diff --git a/.env.example b/.env.example
index b0a3bfb..c65fca8 100644
--- a/.env.example
+++ b/.env.example
@@ -24,8 +24,10 @@ INDEXER_HOST=0.0.0.0
 INDEXER_PORT=6004
 
 # Optional service ports
+FRONTEND_PORT=6003
 EMBEDDING_PORT=6005
 TEI_PORT=8080
+CNCLIP_PORT=51000
 TRANSLATION_PORT=6006
 RERANKER_PORT=6007
 EMBEDDING_SERVICE_URL=http://127.0.0.1:6005
@@ -35,7 +37,7 @@ TRANSLATION_PROVIDER=direct
 TRANSLATION_MODEL=qwen
 EMBEDDING_BACKEND=tei
 TEI_BASE_URL=http://127.0.0.1:8080
-TEI_USE_GPU=1
+TEI_DEVICE=cuda
 TEI_VERSION=1.9
 TEI_MAX_BATCH_TOKENS=2048
 TEI_MAX_CLIENT_BATCH_SIZE=8
@@ -43,13 +45,6 @@ TEI_HEALTH_TIMEOUT_SEC=300
 RERANK_PROVIDER=http
 RERANK_BACKEND=qwen3_vllm
 
-# Optional startup switches (run.sh / scripts/service_ctl.sh)
-START_EMBEDDING=0
-START_TRANSLATOR=0
-START_RERANKER=0
-START_TEI=0
-START_CNCLIP=0
-
 # Cache Directory
 CACHE_DIR=.cache
 
diff --git a/README.md b/README.md
index e6a7641..d63ce8c 100644
--- a/README.md
+++ b/README.md
@@ -29,12 +29,15 @@ source activate.sh
 ./run.sh
 
 # 可选：附加能力服务（按需开启）
-START_EMBEDDING=1 START_TRANSLATOR=1 START_RERANKER=1 START_TEI=1 START_CNCLIP=1 ./run.sh
+./scripts/service_ctl.sh start embedding translator reranker tei cnclip
 
 # 查看状态
 ./scripts/service_ctl.sh status
 ```
 
+服务管理全盘说明（入口职责、默认行为、全量启停方式）见：
+- `docs/Usage-Guide.md` -> `服务管理总览`
+
 核心端口：
 
 - `6002` backend（`/search/*`, `/admin/*`）
diff --git a/config/env_config.py b/config/env_config.py
index fcda748..936c388 100644
--- a/config/env_config.py
+++ b/config/env_config.py
@@ -62,7 +62,7 @@ INDEXER_PORT = int(os.getenv('INDEXER_PORT', 6004))
 EMBEDDING_HOST = os.getenv('EMBEDDING_HOST', '127.0.0.1')
 EMBEDDING_PORT = int(os.getenv('EMBEDDING_PORT', 6005))
 TRANSLATION_HOST = os.getenv('TRANSLATION_HOST', '127.0.0.1')
-TRANSLATION_PORT = int(os.getenv('TRANSLATION_PORT', os.getenv('TRANSLATOR_PORT', 6006)))
+TRANSLATION_PORT = int(os.getenv('TRANSLATION_PORT', 6006))
 TRANSLATION_PROVIDER = os.getenv('TRANSLATION_PROVIDER', 'direct')
 TRANSLATION_MODEL = os.getenv('TRANSLATION_MODEL', 'qwen')
 RERANKER_HOST = os.getenv('RERANKER_HOST', '127.0.0.1')
diff --git a/docs/CNCLIP_SERVICE说明文档.md b/docs/CNCLIP_SERVICE说明文档.md
index 36c74c1..650630a 100644
--- a/docs/CNCLIP_SERVICE说明文档.md
+++ b/docs/CNCLIP_SERVICE说明文档.md
@@ -70,7 +70,8 @@ cd /data/saas-search
 ### 6.1 通过统一编排启动
 
 ```bash
-START_EMBEDDING=1 START_TEI=1 START_CNCLIP=1 ./scripts/service_ctl.sh start
+./scripts/service_ctl.sh start cnclip
+# 或一次启动可选能力：./scripts/service_ctl.sh start embedding tei cnclip
 ```
 
 ### 6.2 设备选择优先级
@@ -110,6 +111,11 @@ cat third-party/clip-as-service/server/torch-flow-temp.yml
 - GPU 模式：`device: 'cuda'`
 - CPU 模式：`device: 'cpu'`
 
+### 7.2.1 日志与 PID 文件
+
+- 日志：`logs/cnclip.log`
+- PID：`logs/cnclip.pid`
+
 ### 7.3 发送一次编码请求（触发模型加载）
 
 ```bash
diff --git a/docs/QUICKSTART.md b/docs/QUICKSTART.md
index d921996..7d2d603 100644
--- a/docs/QUICKSTART.md
+++ b/docs/QUICKSTART.md
@@ -66,18 +66,20 @@ source activate.sh
 ```bash
 ./run.sh
 # 启动全部能力
-START_EMBEDDING=1 START_TRANSLATOR=1 START_RERANKER=1 START_TEI=1 START_CNCLIP=1 ./run.sh
-# 等价方式（直接使用服务控制器）
-START_EMBEDDING=1 START_TRANSLATOR=1 START_RERANKER=1 START_TEI=1 START_CNCLIP=1 ./scripts/service_ctl.sh start
+# 追加可选能力服务（显式指定）
+TEI_DEVICE=cuda CNCLIP_DEVICE=cuda ./scripts/service_ctl.sh start tei cnclip embedding translator reranker
 # 说明：
 # - reranker 为 GPU 强制模式（资源不足会直接启动失败）
-# - TEI 默认使用 GPU；当 TEI_USE_GPU=1 且 GPU 不可用时会直接失败（不会自动降级到 CPU）
+# - TEI 默认使用 GPU；当 TEI_DEVICE=cuda 且 GPU 不可用时会直接失败（不会自动降级到 CPU）
 # - cnclip 默认使用 cuda；若显式配置为 cuda 且 GPU 不可用会直接失败（不会自动降级到 cpu）
 
 ./scripts/service_ctl.sh status
 ./scripts/stop.sh
 ```
 
+服务管理方式（入口职责、默认行为、全量拉起顺序）见：
+- `docs/Usage-Guide.md` -> `服务管理总览`
+
 ### 1.3 常用 API 请求示例
 
 #### 搜索 API（backend 6002）
@@ -135,7 +137,7 @@ API 文档：`http://localhost:6004/docs`
 ```bash
 # TEI（文本向量后端，默认）
 # GPU（需 nvidia-container-toolkit）
-TEI_USE_GPU=1 ./scripts/start_tei_service.sh
+TEI_DEVICE=cuda ./scripts/start_tei_service.sh
 
 # Embedding API（会校验 TEI /health）
 ./scripts/start_embedding_service.sh
@@ -151,7 +153,7 @@ curl -X POST http://localhost:6005/embed/image \
 
 说明：
 - TEI 默认镜像按 `TEI_VERSION` 组装：`cuda-<version>`（默认 `1.9`）。
-- `TEI_USE_GPU=1` 时会严格校验 Docker GPU runtime；未配置会直接报错退出。
+- `TEI_DEVICE=cuda` 时会严格校验 Docker GPU runtime；未配置会直接报错退出。
 - `/embed/image` 依赖 `cnclip`（`grpc://127.0.0.1:51000`），未启动时 embedding 服务会启动失败。
 
 #### Translator 服务（6006）
@@ -514,6 +516,8 @@ curl http://localhost:6007/health
 - `logs/embedding.log`
 - `logs/translator.log`
 - `logs/reranker.log`
+- `logs/tei.log`
+- `logs/cnclip.log`
 - `logs/search_engine.log`
 - `logs/errors.log`
 
diff --git a/docs/TEI_SERVICE说明文档.md b/docs/TEI_SERVICE说明文档.md
index 103a9fa..b20d3ef 100644
--- a/docs/TEI_SERVICE说明文档.md
+++ b/docs/TEI_SERVICE说明文档.md
@@ -63,13 +63,13 @@ docker info --format '{{json .Runtimes}}'
 ### 5.1 GPU 模式启动（默认）
 
