onesvm-browser-server/stacks/browser-server.yml

# stacks/browser-server.yml — onesvm-browser-server Swarm stack（Contract §5-A5）。
#
# 权威对齐：docs/deploy-prod-preset-20260901.md §2 服务清单/镜像/资源/端口（逐字段）；
# stack 必备字段对齐 onesvm-dev-md/base/deployment-rules.md §1.3
# （每服务 TZ / placement / restart_policy / resources.limits / healthcheck，backup 除外）。
# 测试期全钉 mgr1（placement node.hostname==swarm-mgr1）；scheduler 因 SQLite 卷必须钉。
#
# 密钥注入面（deploy.env 提供，禁默认值禁入 git）：
#   BROWSER_SERVER_KEYS_SEED / BROWSER_SERVER_ADMIN_TOKEN / PROXY_SUB_URLS / SEARXNG_SECRET
# 部署：docker stack deploy -c stacks/browser-server.yml --resolve-image never browser-server
# 红线：本文件本轮仅交付，不执行部署（Contract §9）。
#
# O13 镜像 tag 约定：自研镜像 onesvm/browser-server:dev / onesvm/trafilatura-http:dev
# 日常迭代复用覆盖；第三方镜像 tag + digest 双锁（deployment-rules §2.3）。

version: "3.8"

x-restart: &restart
  restart_policy:
    condition: on-failure
    delay: 5s
    max_attempts: 3
    window: 60s

x-placement-mgr1: &placement-mgr1
  placement:
    constraints:
      - node.hostname==swarm-mgr1

services:
  # ---------- 控制层（自研三角色，单二进制 role 分流） ----------

  gateway:
    image: onesvm/browser-server:dev
    command: ["-role=gateway"]
    environment:
      TZ: Asia/Shanghai
      BROWSER_SERVER_KEYS_SEED: ${BROWSER_SERVER_KEYS_SEED:?required}
      BROWSER_SERVER_ADMIN_TOKEN: ${BROWSER_SERVER_ADMIN_TOKEN:?required}
      BROWSER_SERVER_LISTEN_gateway: ":8640"
      BROWSER_SERVER_SCHEDULER_URL: http://scheduler:8641
      # gateway 与 scheduler 共库（gateway 只读 consumers/api_keys，写 audit/quota）。
      BROWSER_SERVER_DB_PATH: /data/browser-server.db
    volumes:
      - browser-server-data:/data      # 与 scheduler 共库（gateway 只读写 consumers/api_keys/audit/quota；
                                       # SQLite 需目录可写以建 WAL 伴生文件，读写纪律由应用层单写者约定保证）
    ports:
      - "8640:8640"                    # host ingress（对外唯一入口；反代前缀 /bs-api）
    networks:
      # vlepontas-casa-net 附加（D10，对齐 WSG 先例）；长格式显式声明 alias
      # browser-server（V3 arch-review D-2：compose 校验器不接受列表内映射，
      # 须用长格式；alias 防 vlepontas 消费方 DNS 解析失败）
      browser-server-net:
        aliases: []
      casa-net:
        aliases:
          - browser-server
    deploy:
      replicas: 1
      <<: *restart
      placement:
        constraints:
          - node.hostname==swarm-mgr1        # 测试期钉 mgr1（与共库卷同节点）
      resources:
        limits:
          memory: 64m
          cpus: "0.25"
    healthcheck:
      test: ["CMD", "/browser-server", "-healthz-probe"]   # 见下方说明（distroless 无 curl/wget）
      interval: 30s
      timeout: 3s
      retries: 3
      start_period: 10s

