feat: 部署改为 git 驱动，容器以宿主用户运行

服务器实测可经 Cloudflare 443 访问 Gitea 且 git 协议正常（此前我只测了 13000 端口就断言不可达，是错的），因此不再需要 tar + SFTP。

- docker-compose: 增加 user: "1000:1000"。容器此前以 root 运行，写进挂载目录的每轮 jsonl 产物都是 root 属主，导致宿主用户连自己的部署目录都挪不动 —— 这在把部署迁到 /mnt/data 时实际发生了
- deploy.sh: 一条命令走完 拉代码 →（webui/ 有改动时）重建前端 → 重启容器。前端产物 api/webui 是 gitignore 的，git pull 带不过来，必须在服务器上重建一次
- Dockerfile: 补 npm 包。corepack 只管 yarn/pnpm 不管 npm，而前端要在服务器上重建；这样服务器只需要 Docker，不必另配 Node 环境

Dockerfile

@@ -29,7 +29,9 @@ ENV PYTHONUNBUFFERED=1 \
 # every stored timestamp in UTC. nodejs is for PyExecJS: douyin/help.py compiles
 # libs/douyin.js at *import* time, and because main.py imports every platform,
 # that single platform being importable-or-not decides whether the whole app
-# (and the environment self-check) comes up.
+# (and the environment self-check) comes up. npm rides along so the WebUI can be
+# rebuilt on the server (see deploy.sh) instead of only on a workstation --
+# corepack is present but does not cover npm, only yarn and pnpm.
 #
 # The pip mirror is set for the same reason as the apt one: this host's route to
 # the public index is slow.
@@ -44,7 +46,7 @@ RUN set -eux; \
     apt-get install -y --no-install-recommends \
         build-essential pkg-config default-libmysqlclient-dev \
         libgl1 libglib2.0-0 libsm6 libxext6 libxrender1 libxcb1 libgomp1 \
-        tzdata nodejs; \
+        tzdata nodejs npm; \
     rm -rf /var/lib/apt/lists/*
 
 WORKDIR /app

deploy.sh

@@ -0,0 +1,60 @@
+#!/usr/bin/env bash
+#
+# 更新这台机器上的部署。用法：
+#
+#     ./deploy.sh
+#
+# 为什么需要脚本而不是一句 `git pull && docker compose up -d`：
+# 前端产物 api/webui 是 gitignore 的（它由 vite 生成），git pull 带不过来。
+# 所以代码更新之后必须在服务器上重建一次前端，否则页面还是旧的。
+# 这一步在容器里做，好处是服务器不需要装 Node —— 只有 Docker。
+#
+# node_modules 和 npm 缓存都留在挂载目录内，重复构建不会重新下载。
+set -euo pipefail
+cd "$(dirname "$0")"
+
+# 用镜像里的解释器跑，宿主机的 Python 版本无关。
+IMAGE=mediacrawler:latest
+
+before=$(git rev-parse HEAD)
+git pull --ff-only
+after=$(git rev-parse HEAD)
+
+if [ "$before" = "$after" ]; then
+    echo "== 代码已是最新（$after）"
+else
+    echo "== 代码更新 $before -> $after"
+    git --no-pager log --oneline "$before..$after" | sed 's/^/   /'
+fi
+
+# 前端重建的两种情况：产物根本不存在（首次部署），或 webui/ 有改动。
+if [ ! -f api/webui/index.html ]; then
+    need_build=1
+    reason="前端产物不存在"
+elif [ "$before" != "$after" ] && ! git diff --quiet "$before" "$after" -- webui/; then
+    need_build=1
+    reason="webui/ 有改动"
+else
+    need_build=0
+    reason=""
+fi
+
+if [ "$need_build" = "1" ]; then
+    echo "== 重建前端（$reason）"
+    # -u 1000:1000 而不是 root：这里产出的文件要留在这个目录里给后面用，
+    # 以 root 生成的 node_modules 会让下次构建和人工清理都变得别扭。
+    # HOME 指向挂载目录，这样 npm 的缓存在宿主机上，重建时能复用。
+    docker run --rm \
+        -u 1000:1000 \
+        -w /app/webui \
+        -v "$PWD:/app" \
+        -e HOME=/app/webui \
+        -e npm_config_registry=https://registry.npmmirror.com \
+        "$IMAGE" sh -c 'npm ci --no-audit --no-fund && npm run build'
+else
+    echo "== 前端无改动，跳过构建"
+fi
+
+echo "== 重启容器"
+docker compose up -d
+docker compose ps

docker-compose.yml

@@ -5,6 +5,13 @@ services:
     container_name: mediacrawler
     restart: unless-stopped
 
+    # Run as the user that owns this checkout. Without it the container is root,
+    # and every file it writes into the mounted tree -- the crawler's per-run
+    # jsonl output above all -- comes out root-owned. That does not break the app,
+    # but it does lock the operator out of moving or deleting their own
+    # deployment, which is exactly what happened the first time this was deployed.
+    user: "1000:1000"
+
     # host networking is a requirement, not a convenience: the crawler attaches
     # to the operator's Chrome at 127.0.0.1:9222, and inside a bridge network
     # that loopback is the container's own, where no browser is listening.
@@ -20,8 +27,7 @@ services:
 
     volumes:
       # The code is mounted rather than baked in, so shipping a change is
-      # "re-upload the tarball, restart" instead of an image rebuild. Only the
-      # dependencies live in the image, because those are the expensive part and
-      # they change rarely -- rebuild only when requirements.txt or the
-      # Dockerfile itself changes.
+      # "git pull, restart" instead of an image rebuild. Only the dependencies
+      # live in the image, because those are the expensive part and they change
+      # rarely -- rebuild only when requirements.txt or the Dockerfile changes.
       - ./:/app