 ```bash
-TEI_USE_GPU=1 ./scripts/start_tei_service.sh
+TEI_DEVICE=cuda ./scripts/start_tei_service.sh
 ```
 
 预期输出包含：
 
 - `Image: ghcr.io/huggingface/text-embeddings-inference:turing-...` 或 `cuda-...`（脚本按 GPU 架构自动选择）
-- `Mode: gpu`
+- `Mode: cuda`
 - `TEI is ready and output probe passed: http://127.0.0.1:8080`
 
 说明：
@@ -79,7 +79,7 @@ TEI_USE_GPU=1 ./scripts/start_tei_service.sh
 ### 5.2 CPU 模式启动（显式）
 
 ```bash
-TEI_USE_GPU=0 ./scripts/start_tei_service.sh
+TEI_DEVICE=cpu ./scripts/start_tei_service.sh
 ```
 
 预期输出包含：
@@ -135,7 +135,7 @@ curl -sS -X POST "http://127.0.0.1:6005/embed/text" \
 
 `scripts/start_tei_service.sh` 支持下列变量：
 
-- `TEI_USE_GPU`：`1/0`（或 `true/false`），默认 `1`
+- `TEI_DEVICE`：`cuda/cpu`，默认 `cuda`
 - `TEI_CONTAINER_NAME`：容器名，默认 `saas-search-tei`
 - `TEI_PORT`：宿主机端口，默认 `8080`
 - `TEI_MODEL_ID`：默认 `Qwen/Qwen3-Embedding-0.6B`
@@ -152,7 +152,7 @@ curl -sS -X POST "http://127.0.0.1:6005/embed/text" \
 启动全套（含 TEI）：
 
 ```bash
-START_EMBEDDING=1 START_TRANSLATOR=1 START_RERANKER=1 START_TEI=1 START_CNCLIP=1 TEI_USE_GPU=1 ./scripts/service_ctl.sh start
+TEI_DEVICE=cuda ./scripts/service_ctl.sh start embedding translator reranker tei cnclip
 ```
 
 仅启动 TEI：
@@ -167,9 +167,11 @@ START_EMBEDDING=1 START_TRANSLATOR=1 START_RERANKER=1 START_TEI=1 START_CNCLIP=1
 ./scripts/service_ctl.sh status tei
 ```
 
+日志文件：`logs/tei.log`
+
 ## 9. 常见问题与排障
 
-### 9.1 `ERROR: TEI_USE_GPU=1 but Docker nvidia runtime is not configured`
+### 9.1 `ERROR: TEI_DEVICE=cuda but Docker nvidia runtime is not configured`
 
 - 原因：Docker 未配置 NVIDIA runtime。
 - 处理：按本文 4.2 配置后重启 Docker。
@@ -181,7 +183,7 @@ START_EMBEDDING=1 START_TRANSLATOR=1 START_RERANKER=1 START_TEI=1 START_CNCLIP=1
 
 ```bash
 ./scripts/stop_tei_service.sh
-TEI_USE_GPU=0 ./scripts/start_tei_service.sh   # 或改为 1
+TEI_DEVICE=cpu ./scripts/start_tei_service.sh   # 或改为 cuda
 ```
 
 ### 9.3 embedding 服务报 TEI 不可达
diff --git a/docs/Usage-Guide.md b/docs/Usage-Guide.md
index e4c53d8..9cd4786 100644
--- a/docs/Usage-Guide.md
+++ b/docs/Usage-Guide.md
@@ -8,10 +8,11 @@
 
 1. [环境准备](#环境准备)
 2. [服务启动](#服务启动)
-3. [配置说明](#配置说明)
-4. [查看日志](#查看日志)
-5. [测试验证](#测试验证)
-6. [常见问题](#常见问题)
+3. [服务管理总览](#全盘串讲服务管理方式)
+4. [配置说明](#配置说明)
+5. [查看日志](#查看日志)
+6. [测试验证](#测试验证)
+7. [常见问题](#常见问题)
 
 ---
 
@@ -50,10 +51,10 @@ TEI 文本向量服务使用 Docker 容器：
 
 ```bash
 # GPU（需 nvidia-container-toolkit）
-TEI_USE_GPU=1 ./scripts/start_tei_service.sh
+TEI_DEVICE=cuda ./scripts/start_tei_service.sh
 
 # CPU
-TEI_USE_GPU=0 ./scripts/start_tei_service.sh
+TEI_DEVICE=cpu ./scripts/start_tei_service.sh
 ```
 
 专项说明：
@@ -138,7 +139,7 @@ cd /data/saas-search
 可选：全功能模式（同时启动 embedding/translator/reranker/tei/cnclip）：
 
 ```bash
-START_EMBEDDING=1 START_TRANSLATOR=1 START_RERANKER=1 START_TEI=1 START_CNCLIP=1 ./run.sh
+TEI_DEVICE=cuda CNCLIP_DEVICE=cuda ./scripts/service_ctl.sh start tei cnclip embedding translator reranker
 ```
 
 ### 方式2: 统一控制脚本（推荐）
@@ -241,6 +242,49 @@ cd frontend
 python -m http.server 6003
 ```
 
+## 服务管理总览
+
+### 1) 入口脚本职责
+
+- `./run.sh`：仅启动核心服务（`backend/indexer/frontend`）。
+- `./restart.sh`：重启逻辑为“先停所有已知服务，再启动核心服务”。
+- `./scripts/stop.sh`：停止所有已知服务。
+- `./scripts/service_ctl.sh`：统一控制器，支持 `start/stop/restart/status`，是唯一推荐入口。
+
+### 2) `service_ctl.sh` 的默认行为
+
+- `start`（不带服务名）：启动核心服务 `backend/indexer/frontend`。
+- `stop`（不带服务名）：停止全部已知服务（含可选服务）。
+- `restart`（不带服务名）：先停全部，再只启动核心服务。
+- `status`（不带服务名）：显示全部已知服务状态。
+
+### 3) 全量服务一键拉起
+
+```bash
+TEI_DEVICE=cuda CNCLIP_DEVICE=cuda ./scripts/service_ctl.sh start tei cnclip embedding translator reranker
+```
+
+说明：
+- `TEI_DEVICE` / `CNCLIP_DEVICE` 统一使用 `cuda|cpu`。
+- 显式把 `tei`、`cnclip` 放在前面，避免 `embedding` 因依赖未就绪启动失败。
+
+### 4) 常用运维命令
+
+```bash
+# 先重启核心，再拉起可选服务（最常用）
+./restart.sh
+TEI_DEVICE=cuda CNCLIP_DEVICE=cuda ./scripts/service_ctl.sh start tei cnclip embedding translator reranker
+
+# 查看全量状态
+./scripts/service_ctl.sh status
+
+# 仅重启某个服务
+./scripts/service_ctl.sh restart embedding
+
+# 停止全部
+./scripts/service_ctl.sh stop
+```
+
 ### 停止服务
 
