API" "${API_PID}" "${SHUTDOWN_TIMEOUT}"
fi
if is_pid_alive "${CELERY_PID}"; then
stop_process_gracefully "Celery" "${CELERY_PID}" "${SHUTDOWN_TIMEOUT}"
fi
if [[ "${INFRA_STARTED}" == "1" ]]; then
print_warn "正在停止 PostgreSQL / Redis 容器..."
(
cd "${ROOT_DIR}" && docker compose -f docker-compose.dev.yml stop
) >/dev/null 2>&1 || true
print_ok "PostgreSQL/Redis 容器已停止"
fi
}
require_cmd() {
local cmd="$1"
if ! command -v "${cmd}" >/dev/null 2>&1; then
print_err "未找到命令: ${cmd}"
exit 1
fi
}
start_infra() {
print_header "启动 PostgreSQL 和 Redis"
cd "${ROOT_DIR}"
docker compose -f docker-compose.dev.yml up -d
INFRA_STARTED=1
print_ok "基础设施已就绪"
}
wait_postgres_ready() {
local retries=30
local i=0
print_header "等待 PostgreSQL 就绪"
cd "${ROOT_DIR}"
while (( i < retries )); do
if docker compose -f docker-compose.dev.yml exec -T postgres \
pg_isready -U postgres >/dev/null 2>&1; then
print_ok "PostgreSQL 已就绪"
return 0
fi
sleep 1
i=$((i + 1))
done
print_warn "PostgreSQL 在 ${retries}s 内未就绪,迁移可能失败"
return 1
}
get_effective_database_url() {
if [[ -n "${DATABASE_URL:-}" ]]; then
printf '%s\n' "${DATABASE_URL}"
return 0
fi
if [[ -f "${ROOT_DIR}/.env" ]]; then
local line
line="$(sed -n 's/^DATABASE_URL=//p' "${ROOT_DIR}/.env" | sed -n '1p')"
line="${line%\"}"
line="${line#\"}"
line="${line%\'}"
line="${line#\'}"
if [[ -n "${line}" ]]; then
printf '%s\n' "${line}"
return 0
fi
fi
return 1
}
warn_database_url_host_pitfall() {
local database_url
local host
if ! database_url="$(get_effective_database_url)"; then
return 0
fi
if [[ "${database_url}" =~ @([^:/?#]+) ]]; then
host="${BASH_REMATCH[1]}"
case "${host}" in
postgres|db|postgres-dev|postgresql)
print_warn "检测到 DATABASE_URL 主机为 ${host};在宿主机执行 Alembic/uvicorn 时通常应使用 localhost"
;;
esac
fi
}
print_alembic_failure_hint() {
local log_file="$1"
local log_output
log_output="$(sed -n '1,200p' "${log_file}")"
if [[ "${log_output}" == *'could not translate host name "postgres"'* ]] || [[ "${log_output}" == *"Name or service not known"* ]]; then
print_warn "看起来 DATABASE_URL 指向了容器内主机名;在宿主机运行时请改用 localhost:5432"
elif [[ "${log_output}" == *"Connection refused"* ]] || [[ "${log_output}" == *"could not connect to server"* ]]; then
print_warn "PostgreSQL 连接被拒绝;请确认容器已启动且 DATABASE_URL 与 docker-compose.dev.yml 暴露端口一致"
elif [[ "${log_output}" == *"password authentication failed"* ]]; then
print_warn "PostgreSQL 用户名或密码不匹配;请核对 .env.development 中的 DATABASE_URL"
elif [[ "${log_output}" == *"No such file or directory"* ]] || [[ "${log_output}" == *"can't open file"* ]]; then
print_warn "Alembic 依赖的文件或工作目录可能不正确;请确认在 api/ 目录运行脚本"
fi
}
is_port_listening() {
local port="$1"
if command -v lsof >/dev/null 2>&1; then
lsof -nP -iTCP:"${port}" -sTCP:LISTEN >/dev/null 2>&1
return $?
fi
if [[ -x "${VENV_DIR}/bin/python" ]]; then
"${VENV_DIR}/bin/python" - "${port}" <<'PY' >/dev/null 2>&1
import socket
import sys
sock = socket.socket()
sock.settimeout(0.2)
try:
sock.connect(("127.0.0.1", int(sys.argv[1])))
except OSError:
raise SystemExit(1)
finally:
sock.close()
raise SystemExit(0)
PY
return $?