  scheduler:
    image: onesvm/browser-server:dev
    command: ["-role=scheduler"]
    environment:
      TZ: Asia/Shanghai
      BROWSER_SERVER_LISTEN_scheduler: ":8641"
      BROWSER_SERVER_DB_PATH: /data/browser-server.db
      BROWSER_SERVER_ADMIT_MAX: "60"
      BROWSER_SERVER_WORKERS: "8"
      BROWSER_SERVER_SEARXNG_CN_URL: http://searxng-cn:8080
      BROWSER_SERVER_SEARXNG_GLOBAL_URL: http://searxng-global:8080
      BROWSER_SERVER_TRAFILATURA_URL: http://trafilatura:8080
      BROWSER_SERVER_LIGHTPANDA_CDP_HOST: lightpanda:9222
      BROWSER_SERVER_SHELL_CDP_HOST: headless-shell:9222
      BROWSER_SERVER_PROXYMANAGER_URL: http://proxymanager:8642
      # MIHOMO 配置目录：proxymanager 写 provider 到挂载卷，mihomo 同卷读取。
      BROWSER_SERVER_MIHOMO_CONFIG_DIR: /data/mihomo
      MIHOMO_CONTROLLER_URL: http://mihomo:19090
    volumes:
      - browser-server-data:/data      # named volume 钉 mgr1（SQLite 单写者，禁跨节点）
    networks:
      - browser-server-net
    deploy:
      replicas: 1
      <<: *restart
      placement:
        constraints:
          - node.hostname==swarm-mgr1
      resources:
        limits:
          memory: 64m
          cpus: "0.5"
    healthcheck:
      test: ["CMD", "/browser-server", "-role=scheduler", "-healthz-probe"]
      interval: 30s
      timeout: 3s
      retries: 3
      start_period: 10s

  proxymanager:
    image: onesvm/browser-server:dev
    command: ["-role=proxymanager"]
    environment:
      TZ: Asia/Shanghai
      BROWSER_SERVER_LISTEN_proxymanager: ":8642"
      PROXY_SUB_URLS: ${PROXY_SUB_URLS:?required}
      # 订阅拉取专用 bootstrap（非密钥；缺省空=直连）。mgr1 直连机场被墙，
      # 复用 mgr3 central-proxy :7890（deploy-preset §1）。探活仍走自家 mihomo。
      BROWSER_SERVER_SUB_FETCH_PROXY: ${BROWSER_SERVER_SUB_FETCH_PROXY:-}
      MIHOMO_CONTROLLER_URL: http://mihomo:19090
      # provider 文件写到 mihomo 挂载卷（stack 把默认 /data/mihomo 指到 mihomo 同卷挂载点）。
      BROWSER_SERVER_MIHOMO_CONFIG_DIR: /data/mihomo
      # 热载路径必须是 mihomo 容器内视角（同卷挂 /root/.config/mihomo）。
      BROWSER_SERVER_MIHOMO_RELOAD_PATH: /root/.config/mihomo/config.yaml
      BROWSER_SERVER_DB_PATH: /data/browser-server.db
    volumes:
      - browser-server-data:/data      # rules 表共享（gateway policy 联动）
      - browser-server-mihomo:/data/mihomo   # mihomo provider 写入点（与 mihomo 服务同卷）
    networks:
      - browser-server-net
    deploy:
      replicas: 1
      <<: *restart
      placement:
        constraints:
          - node.hostname==swarm-mgr1
      resources:
        limits:
          memory: 32m
          cpus: "0.1"
    healthcheck:
      test: ["CMD", "/browser-server", "-role=proxymanager", "-healthz-probe"]
      interval: 30s
      timeout: 3s
      retries: 3
      start_period: 10s

  # ---------- 数据面 ----------

  mihomo:
    image: metacubex/mihomo:v1.19.24@sha256:9e37208fae8afa4c8b83d14ff2e9771b99178ebaf65a2c4fb388bc67ecefe4dc
    volumes:
      - browser-server-mihomo:/root/.config/mihomo   # proxymanager 写 config.yaml/providers 到此（PM 侧 BROWSER_SERVER_MIHOMO_CONFIG_DIR=/data/mihomo 指同卷）
    networks:
      - browser-server-net
    deploy:
      replicas: 1
      <<: *restart
      placement:
        constraints:
          - node.hostname==swarm-mgr1
      resources:
        limits:
          # 69 节点 + 订阅自带 dns/geosite 会加载 geoip（64m 实测 OOM 137）。
          # 档标仍按 L2 超配；瘦 dns 后可回 64m（O12 校准）。
          memory: 128m
    healthcheck:
      # mihomo controller 19090 REST 探活（仅 overlay；busybox sh 无 /dev/tcp，
      # 用 busybox wget 打 /version——controller 无 secret 时返回 JSON）。
      test: ["CMD-SHELL", "wget -q -O /dev/null http://127.0.0.1:19090/version || exit 1"]
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 15s