 ```bash
@@ -303,16 +347,12 @@ INDEXER_HOST=0.0.0.0
 INDEXER_PORT=6004
 
 # Optional service ports
+FRONTEND_PORT=6003
 EMBEDDING_PORT=6005
+TEI_PORT=8080
+CNCLIP_PORT=51000
 TRANSLATION_PORT=6006
 RERANKER_PORT=6007
-
-# Optional startup switches (for run.sh / service_ctl.sh)
-START_EMBEDDING=0
-START_TRANSLATOR=0
-START_RERANKER=0
-START_TEI=0
-START_CNCLIP=0
 ```
 
 ### 修改配置
@@ -334,6 +374,8 @@ START_CNCLIP=0
 - `logs/embedding.log` - 向量服务日志（可选）
 - `logs/translator.log` - 翻译服务日志（可选）
 - `logs/reranker.log` - 重排服务日志（可选）
+- `logs/tei.log` - TEI 启停日志（可选）
+- `logs/cnclip.log` - CN-CLIP 启停日志（可选）
 - `logs/search_engine.log` - 应用主日志（按天轮转）
 - `logs/errors.log` - 错误日志（按天轮转）
 
diff --git a/docs/搜索API对接指南.md b/docs/搜索API对接指南.md
index d98d811..e62a8c6 100644
--- a/docs/搜索API对接指南.md
+++ b/docs/搜索API对接指南.md
@@ -1655,6 +1655,7 @@ curl -X POST "http://localhost:6004/indexer/enrich-content" \
 | 重排服务 | 6007 | `http://localhost:6007` | 对检索结果进行二次排序 |
 
 生产环境请将 `localhost` 替换为实际服务地址。
+服务管理入口与完整启停规则见：`docs/Usage-Guide.md` -> `服务管理总览`。
 
 ### 7.1 向量服务（Embedding）
 
@@ -1663,7 +1664,7 @@ curl -X POST "http://localhost:6004/indexer/enrich-content" \
 - **依赖**:
   - 文本向量后端默认走 TEI（`http://127.0.0.1:8080`）
   - 图片向量依赖 `cnclip`（`grpc://127.0.0.1:51000`）
-  - TEI 默认使用 GPU（`TEI_USE_GPU=1`）；当配置为 GPU 且不可用时会启动失败（不会自动降级到 CPU）
+  - TEI 默认使用 GPU（`TEI_DEVICE=cuda`）；当配置为 GPU 且不可用时会启动失败（不会自动降级到 CPU）
   - cnclip 默认使用 `cuda`；若配置为 `cuda` 但 GPU 不可用会启动失败（不会自动降级到 `cpu`）
 
 #### 7.1.1 `POST /embed/text` — 文本向量化
diff --git a/embeddings/README.md b/embeddings/README.md
index 6c3bb9b..730f515 100644
--- a/embeddings/README.md
+++ b/embeddings/README.md
@@ -65,10 +65,10 @@
 
 ```bash
 # GPU（需 nvidia-container-toolkit）
-TEI_USE_GPU=1 ./scripts/start_tei_service.sh
+TEI_DEVICE=cuda ./scripts/start_tei_service.sh
 
 # CPU
-TEI_USE_GPU=0 ./scripts/start_tei_service.sh
+TEI_DEVICE=cpu ./scripts/start_tei_service.sh
 
 ./scripts/start_embedding_service.sh
 ```
@@ -82,4 +82,4 @@ TEI_USE_GPU=0 ./scripts/start_tei_service.sh
 - `IMAGE_NORMALIZE_EMBEDDINGS`（默认 true）
 - `USE_CLIP_AS_SERVICE`, `CLIP_AS_SERVICE_SERVER`：图片向量（clip-as-service）
 - `IMAGE_MODEL_NAME`, `IMAGE_DEVICE`：本地 CN-CLIP（当 `USE_CLIP_AS_SERVICE=false` 时）
-- TEI 相关：`TEI_USE_GPU`、`TEI_VERSION`、`TEI_MAX_BATCH_TOKENS`、`TEI_MAX_CLIENT_BATCH_SIZE`、`TEI_HEALTH_TIMEOUT_SEC`
+- TEI 相关：`TEI_DEVICE`、`TEI_VERSION`、`TEI_MAX_BATCH_TOKENS`、`TEI_MAX_CLIENT_BATCH_SIZE`、`TEI_HEALTH_TIMEOUT_SEC`
diff --git a/reranker/DEPLOYMENT_AND_TUNING.md b/reranker/DEPLOYMENT_AND_TUNING.md
index 48a82e0..54b03a2 100644
--- a/reranker/DEPLOYMENT_AND_TUNING.md
+++ b/reranker/DEPLOYMENT_AND_TUNING.md
@@ -78,7 +78,7 @@ services:
 