fi
return 1
}
wait_for_tcp_listener() {
local pid="$1"
local port="$2"
local timeout="${3:-8}"
local waited=0
while (( waited < timeout )); do
if is_port_listening "${port}"; then
return 0
fi
if ! is_pid_alive "${pid}"; then
return 1
fi
sleep 1
waited=$((waited + 1))
done
return 2
}
ensure_background_process_alive() {
local name="$1"
local pid="$2"
sleep 1
if ! is_pid_alive "${pid}"; then
print_err "${name} 启动后立即退出,请查看上方日志"
exit 1
fi
}
ensure_venv() {
print_header "检查 Python 虚拟环境"
if [[ ! -d "${VENV_DIR}" ]]; then
print_warn ".venv 不存在,正在创建"
uv venv "${VENV_DIR}"
fi
if [[ "${SKIP_INSTALL}" != "1" ]]; then
print_header "安装 Python 依赖"
uv sync
print_ok "依赖安装完成"
else
print_warn "已跳过依赖安装 (SKIP_INSTALL=1)"
fi
}
ensure_dotenv_from_development() {
print_header "准备本地 .env"
if [[ -f "${ROOT_DIR}/.env.development" ]]; then
cp "${ROOT_DIR}/.env.development" "${ROOT_DIR}/.env"
print_ok "已从 .env.development 同步为 .env"
return 0
fi
print_warn "未找到 .env.development将使用现有 .env若存在"
}
check_internal_eval_key() {
print_header "检查内部评测密钥"
if [[ -f "${ROOT_DIR}/.env" ]] && grep -qE '^INTERNAL_EVAL_API_KEY=.+' "${ROOT_DIR}/.env" 2>/dev/null; then
print_ok "已在 .env 中配置 INTERNAL_EVAL_API_KEY"
return 0
fi
if [[ -n "${INTERNAL_EVAL_API_KEY:-}" ]]; then
print_ok "已从环境变量传入 INTERNAL_EVAL_API_KEY"
return 0
fi
print_err "未配置 INTERNAL_EVAL_API_KEY内部评测接口将返回 503。"
print_err "请在 api/.env.development或 .env中加入一行例如"
print_err " INTERNAL_EVAL_API_KEY=\"your-long-random-secret\""
exit 1
}
check_env_file() {
print_header "检查环境变量文件"
if [[ ! -f "${ROOT_DIR}/.env" ]]; then
print_warn "未找到 .env应用可能因缺少配置启动失败"
else
print_ok "检测到 .env"
warn_database_url_host_pitfall
fi
}
run_migrations() {
print_header "执行数据库迁移"
cd "${ROOT_DIR}"
local log_file
log_file="$(mktemp -t life-echo-alembic.XXXXXX.log)"
if uv run alembic upgrade head >"${log_file}" 2>&1; then
print_ok "Alembic 迁移已就绪"
rm -f "${log_file}"
else
print_warn "Alembic 迁移失败(可能数据库未启动或 DATABASE_URL 未配置),应用启动可能失败"
print_alembic_failure_hint "${log_file}"
print_warn "Alembic 输出(最近 40 行):"
tail -n 40 "${log_file}"
rm -f "${log_file}"
fi
}
start_eval_web() {
print_header "启动 app-eval-web (Vite)"
if [[ ! -d "${EVAL_WEB_DIR}" ]]; then
print_err "未找到 ${EVAL_WEB_DIR}"
exit 1
fi
if [[ ! -d "${EVAL_WEB_DIR}/node_modules" ]]; then
print_err "请先执行: cd app-eval-web && npm install"
exit 1
fi
require_cmd "npm"
local api_key="${INTERNAL_EVAL_API_KEY:-}"
if [[ -z "${api_key}" ]] && [[ -f "${ROOT_DIR}/.env" ]]; then
api_key="$(grep -E '^INTERNAL_EVAL_API_KEY=' "${ROOT_DIR}/.env" | head -1 | cut -d= -f2- | tr -d '\r' | sed 's/^"//;s/"$//')"
fi
if [[ -z "${api_key}" ]]; then
print_err "无法解析 INTERNAL_EVAL_API_KEY无法为 Vite 注入 VITE_EVAL_API_KEY"
exit 1
fi
local vite_extra=()
if [[ "${OPEN_EVAL_WEB}" == "1" ]]; then
vite_extra+=(--open)
fi
# 不设 VITE_EVAL_API_BASE前端走 Vite proxyapp-eval-web/vite.config.ts转发到 :${INTERNAL_EVAL_PORT},减少直连/CORS/误指主站问题。
# 若需直连远端 APIexport VITE_EVAL_API_BASE=https://... 后再手动 npm run dev。
(
cd "${EVAL_WEB_DIR}"
VITE_EVAL_API_KEY="${api_key}" \
npm run dev -- --host 127.0.0.1 --port "${EVAL_WEB_PORT}" "${vite_extra[@]}"
) &
EVAL_WEB_PID=$!