  # ---------- Worker 层（第三方/自构建引擎，全部 pin） ----------

  searxng-cn:
    image: searxng/searxng:2026.8.29-d226b78bc@sha256:61ad36fcf71eb300f7d64216a6a04e010754755c7f5d33a526bd52e597e78c69
    volumes:
      - ./searxng-cn-settings.yml:/etc/searxng/settings.yml:ro
    environment:
      TZ: Asia/Shanghai
      SEARXNG_BASE_URL: http://searxng-cn:8080/
    networks:
      - browser-server-net
    deploy:
      replicas: 1
      <<: *restart
      placement:
        constraints:
          - node.hostname==swarm-mgr1
      resources:
        limits:
          memory: 192m
    healthcheck:
      test: ["CMD", "wget", "-q", "-O", "/dev/null", "http://127.0.0.1:8080/healthz"]
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 20s

  searxng-global:
    image: searxng/searxng:2026.8.29-d226b78bc@sha256:61ad36fcf71eb300f7d64216a6a04e010754755c7f5d33a526bd52e597e78c69
    volumes:
      - ./searxng-global-settings.yml:/etc/searxng/settings.yml:ro
    environment:
      TZ: Asia/Shanghai
      SEARXNG_BASE_URL: http://searxng-global:8080/
    networks:
      - browser-server-net
    deploy:
      replicas: 1
      <<: *restart
      placement:
        constraints:
          - node.hostname==swarm-mgr1
      resources:
        limits:
          memory: 192m
    healthcheck:
      test: ["CMD", "wget", "-q", "-O", "/dev/null", "http://127.0.0.1:8080/healthz"]
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 20s

  trafilatura:
    build:
      context: ./trafilatura
    image: onesvm/trafilatura-http:dev
    environment:
      TZ: Asia/Shanghai
    networks:
      - browser-server-net
    deploy:
      replicas: 1
      <<: *restart
      placement:
        constraints:
          - node.hostname==swarm-mgr1
      resources:
        limits:
          memory: 64m
    healthcheck:
      test: ["CMD", "python", "-c", "import urllib.request;urllib.request.urlopen('http://127.0.0.1:8080/health', timeout=3)"]
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 15s

  lightpanda:
    image: lightpanda/browser:0.3.7@sha256:447598e4c103b70b0fb610813e72795da44341c564e1ec872c3a490c1bc6aebe
    command: ["/bin/lightpanda", "serve", "--host", "0.0.0.0", "--port", "9222", "--log_level", "info"]
    environment:
      TZ: Asia/Shanghai
      # 境外出口统一走 mihomo（对齐 bench compose-lightpanda-proxy.yml 的 proxy env；
      # 域名级直连/代理由 mihomo 规则裁决）。
      HTTP_PROXY: http://mihomo:17890
      HTTPS_PROXY: http://mihomo:17890
      http_proxy: http://mihomo:17890
      https_proxy: http://mihomo:17890
      NO_PROXY: 127.0.0.1,localhost
      no_proxy: 127.0.0.1,localhost
    networks:
      - browser-server-net
    deploy:
      replicas: 1
      <<: *restart
      placement:
        constraints:
          - node.hostname==swarm-mgr1
      resources:
        limits:
          memory: 320m
    # lightpanda 镜像内无 curl/wget/sh 工具链（distroless 风格），CDP 探活由
    # scheduler 适配器 Health 上报承担（healthcheck 缺省即 deployment-rules 例外，注明）。