 ```bash
 ./scripts/start_reranker.sh
-./scripts/stop_reranker.sh
+./scripts/service_ctl.sh stop reranker
 ```
 
 健康检查：
diff --git a/restart.sh b/restart.sh
index cf2d62a..49c036f 100755
--- a/restart.sh
+++ b/restart.sh
@@ -2,6 +2,8 @@
 
 # Unified restart script for saas-search services
 
+set -euo pipefail
+
 cd "$(dirname "$0")"
 
 ./scripts/service_ctl.sh restart
diff --git a/run.sh b/run.sh
index 2afccb8..403554b 100755
--- a/run.sh
+++ b/run.sh
@@ -2,6 +2,8 @@
 
 # Unified startup script for saas-search services
 
+set -euo pipefail
+
 cd "$(dirname "$0")"
 
 ./scripts/service_ctl.sh start
diff --git a/scripts/benchmark_reranker_1000docs.sh b/scripts/benchmark_reranker_1000docs.sh
index 8f06e85..d898740 100755
--- a/scripts/benchmark_reranker_1000docs.sh
+++ b/scripts/benchmark_reranker_1000docs.sh
@@ -33,7 +33,7 @@ TMP_CASES="/tmp/rerank_1000_shortdocs_cases.json"
 mkdir -p "${OUT_DIR}"
 
 cleanup() {
-  ./scripts/stop_reranker.sh >/dev/null 2>&1 || true
+  ./scripts/service_ctl.sh stop reranker >/dev/null 2>&1 || true
 }
 trap cleanup EXIT
 
diff --git a/scripts/frontend_server.py b/scripts/frontend_server.py
index 0e0ab04..77dac02 100755
--- a/scripts/frontend_server.py
+++ b/scripts/frontend_server.py
@@ -30,8 +30,8 @@ BACKEND_PROXY_URL = os.getenv('BACKEND_PROXY_URL', 'http://127.0.0.1:6002').rstr
 frontend_dir = os.path.join(os.path.dirname(__file__), '../frontend')
 os.chdir(frontend_dir)
 
-# Get port from environment variable or default
-PORT = int(os.getenv('PORT', 6003))
+# FRONTEND_PORT is the canonical config; keep PORT as a secondary fallback.
+PORT = int(os.getenv('FRONTEND_PORT', os.getenv('PORT', 6003)))
 
 # Configure logging to suppress scanner noise
 logging.basicConfig(level=logging.ERROR, format='%(asctime)s - %(levelname)s - %(message)s')
diff --git a/scripts/service_ctl.sh b/scripts/service_ctl.sh
index 8d99aae..a018cb9 100755
--- a/scripts/service_ctl.sh
+++ b/scripts/service_ctl.sh
@@ -28,7 +28,7 @@ get_port() {
     indexer) echo "${INDEXER_PORT:-6004}" ;;
     frontend) echo "${FRONTEND_PORT:-6003}" ;;
     embedding) echo "${EMBEDDING_PORT:-6005}" ;;
-    translator) echo "${TRANSLATION_PORT:-${TRANSLATOR_PORT:-6006}}" ;;
+    translator) echo "${TRANSLATION_PORT:-6006}" ;;
     reranker) echo "${RERANKER_PORT:-6007}" ;;
     tei) echo "${TEI_PORT:-8080}" ;;
     cnclip) echo "${CNCLIP_PORT:-51000}" ;;
@@ -38,10 +38,7 @@ get_port() {
 
 pid_file() {
   local service="$1"
-  case "${service}" in
-    cnclip) echo "${LOG_DIR}/cnclip_service.pid" ;;
-    *) echo "${LOG_DIR}/${service}.pid" ;;
-  esac
+  echo "${LOG_DIR}/${service}.pid"
 }
 
 log_file() {
@@ -64,6 +61,24 @@ service_start_cmd() {
   esac
 }
 
+service_exists() {
+  local service="$1"
+  case "${service}" in
+    backend|indexer|frontend|embedding|translator|reranker|tei|cnclip) return 0 ;;
+    *) return 1 ;;
+  esac
+}
+
+validate_targets() {
+  local targets="$1"
+  for svc in ${targets}; do
+    if ! service_exists "${svc}"; then
+      echo "[error] unknown service: ${svc}" >&2
+      return 1
+    fi
+  done
+}
+
 wait_for_health() {
   local service="$1"
   local max_retries="${2:-30}"
@@ -188,9 +203,15 @@ start_one() {
     cnclip|tei)
       echo "[start] ${service} (managed by native script)"
       if [ "${service}" = "cnclip" ]; then
-        CNCLIP_DEVICE="${CNCLIP_DEVICE:-cuda}" bash -lc "${cmd}" >> "${lf}" 2>&1
+        if ! CNCLIP_DEVICE="${CNCLIP_DEVICE:-cuda}" bash -lc "${cmd}" >> "${lf}" 2>&1; then
+          echo "[error] ${service} start script failed, inspect ${lf}" >&2
+          return 1
+        fi
       else
-        bash -lc "${cmd}" >> "${lf}" 2>&1
+        if ! bash -lc "${cmd}" >> "${lf}" 2>&1; then
+          echo "[error] ${service} start script failed, inspect ${lf}" >&2
+          return 1
+        fi
       fi
       if [ "${service}" = "tei" ]; then
         if is_running_tei_container; then
@@ -244,6 +265,24 @@ start_one() {
   esac
 }
 
+cleanup_reranker_orphans() {
+  local engine_pids
+  engine_pids="$(pgrep -f 'VLLM::EngineCore' 2>/dev/null || true)"
+  if [ -z "${engine_pids}" ]; then
+    return 0
+  fi
+
+  echo "[stop] reranker orphan engines=${engine_pids}"
+  for pid in ${engine_pids}; do
+    kill -TERM "${pid}" 2>/dev/null || true
+  done
+  sleep 1
+  engine_pids="$(pgrep -f 'VLLM::EngineCore' 2>/dev/null || true)"
+  for pid in ${engine_pids}; do
+    kill -KILL "${pid}" 2>/dev/null || true
+  done
+}
+
 stop_one() {
   local service="$1"
   cd "${PROJECT_ROOT}"
@@ -257,11 +296,6 @@ stop_one() {
     bash -lc "./scripts/stop_tei_service.sh" || true
     return 0
   fi
-  if [ "${service}" = "reranker" ]; then
-    echo "[stop] reranker (managed by native script)"
-    bash -lc "./scripts/stop_reranker.sh" || true
-    return 0
-  fi
 
   local pf
   pf="$(pid_file "${service}")"
@@ -297,6 +331,10 @@ stop_one() {
       done
     fi
   fi
+
+  if [ "${service}" = "reranker" ]; then
+    cleanup_reranker_orphans
+  fi
 }
 
 status_one() {
@@ -340,13 +378,7 @@ resolve_targets() {
 
   case "${scope}" in
     start)
-      local targets=("${CORE_SERVICES[@]}")
-      if [ "${START_TEI:-0}" = "1" ]; then targets+=("tei"); fi
-      if [ "${START_CNCLIP:-0}" = "1" ]; then targets+=("cnclip"); fi
-      if [ "${START_EMBEDDING:-0}" = "1" ]; then targets+=("embedding"); fi
-      if [ "${START_TRANSLATOR:-0}" = "1" ]; then targets+=("translator"); fi
-      if [ "${START_RERANKER:-0}" = "1" ]; then targets+=("reranker"); fi
-      echo "${targets[@]}"
+      echo "${CORE_SERVICES[@]}"
       ;;
     stop|status)
       echo "$(all_services)"
@@ -370,14 +402,14 @@ Usage:
   ./scripts/service_ctl.sh status [service...]
 
 Default target set (when no service provided):
-  start   -> backend indexer frontend (+ optional by env flags)
+  start   -> backend indexer frontend
   stop    -> all known services
   restart -> stop all known services, then start with start targets
   status  -> all known services
 
-Optional startup flags:
-  START_EMBEDDING=1 START_TRANSLATOR=1 START_RERANKER=1 START_TEI=1 START_CNCLIP=1 ./run.sh
-  START_EMBEDDING=1 START_TRANSLATOR=1 START_RERANKER=1 START_TEI=1 START_CNCLIP=1 ./scripts/service_ctl.sh start
+Optional service startup:
+  ./scripts/service_ctl.sh start embedding translator reranker tei cnclip
+  TEI_DEVICE=cuda|cpu ./scripts/service_ctl.sh start tei