print_ok "eval-web 已启动 (PID: ${EVAL_WEB_PID}) → http://127.0.0.1:${EVAL_WEB_PORT}/"
}
start_services() {
print_header "启动 Internal Eval API 与 Celery"
cd "${ROOT_DIR}"
if is_port_listening "${INTERNAL_EVAL_PORT}"; then
print_err "端口 ${INTERNAL_EVAL_PORT} 已被占用,无法启动内部评测 Uvicorn。"
print_err "请先结束占用进程,或设置 INTERNAL_EVAL_PORT 为其他端口"
exit 1
fi
# 与主开发脚本一致:评审/生产 LLM 等从 .env 读取;文档默认关闭,本地可 export INTERNAL_EVAL_ENABLE_DOCS=1
"${UVICORN_BIN}" app.internal_main:internal_app --reload \
--reload-exclude 'alembic/**' \
--reload-exclude 'alembic.ini' \
--host "${INTERNAL_EVAL_HOST}" --port "${INTERNAL_EVAL_PORT}" &
API_PID=$!
local api_start_status=0
if wait_for_tcp_listener "${API_PID}" "${INTERNAL_EVAL_PORT}" 8; then
api_start_status=0
else
api_start_status=$?
fi
case "${api_start_status}" in
0)
print_ok "Internal Eval API 已启动 (PID: ${API_PID})"
;;
1)
print_err "Internal Eval API 启动失败,进程已退出;请查看上方 Uvicorn 日志"
exit 1
;;
*)
print_err "Internal Eval API 进程仍存活,但端口 ${INTERNAL_EVAL_PORT} 未在预期时间内开始监听"
exit 1
;;
esac
if [[ "${SKIP_CELERY}" != "1" ]]; then
"${CELERY_BIN}" -A app.tasks.celery_app worker --loglevel=info --pool="${CELERY_POOL}" &
CELERY_PID=$!
ensure_background_process_alive "Celery" "${CELERY_PID}"
print_ok "Celery 已启动 (PID: ${CELERY_PID})"
else
print_warn "已跳过 Celery (SKIP_CELERY=1);实验 run 接口需要 worker 才能执行"
fi
if [[ "${START_EVAL_WEB}" == "1" ]]; then
start_eval_web
fi
echo
echo -e "${GREEN}内部评测环境启动完成${NC}"
echo "【请用浏览器打开】评测 Web UI: http://127.0.0.1:${EVAL_WEB_PORT}/ /internal 会代理到 API :${INTERNAL_EVAL_PORT}"
echo "内部评测 API: http://127.0.0.1:${INTERNAL_EVAL_PORT}/health"
echo "评测 REST 前缀: http://127.0.0.1:${INTERNAL_EVAL_PORT}/internal/api/evaluation"
if [[ "${INTERNAL_EVAL_ENABLE_DOCS:-}" == "1" ]] || grep -qE '^INTERNAL_EVAL_ENABLE_DOCS=true' "${ROOT_DIR}/.env" 2>/dev/null; then
echo "API 文档: http://127.0.0.1:${INTERNAL_EVAL_PORT}/docs"
fi
echo "说明文档: api/docs/internal-eval.md"
echo "按 Ctrl+C 停止所有进程"
}
main() {
print_header "Life Echo 内部回归评测 — 一键启动"
echo -e "${BLUE}说明:${NC} 不启动主站 APImain:app / 默认 8000仅启动 internal_main:${INTERNAL_EVAL_PORT})。"
echo ""
require_cmd "uv"
trap cleanup EXIT INT TERM
if [[ "${SKIP_INFRA}" != "1" ]]; then
require_cmd "docker"
start_infra
wait_postgres_ready || true
else
print_warn "已跳过 docker 基础设施 (SKIP_INFRA=1)"
fi
ensure_venv
ensure_dotenv_from_development
check_env_file
check_internal_eval_key
run_migrations
start_services
local wait_pids=("${API_PID}")
if [[ "${SKIP_CELERY}" != "1" ]]; then
wait_pids+=("${CELERY_PID}")
fi
if [[ "${START_EVAL_WEB}" == "1" ]] && [[ -n "${EVAL_WEB_PID}" ]]; then
wait_pids+=("${EVAL_WEB_PID}")
fi
wait "${wait_pids[@]}"
}
main "$@"
exec "${ROOT_DIR}/development.sh" "$@"