  headless-shell:
    image: chromedp/headless-shell:151.0.7922.109@sha256:5f877a2a559dea1a99fb750da695d28a020cdd49db660aead6c78a46e3c7dd50
    command:
      - --disable-dev-shm-usage
      # 境外出口统一走 mihomo（对齐 bench compose：--proxy-server=http://mihomo:17890）。
      - --proxy-server=http://mihomo:17890
    # ENTRYPOINT /headless-shell/run.sh 已带 --no-sandbox、--remote-debugging-port=9223
    # 与 socat 9222→9223 转发（bench compose 实测复用），不得覆盖端口。
    shm_size: 128m
    networks:
      - browser-server-net
    deploy:
      replicas: 0                      # 按需槽：平时 0，scheduler 经 Swarm API 拉起（空闲 10min 回收）
      <<: *restart
      placement:
        constraints:
          - node.hostname==swarm-mgr1
      resources:
        limits:
          memory: 400m
    healthcheck:
      # headless-shell 基于 debian + socat（有 shell），探 CDP 调试端口。
      test: ["CMD", "sh", "-c", "exec 3<>/dev/tcp/127.0.0.1/9222 || exit 1"]
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 10s

  # ---------- 备份 sidecar（对齐组织 NAS sidecar 惯例） ----------

  backup:
    # 简化实现：每日 01:00（Asia/Shanghai）sqlite3 .backup → tar 到 /backup 卷。
    # 说明：alpine 基础镜像无 sqlite3 包，这里用 busybox sleep 循环 + cp 快照
    #（WAL 模式下 cp 非原子；严格一致性备份轮换 sqlite3/amd64 镜像后升级 .backup）。
    image: alpine:3.20@sha256:c64c687cbea9300178b30c95835354e34c4e4febc4badfe27102879de0483b5e
    command:
      - /bin/sh
      - -c
      - |
        # 每日 01:00 Asia/Shanghai 备份（TZ 已设 Asia/Shanghai）：
        # 1) sleep 到次日 01:00；2) cp 数据库（WAL 模式 + checkpoint 后主库自洽）；
        # 3) tar 打包进 /backup，保留最近 14 份。
        while true; do
          now=$$(date +%s); target=$$(date -d "tomorrow 01:00" +%s 2>/dev/null || echo $$((now + 86400)));
          sleep $$((target - now));
          ts=$$(date +%Y%m%d-%H%M%S);
          cp /data/browser-server.db /data/browser-server.db-wal /tmp/ 2>/dev/null || true;
          tar -czf /backup/browser-server-$$ts.tar.gz -C /data browser-server.db browser-server.db-wal browser-server.db-shm 2>/dev/null \
            || tar -czf /backup/browser-server-$$ts.tar.gz -C /data browser-server.db;
          ls -1t /backup/browser-server-*.tar.gz | tail -n +15 | xargs -r rm -f;
        done        
    environment:
      TZ: Asia/Shanghai
    volumes:
      - browser-server-data:/data:ro
      - browser-server-backup:/backup
    networks:
      - browser-server-net
    deploy:
      replicas: 1
      <<: *restart
      placement:
        constraints:
          - node.hostname==swarm-mgr1
      resources:
        limits:
          memory: 32m
    # 备份 sidecar 无对外服务，healthcheck 缺省（deployment-rules §1.3 例外，对齐 pg 备份 sidecar 惯例）。

  # ---------- 网络 / 卷 ----------

networks:
  browser-server-net:
    driver: overlay
    name: browser-server-net
  casa-net:
    external: true
    name: vlepontas-casa-prod-net

volumes:
  browser-server-data: {}              # SQLite 库（scheduler 单写者；gateway 只读；钉 mgr1 由 placement 保证）
  browser-server-mihomo: {}            # mihomo 配置/provider（PM 写、mihomo 读）
  browser-server-backup: {}            # 备份产物（NAS 侧同步由部署轮另议）