   CNCLIP_DEVICE=cuda|cpu ./scripts/service_ctl.sh start cnclip
 EOF
 }
@@ -404,6 +436,7 @@ main() {
     usage
     exit 1
   fi
+  validate_targets "${targets}"
 
   case "${action}" in
     start)
diff --git a/scripts/start.sh b/scripts/start.sh
index e7184cf..5c30615 100755
--- a/scripts/start.sh
+++ b/scripts/start.sh
@@ -1,9 +1,9 @@
 #!/bin/bash
 
-# Backward-compatible start entrypoint.
+# Service start entrypoint.
 # Delegates to unified service controller.
 
-set -e
+set -euo pipefail
 
 cd "$(dirname "$0")/.."
 
@@ -11,12 +11,12 @@ echo "========================================"
 echo "saas-search 服务启动"
 echo "========================================"
 echo "默认启动核心服务: backend/indexer/frontend"
-echo "可选服务通过环境变量开启:"
-echo "  START_EMBEDDING=1 START_TRANSLATOR=1 START_RERANKER=1 START_TEI=1 START_CNCLIP=1 ./run.sh"
+echo "可选服务请显式指定:"
+echo "  ./scripts/service_ctl.sh start embedding translator reranker tei cnclip"
 echo
 
 ./scripts/service_ctl.sh start
 
 echo
 echo "当前服务状态："
-./scripts/service_ctl.sh status backend indexer frontend embedding translator reranker tei
+./scripts/service_ctl.sh status backend indexer frontend embedding translator reranker tei cnclip
diff --git a/scripts/start_backend.sh b/scripts/start_backend.sh
index 49c2d27..f6be6c0 100755
--- a/scripts/start_backend.sh
+++ b/scripts/start_backend.sh
@@ -2,7 +2,7 @@
 
 # Start Backend API Service
 
-set -e
+set -euo pipefail
 
 cd "$(dirname "$0")/.."
 source ./activate.sh
@@ -15,24 +15,25 @@ echo -e "${GREEN}========================================${NC}"
 echo -e "${GREEN}Starting Backend API Service${NC}"
 echo -e "${GREEN}========================================${NC}"
 
+API_HOST="${API_HOST:-0.0.0.0}"
+API_PORT="${API_PORT:-6002}"
+ES_HOST="${ES_HOST:-http://localhost:9200}"
+ES_USERNAME="${ES_USERNAME:-}"
+ES_PASSWORD="${ES_PASSWORD:-}"
+
 echo -e "\n${YELLOW}Configuration:${NC}"
-echo "  API Host: ${API_HOST:-0.0.0.0}"
-echo "  API Port: ${API_PORT:-6002}"
-echo "  ES Host: ${ES_HOST:-http://localhost:9200}"
+echo "  API Host: ${API_HOST}"
+echo "  API Port: ${API_PORT}"
+echo "  ES Host: ${ES_HOST}"
 echo "  ES Username: ${ES_USERNAME:-not set}"
 
 echo -e "\n${YELLOW}Starting backend API service (search + admin)...${NC}"
 
 # Export environment variables for the Python process
-export API_HOST=${API_HOST:-0.0.0.0}
-export API_PORT=${API_PORT:-6002}
-export ES_HOST=${ES_HOST:-http://localhost:9200}
-export ES_USERNAME=${ES_USERNAME:-}
-export ES_PASSWORD=${ES_PASSWORD:-}
-
-python main.py serve \
-  --host $API_HOST \
-  --port $API_PORT \
-  --es-host $ES_HOST
+export API_HOST API_PORT ES_HOST ES_USERNAME ES_PASSWORD
 
+exec python main.py serve \
+  --host "${API_HOST}" \
+  --port "${API_PORT}" \
+  --es-host "${ES_HOST}"
 
diff --git a/scripts/start_cnclip_service.sh b/scripts/start_cnclip_service.sh
index e308c38..3069b28 100755
--- a/scripts/start_cnclip_service.sh
+++ b/scripts/start_cnclip_service.sh
@@ -12,9 +12,6 @@
 # 选项：
 #   --port PORT           服务端口（默认：51000）
 #   --device DEVICE       设备类型：cuda 或 cpu（默认：cuda）
-#   --batch-size SIZE     批处理大小（默认：32）
-#   --num-workers NUM     预处理线程数（默认：4）
-#   --dtype TYPE          数据类型：float16 或 float32（默认：float16）
 #   --model-name NAME     模型名称（默认：CN-CLIP/ViT-H-14）
 #   --replicas NUM        副本数（默认：1）
 #   --help                显示帮助信息
@@ -22,11 +19,10 @@
 # 示例：
 #   ./scripts/start_cnclip_service.sh
 #   ./scripts/start_cnclip_service.sh --port 52000 --device cuda
-#   ./scripts/start_cnclip_service.sh --batch-size 16 --dtype float32
 #
 ###############################################################################
 
-set -e  # 遇到错误立即退出
+set -euo pipefail
 
 # 颜色定义
 RED='\033[0;31m'
@@ -38,9 +34,6 @@ NC='\033[0m' # No Color
 # 默认配置
 DEFAULT_PORT=51000
 DEFAULT_DEVICE="cuda"
-DEFAULT_BATCH_SIZE=32
-DEFAULT_NUM_WORKERS=4
-DEFAULT_DTYPE="float16"
 DEFAULT_MODEL_NAME="CN-CLIP/ViT-H-14"
 # DEFAULT_MODEL_NAME="CN-CLIP/ViT-L-14-336"
 DEFAULT_REPLICAS=1  # 副本数
@@ -49,8 +42,8 @@ DEFAULT_REPLICAS=1  # 副本数
 PROJECT_ROOT="$(cd "$(dirname "$0")/.." && pwd)"
 CLIP_SERVER_DIR="${PROJECT_ROOT}/third-party/clip-as-service/server"
 LOG_DIR="${PROJECT_ROOT}/logs"
-PID_FILE="${LOG_DIR}/cnclip_service.pid"
-LOG_FILE="${LOG_DIR}/cnclip_service.log"
+PID_FILE="${LOG_DIR}/cnclip.pid"
+LOG_FILE="${LOG_DIR}/cnclip.log"
 
 # 帮助信息
 show_help() {
@@ -59,11 +52,8 @@ show_help() {
     echo "用法: $0 [选项]"
     echo ""
     echo "选项:"
-    echo "  --port PORT           服务端口（默认：${DEFAULT_PORT}）"
+    echo "  --port PORT           服务端口（默认：${CNCLIP_PORT:-${DEFAULT_PORT}}）"
     echo "  --device DEVICE       设备类型：cuda 或 cpu（默认：cuda）"
-    echo "  --batch-size SIZE     批处理大小（默认：${DEFAULT_BATCH_SIZE}）"
-    echo "  --num-workers NUM     预处理线程数（默认：${DEFAULT_NUM_WORKERS}）"
-    echo "  --dtype TYPE          数据类型：float16 或 float32（默认：${DEFAULT_DTYPE}）"
     echo "  --model-name NAME     模型名称（默认：${DEFAULT_MODEL_NAME}）"
     echo "  --replicas NUM        副本数（默认：${DEFAULT_REPLICAS}）"
     echo "  --help                显示此帮助信息"
@@ -72,23 +62,19 @@ show_help() {
     echo "  $0                                          # 使用默认配置启动"
     echo "  $0 --port 52000 --device cuda               # 指定 CUDA 模式，端口 52000"
     echo "  $0 --port 52000 --device cpu                # 显式使用 CPU 模式"
-    echo "  $0 --batch-size 16 --dtype float32          # 小批处理，float32 精度"
     echo "  $0 --replicas 2                            # 启动2个副本（需8-10GB显存）"
     echo ""
     echo "支持的模型:"
     echo "  - CN-CLIP/ViT-B-16      基础版本，速度快"
     echo "  - CN-CLIP/ViT-L-14      平衡版本"
-    echo "  - CN-CLIP/ViT-L-14-336  高分辨率版本（默认）"
-    echo "  - CN-CLIP/ViT-H-14      大型版本，精度高"
+    echo "  - CN-CLIP/ViT-L-14-336  高分辨率版本"
+    echo "  - CN-CLIP/ViT-H-14      大型版本，精度高（默认）"
     echo "  - CN-CLIP/RN50          ResNet-50 版本"
 }
 
 # 解析命令行参数
-PORT=${DEFAULT_PORT}
+PORT="${CNCLIP_PORT:-${DEFAULT_PORT}}"
 DEVICE=${DEFAULT_DEVICE}
-BATCH_SIZE=${DEFAULT_BATCH_SIZE}
-NUM_WORKERS=${DEFAULT_NUM_WORKERS}
-DTYPE=${DEFAULT_DTYPE}
 MODEL_NAME=${DEFAULT_MODEL_NAME}
 REPLICAS=${DEFAULT_REPLICAS}
 
@@ -102,18 +88,6 @@ while [[ $# -gt 0 ]]; do
             DEVICE="$2"
             shift 2
             ;;
-        --batch-size)
-            BATCH_SIZE="$2"
-            shift 2
-            ;;
-        --num-workers)
-            NUM_WORKERS="$2"
-            shift 2
-            ;;
-        --dtype)
-            DTYPE="$2"
-            shift 2
-            ;;
         --model-name)
             MODEL_NAME="$2"
             shift 2
@@ -197,7 +171,7 @@ python -c "import cn_clip" 2>/dev/null || {
 }
 
 # clip_server 通过 PYTHONPATH 加载（见下方启动命令），此处仅做可导入性检查
-export PYTHONPATH="${CLIP_SERVER_DIR}:${PYTHONPATH}"
+export PYTHONPATH="${CLIP_SERVER_DIR}${PYTHONPATH:+:${PYTHONPATH}}"
 python -c "import clip_server" 2>/dev/null || {
     echo -e "${RED}错误: clip_server 不可用${NC}"
     echo -e "${YELLOW}请重建专用环境: ./scripts/setup_cnclip_venv.sh${NC}"
@@ -251,7 +225,7 @@ fi
 cd "${CLIP_SERVER_DIR}"
 
 # 设置环境变量
-export PYTHONPATH="${CLIP_SERVER_DIR}:${PYTHONPATH}"
+export PYTHONPATH="${CLIP_SERVER_DIR}${PYTHONPATH:+:${PYTHONPATH}}"
 export NO_VERSION_CHECK=1  # 跳过版本检查
 
 # 启动服务
diff --git a/scripts/start_frontend.sh b/scripts/start_frontend.sh
index 5abe75f..a87b1cb 100755
--- a/scripts/start_frontend.sh
+++ b/scripts/start_frontend.sh
@@ -2,7 +2,7 @@
 
 # Start Frontend Server
 
-set -e
+set -euo pipefail
 
 cd "$(dirname "$0")/.."
 source ./activate.sh
@@ -17,6 +17,7 @@ echo -e "${GREEN}========================================${NC}"
 
 FRONTEND_PORT="${FRONTEND_PORT:-6003}"
 API_PORT="${API_PORT:-6002}"
+PORT="${FRONTEND_PORT}"
 
 echo -e "\n${YELLOW}Frontend will be available at:${NC}"
 echo -e "  ${GREEN}http://localhost:${FRONTEND_PORT}${NC}"
@@ -25,4 +26,5 @@ echo -e "${YELLOW}Make sure the backend API is running at:${NC}"
 echo -e "  ${GREEN}http://localhost:${API_PORT}${NC}"
 echo ""
 
-python scripts/frontend_server.py
+export FRONTEND_PORT API_PORT PORT
+exec python scripts/frontend_server.py
diff --git a/scripts/start_indexer.sh b/scripts/start_indexer.sh
index 8924ae8..5878680 100755
--- a/scripts/start_indexer.sh
+++ b/scripts/start_indexer.sh
@@ -2,7 +2,7 @@
 
 # Start dedicated Indexer API Service
 
-set -e
+set -euo pipefail
 
 cd "$(dirname "$0")/.."
 source ./activate.sh
@@ -15,24 +15,25 @@ echo -e "${GREEN}========================================${NC}"
 echo -e "${GREEN}Starting Indexer API Service${NC}"
 echo -e "${GREEN}========================================${NC}"
 
+INDEXER_HOST="${INDEXER_HOST:-0.0.0.0}"
+INDEXER_PORT="${INDEXER_PORT:-6004}"
+ES_HOST="${ES_HOST:-http://localhost:9200}"
+ES_USERNAME="${ES_USERNAME:-}"
+ES_PASSWORD="${ES_PASSWORD:-}"
+
 echo -e "\n${YELLOW}Configuration:${NC}"
-echo "  INDEXER Host: ${INDEXER_HOST:-0.0.0.0}"
-echo "  INDEXER Port: ${INDEXER_PORT:-6004}"
-echo "  ES Host: ${ES_HOST:-http://localhost:9200}"
+echo "  INDEXER Host: ${INDEXER_HOST}"
+echo "  INDEXER Port: ${INDEXER_PORT}"
+echo "  ES Host: ${ES_HOST}"
 echo "  ES Username: ${ES_USERNAME:-not set}"
 
 echo -e "\n${YELLOW}Starting indexer service...${NC}"
 
 # Export environment variables for the Python process
-export INDEXER_HOST=${INDEXER_HOST:-0.0.0.0}
-export INDEXER_PORT=${INDEXER_PORT:-6004}
-export ES_HOST=${ES_HOST:-http://localhost:9200}
-export ES_USERNAME=${ES_USERNAME:-}
-export ES_PASSWORD=${ES_PASSWORD:-}
-
-python main.py serve-indexer \
-  --host $INDEXER_HOST \
-  --port $INDEXER_PORT \
-  --es-host $ES_HOST
+export INDEXER_HOST INDEXER_PORT ES_HOST ES_USERNAME ES_PASSWORD
 
+exec python main.py serve-indexer \
+  --host "${INDEXER_HOST}" \
+  --port "${INDEXER_PORT}" \
+  --es-host "${ES_HOST}"
 
diff --git a/scripts/start_servers.py b/scripts/start_servers.py
deleted file mode 100755
index baa9a5b..0000000
--- a/scripts/start_servers.py
+++ /dev/null
@@ -1,249 +0,0 @@
-#!/usr/bin/env python3
-"""
-Production-ready server startup script with proper error handling and monitoring.
-
-[LEGACY]
-This script is kept for historical compatibility.
-Preferred entrypoint is:
-  ./scripts/service_ctl.sh start
-"""
-
-import os
-import sys
-import signal
-import time
-import subprocess
-import logging
-import argparse
-from typing import Dict, List, Optional
-import multiprocessing
-import threading
-
-# Configure logging
-logging.basicConfig(
-    level=logging.INFO,
-    format='%(asctime)s - %(name)s - %(levelname)s - %(message)s',
-    handlers=[
-        logging.StreamHandler(),
-        logging.FileHandler('/tmp/search_engine_startup.log', mode='a')
-    ]
-)
-logger = logging.getLogger(__name__)
-
-class ServerManager:
-    """Manages frontend and API server processes."""
-
-    def __init__(self):
-        self.processes: Dict[str, subprocess.Popen] = {}
-        self.running = True
-
-    def start_frontend_server(self) -> bool:
-        """Start the frontend server."""
-        try:
-            frontend_script = os.path.join(os.path.dirname(__file__), 'frontend_server.py')
-
-            cmd = [sys.executable, frontend_script]
-            env = os.environ.copy()
-            env['PYTHONUNBUFFERED'] = '1'
-
-            process = subprocess.Popen(
-                cmd,
-                env=env,
-                stdout=subprocess.PIPE,
-                stderr=subprocess.STDOUT,
-                universal_newlines=True,
-                bufsize=1
-            )
-
-            self.processes['frontend'] = process
-            logger.info(f"Frontend server started with PID: {process.pid}")
-
-            # Start monitoring thread
-            threading.Thread(
-                target=self._monitor_output,
-                args=('frontend', process),
-                daemon=True
-            ).start()
-
-            return True
-
-        except Exception as e:
-            logger.error(f"Failed to start frontend server: {e}")
-            return False
-
-    def start_api_server(self, es_host: str = "http://localhost:9200") -> bool:
-        """Start the API server."""
-        try:
-            cmd = [
-                sys.executable, 'main.py', 'serve',
-                '--es-host', es_host,
-                '--host', '0.0.0.0',
-                '--port', '6002'
-            ]
-
-            env = os.environ.copy()
-            env['PYTHONUNBUFFERED'] = '1'
-            env['ES_HOST'] = es_host
-
-            process = subprocess.Popen(
-                cmd,
-                env=env,
-                stdout=subprocess.PIPE,
-                stderr=subprocess.STDOUT,
-                universal_newlines=True,
-                bufsize=1
-            )
-
-            self.processes['api'] = process
-            logger.info(f"API server started with PID: {process.pid}")
-
-            # Start monitoring thread
-            threading.Thread(
-                target=self._monitor_output,
-                args=('api', process),
-                daemon=True
-            ).start()
-
-            return True
-
-        except Exception as e:
-            logger.error(f"Failed to start API server: {e}")
-            return False
-
-    def _monitor_output(self, name: str, process: subprocess.Popen):
-        """Monitor process output and log appropriately."""
-        try:
-            for line in iter(process.stdout.readline, ''):
-                if line.strip() and self.running:
-                    # Filter out scanner noise for frontend server
-                    if name == 'frontend':
-                        noise_patterns = [
-                            'code 400',
-                            'Bad request version',
-                            'Bad request syntax',
-                            'Bad HTTP/0.9 request type'
-                        ]
-                        if any(pattern in line for pattern in noise_patterns):
-                            continue
-
-                    logger.info(f"[{name}] {line.strip()}")
-
-        except Exception as e:
-            if self.running:
-                logger.error(f"Error monitoring {name} output: {e}")
-
-    def check_servers(self) -> bool:
-        """Check if all servers are still running."""
-        all_running = True
-
-        for name, process in self.processes.items():
-            if process.poll() is not None:
-                logger.error(f"{name} server has stopped with exit code: {process.returncode}")
-                all_running = False
-
-        return all_running
-
-    def stop_all(self):
-        """Stop all servers gracefully."""
-        logger.info("Stopping all servers...")
-        self.running = False
-
-        for name, process in self.processes.items():
-            try:
-                logger.info(f"Stopping {name} server (PID: {process.pid})...")
-
-                # Try graceful shutdown first
-                process.terminate()
-
-                # Wait up to 10 seconds for graceful shutdown
-                try:
-                    process.wait(timeout=10)
-                    logger.info(f"{name} server stopped gracefully")
-                except subprocess.TimeoutExpired:
-                    # Force kill if graceful shutdown fails
-                    logger.warning(f"{name} server didn't stop gracefully, forcing...")
-                    process.kill()
-                    process.wait()
-                    logger.info(f"{name} server stopped forcefully")
-
-            except Exception as e:
-                logger.error(f"Error stopping {name} server: {e}")
-
-        self.processes.clear()
-        logger.info("All servers stopped")
-
-def signal_handler(signum, frame):
-    """Handle shutdown signals."""
-    logger.info(f"Received signal {signum}, shutting down...")
-    if 'manager' in globals():
-        manager.stop_all()
-    sys.exit(0)
-
-def main():
-    """Main function to start all servers."""
-    global manager
-
-    parser = argparse.ArgumentParser(description='Start saas-search servers (multi-tenant)')
-    parser.add_argument('--es-host', default='http://localhost:9200', help='Elasticsearch host')
-    parser.add_argument('--check-dependencies', action='store_true', help='Check dependencies before starting')
-    args = parser.parse_args()
-
-    logger.info("Starting saas-search servers (multi-tenant)...")
-    logger.info(f"Elasticsearch: {args.es_host}")
-
-    # Check dependencies if requested
-    if args.check_dependencies:
-        logger.info("Checking dependencies...")
-        try:
-            import slowapi
-            import anyio
-            logger.info("✓ All dependencies available")
-        except ImportError as e:
-            logger.error(f"✗ Missing dependency: {e}")
-            logger.info("Please run: pip install -r requirements_server.txt")
-            sys.exit(1)
-
-    manager = ServerManager()
-
-    # Set up signal handlers
-    signal.signal(signal.SIGINT, signal_handler)
-    signal.signal(signal.SIGTERM, signal_handler)
-
-    try:
-        # Start servers
-        if not manager.start_api_server(args.es_host):
-            logger.error("Failed to start API server")
-            sys.exit(1)
-
-        # Wait a moment before starting frontend server
-        time.sleep(2)
-
-        if not manager.start_frontend_server():
-            logger.error("Failed to start frontend server")
-            manager.stop_all()
-            sys.exit(1)
-
-        logger.info("All servers started successfully!")
-        logger.info("Frontend: http://localhost:6003")
-        logger.info("API: http://localhost:6002")
-        logger.info("API Docs: http://localhost:6002/docs")
-        logger.info("Press Ctrl+C to stop all servers")
-
-        # Monitor servers
-        while manager.running:
-            if not manager.check_servers():
-                logger.error("One or more servers have stopped unexpectedly")
-                manager.stop_all()
-                sys.exit(1)
-
-            time.sleep(5)  # Check every 5 seconds
-
-    except KeyboardInterrupt:
-        logger.info("Received interrupt signal")
-    except Exception as e:
-        logger.error(f"Unexpected error: {e}")
-    finally:
-        manager.stop_all()
-
-if __name__ == '__main__':
-    main()
\ No newline at end of file
diff --git a/scripts/start_tei_service.sh b/scripts/start_tei_service.sh
index aaea508..666d11e 100755
--- a/scripts/start_tei_service.sh
+++ b/scripts/start_tei_service.sh
@@ -27,14 +27,10 @@ TEI_DTYPE="${TEI_DTYPE:-float16}"
 HF_CACHE_DIR="${HF_CACHE_DIR:-$HOME/.cache/huggingface}"
 TEI_HEALTH_TIMEOUT_SEC="${TEI_HEALTH_TIMEOUT_SEC:-300}"
 
-USE_GPU_RAW="${TEI_USE_GPU:-1}"
-USE_GPU="$(echo "${USE_GPU_RAW}" | tr '[:upper:]' '[:lower:]')"
-if [[ "${USE_GPU}" == "1" || "${USE_GPU}" == "true" || "${USE_GPU}" == "yes" ]]; then
-  USE_GPU="1"
-elif [[ "${USE_GPU}" == "0" || "${USE_GPU}" == "false" || "${USE_GPU}" == "no" ]]; then
-  USE_GPU="0"
-else
-  echo "ERROR: invalid TEI_USE_GPU=${USE_GPU_RAW}. Use 1/0 (or true/false)." >&2
+TEI_DEVICE_RAW="${TEI_DEVICE:-cuda}"
+TEI_DEVICE="$(echo "${TEI_DEVICE_RAW}" | tr '[:upper:]' '[:lower:]')"
+if [[ "${TEI_DEVICE}" != "cuda" && "${TEI_DEVICE}" != "cpu" ]]; then
+  echo "ERROR: invalid TEI_DEVICE=${TEI_DEVICE_RAW}. Use cuda/cpu." >&2
   exit 1
 fi
 
@@ -50,19 +46,19 @@ detect_gpu_tei_image() {
   fi
 }
 
-if [[ "${USE_GPU}" == "1" ]]; then
+if [[ "${TEI_DEVICE}" == "cuda" ]]; then
   if ! command -v nvidia-smi >/dev/null 2>&1 || ! nvidia-smi >/dev/null 2>&1; then
-    echo "ERROR: TEI_USE_GPU=1 but NVIDIA GPU is not available. No CPU fallback." >&2
+    echo "ERROR: TEI_DEVICE=cuda but NVIDIA GPU is not available. No CPU fallback." >&2
     exit 1
   fi
   if ! docker info --format '{{json .Runtimes}}' 2>/dev/null | grep -q 'nvidia'; then
-    echo "ERROR: TEI_USE_GPU=1 but Docker nvidia runtime is not configured." >&2
+    echo "ERROR: TEI_DEVICE=cuda but Docker nvidia runtime is not configured." >&2
     echo "Install and configure nvidia-container-toolkit, then restart Docker." >&2
     exit 1
   fi
   TEI_IMAGE="${TEI_IMAGE:-$(detect_gpu_tei_image)}"
   GPU_ARGS=(--gpus all)
-  TEI_MODE="gpu"
+  TEI_MODE="cuda"
 else
   TEI_IMAGE="${TEI_IMAGE:-ghcr.io/huggingface/text-embeddings-inference:${TEI_VERSION}}"
   GPU_ARGS=()
@@ -81,9 +77,9 @@ if [[ -n "${existing_id}" ]]; then
     if [[ "${current_image}" == *":cuda-"* || "${current_image}" == *":turing-"* ]]; then
       current_is_gpu_image=1
     fi
-    if [[ "${USE_GPU}" == "1" ]]; then
+    if [[ "${TEI_DEVICE}" == "cuda" ]]; then
       if [[ "${current_is_gpu_image}" -eq 1 ]] && [[ "${device_req}" != "null" ]] && [[ "${current_image}" == "${TEI_IMAGE}" ]]; then
-        echo "TEI already running (GPU): ${TEI_CONTAINER_NAME}"
+        echo "TEI already running (CUDA): ${TEI_CONTAINER_NAME}"
         exit 0
       fi
       echo "TEI running with different mode/image; recreating container ${TEI_CONTAINER_NAME}"
diff --git a/scripts/start_translator.sh b/scripts/start_translator.sh
index 5f504c2..4f45adc 100755
--- a/scripts/start_translator.sh
+++ b/scripts/start_translator.sh
@@ -3,13 +3,13 @@
 # Start Translation Service
 #
 
-set -e
+set -euo pipefail
 
 cd "$(dirname "$0")/.."
 source ./activate.sh
 
 TRANSLATION_HOST="${TRANSLATION_HOST:-0.0.0.0}"
-TRANSLATION_PORT="${TRANSLATION_PORT:-${TRANSLATOR_PORT:-6006}}"
+TRANSLATION_PORT="${TRANSLATION_PORT:-6006}"
 
 echo "========================================"
 echo "Starting Translation Service"
diff --git a/scripts/stop.sh b/scripts/stop.sh
index 3b96402..f4b39da 100755
--- a/scripts/stop.sh
+++ b/scripts/stop.sh
@@ -1,9 +1,9 @@
 #!/bin/bash
 
-# Backward-compatible stop entrypoint.
+# Service stop entrypoint.
 # Delegates to unified service controller.
 
-set -e
+set -euo pipefail
 
 cd "$(dirname "$0")/.."
 
diff --git a/scripts/stop_cnclip_service.sh b/scripts/stop_cnclip_service.sh
index df933db..24af68f 100755
--- a/scripts/stop_cnclip_service.sh
+++ b/scripts/stop_cnclip_service.sh
@@ -11,7 +11,7 @@
 #
 ###############################################################################
 
-set -e
+set -euo pipefail
 
 # 颜色定义
 RED='\033[0;31m'
@@ -23,7 +23,7 @@ NC='\033[0m' # No Color
 # 项目路径
 PROJECT_ROOT="$(cd "$(dirname "$0")/.." && pwd)"
 LOG_DIR="${PROJECT_ROOT}/logs"
-PID_FILE="${LOG_DIR}/cnclip_service.pid"
+PID_FILE="${LOG_DIR}/cnclip.pid"
 
 echo -e "${BLUE}========================================${NC}"
 echo -e "${BLUE}停止 CN-CLIP 服务${NC}"
diff --git a/scripts/stop_reranker.sh b/scripts/stop_reranker.sh
deleted file mode 100755
index e397741..0000000
--- a/scripts/stop_reranker.sh
+++ /dev/null
@@ -1,57 +0,0 @@
-#!/bin/bash
-#
-# Stop Reranker Service
-#
-
-set -e
-
-cd "$(dirname "$0")/.."
-
-PID_FILE="logs/reranker.pid"
-RERANKER_PORT="${RERANKER_PORT:-6007}"
-
-echo "========================================"
-echo "Stopping Reranker Service"
-echo "========================================"
-
-if [ -f "${PID_FILE}" ]; then
-  PID="$(cat "${PID_FILE}" 2>/dev/null || true)"
-  if [ -n "${PID}" ] && kill -0 "${PID}" 2>/dev/null; then
-    echo "Stopping PID from file: ${PID}"
-    kill -TERM "${PID}" 2>/dev/null || true
-    sleep 1
-    if kill -0 "${PID}" 2>/dev/null; then
-      kill -KILL "${PID}" 2>/dev/null || true
-    fi
-  fi
-  rm -f "${PID_FILE}"
-fi
-
-PORT_PIDS="$(lsof -ti:${RERANKER_PORT} 2>/dev/null || true)"
-if [ -n "${PORT_PIDS}" ]; then
-  echo "Stopping process on port ${RERANKER_PORT}: ${PORT_PIDS}"
-  for PID in ${PORT_PIDS}; do
-    kill -TERM "${PID}" 2>/dev/null || true
-  done
-  sleep 1
-  PORT_PIDS="$(lsof -ti:${RERANKER_PORT} 2>/dev/null || true)"
-  for PID in ${PORT_PIDS}; do
-    kill -KILL "${PID}" 2>/dev/null || true
-  done
-fi
-
-# Cleanup orphaned vLLM engine workers that may survive a failed startup.
-ENGINE_PIDS="$(pgrep -f 'VLLM::EngineCore' 2>/dev/null || true)"
-if [ -n "${ENGINE_PIDS}" ]; then
-  echo "Stopping orphaned vLLM engine processes: ${ENGINE_PIDS}"
-  for PID in ${ENGINE_PIDS}; do
-    kill -TERM "${PID}" 2>/dev/null || true
-  done
-  sleep 1
-  ENGINE_PIDS="$(pgrep -f 'VLLM::EngineCore' 2>/dev/null || true)"
-  for PID in ${ENGINE_PIDS}; do
-    kill -KILL "${PID}" 2>/dev/null || true
-  done
-fi
-
-echo "Reranker service stopped."
diff --git a/scripts/stop_translator.sh b/scripts/stop_translator.sh
deleted file mode 100755
index 7966874..0000000
--- a/scripts/stop_translator.sh
+++ /dev/null
@@ -1,43 +0,0 @@
-#!/bin/bash
-#
-# Stop Translation Service
-#
-
-set -e
-
-cd "$(dirname "$0")/.."
-
-PID_FILE="logs/translator.pid"
-TRANSLATION_PORT="${TRANSLATION_PORT:-${TRANSLATOR_PORT:-6006}}"
-
-echo "========================================"
-echo "Stopping Translation Service"
-echo "========================================"
-
-if [ -f "${PID_FILE}" ]; then
-  PID="$(cat "${PID_FILE}" 2>/dev/null || true)"
-  if [ -n "${PID}" ] && kill -0 "${PID}" 2>/dev/null; then
-    echo "Stopping PID from file: ${PID}"
-    kill -TERM "${PID}" 2>/dev/null || true
-    sleep 1
-    if kill -0 "${PID}" 2>/dev/null; then
-      kill -KILL "${PID}" 2>/dev/null || true
-    fi
-  fi
-  rm -f "${PID_FILE}"
-fi
-
-PORT_PIDS="$(lsof -ti:${TRANSLATION_PORT} 2>/dev/null || true)"
-if [ -n "${PORT_PIDS}" ]; then
-  echo "Stopping process on port ${TRANSLATION_PORT}: ${PORT_PIDS}"
-  for PID in ${PORT_PIDS}; do
-    kill -TERM "${PID}" 2>/dev/null || true
-  done
-  sleep 1
-  PORT_PIDS="$(lsof -ti:${TRANSLATION_PORT} 2>/dev/null || true)"
-  for PID in ${PORT_PIDS}; do
-    kill -KILL "${PID}" 2>/dev/null || true
-  done
-fi
-
-echo "Translation service stopped."
--
libgit2 0.21.2