fix(deploy): nginx 上游运行期重新解析 + 就绪探针；内存上限按 8C/32G 重新分配

问题：容器重建换 IP 后，网关与 admin-ui 仍往旧 IP 打，/openapi 与 /api
整片 502，必须手动重启 nginx 容器才恢复（2026-11-03 实机踩到）。

编排
- admin-svr / openapi / admin-ui / docs 增加 healthcheck
- admin-ui、gateway、nginx 改 depends_on: condition: service_healthy，
  保证 nginx 启动时后端已就绪
- 新增 deploy/compose.memlimit.yml（原来只存在于某台演示机上）：
  全套上限 8.5G、实测常驻 2G，留给同机其它服务 20G+；
  MySQL buffer pool 128M→768M、max_connections 151→300

nginx
- gateway / admin-ui / iotlcm.conf 示例 统一改 resolver 127.0.0.11 valid=10s
  + upstream zone/resolve：运行期跟随 IP 变化，不必重启（实测 6s 内自愈）

deploy/compose.memlimit.yml

@@ -0,0 +1,175 @@
+# ============================================================
+# iotlcm 内存上限 override —— 8 vCPU / 32 GB 宿主机，与其他服务共用
+#
+# 定位：这台机器是「一机多环境」，iotlcm 只是其中一个占用方。
+#       所以只给全套 20 个服务留 **约 8.5 GB 上限**（实测常驻约 2 GB），
+#       剩余 20+ GB 留给宿主机上要新增的其它服务。
+#
+# 没有本文件会怎样：编排里没有内存限制，任何一个服务泄漏/跑批都会
+# 把整机内存吃到 OOM Killer 随机杀进程（表现为容器反复 Restarting
+# 且日志里看不到业务异常，只能 dmesg 里看到 oom-kill）。
+#
+# 用法（顺序不能反）：
+#   docker compose -f docker-compose.yml -f compose.memlimit.yml up -d
+#
+# 只调 resources（+ MySQL 自身缓存参数），不动镜像、端口、卷、依赖。
+# 换成更小的机器（4 GB）时按比例下调：mysql 1024M / tdengine 512M /
+# worker-svr 768M / 其余 320M；8 GB 机器可用 1024M / 768M / 512M。
+# ============================================================
+
+services:
+  # ---------- 基础组件 ----------
+  mysql:
+    # buffer pool 由 128M（默认）提到 768M，连接数 151 → 300
+    command:
+      - --character-set-server=utf8mb4
+      - --collation-server=utf8mb4_general_ci
+      - --default-time-zone=+08:00
+      - --innodb-buffer-pool-size=768M
+      - --max-connections=300
+    deploy:
+      resources:
+        limits:
+          memory: 2048M
+        reservations:
+          memory: 768M
+
+  redis:
+    deploy:
+      resources:
+        limits:
+          memory: 384M
+        reservations:
+          memory: 96M
+
+  rabbitmq:
+    deploy:
+      resources:
+        limits:
+          memory: 512M
+        reservations:
+          memory: 192M
+
+  tdengine:
+    deploy:
+      resources:
+        limits:
+          memory: 1024M
+        reservations:
+          memory: 256M
+
+  # 一次性建表任务，跑完即退出
+  tdengine-init:
+    deploy:
+      resources:
+        limits:
+          memory: 128M
+
+  # 单节点 consul 常驻约 110~120 MB，给足余量避免贴边
+  consul:
+    deploy:
+      resources:
+        limits:
+          memory: 256M
+        reservations:
+          memory: 96M
+
+  # ---------- 业务服务 ----------
+  admin-svr:
+    deploy:
+      resources:
+        limits:
+          memory: 512M
+        reservations:
+          memory: 128M
+
+  admin-ui:
+    deploy:
+      resources:
+        limits:
+          memory: 64M
+
+  openapi:
+    deploy:
+      resources:
+        limits:
+          memory: 384M
+        reservations:
+          memory: 128M
+
+  upstream-chan:
+    deploy:
+      resources:
+        limits:
+          memory: 384M
+        reservations:
+          memory: 128M
+
+  upstream-proxy:
+    deploy:
+      resources:
+        limits:
+          memory: 384M
+        reservations:
+          memory: 128M
+
+  # 全量同步（流量/状态/IMEI）时内存抬升最明显，按文档「500 MB ~ 1 GB」给 1 GB
+  worker-svr:
+    deploy:
+      resources:
+        limits:
+          memory: 1024M
+        reservations:
+          memory: 256M
+
+  app-server:
+    deploy:
+      resources:
+        limits:
+          memory: 512M
+        reservations:
+          memory: 128M
+
+  app-web:
+    deploy:
+      resources:
+        limits:
+          memory: 64M
+
+  sms:
+    deploy:
+      resources:
+        limits:
+          memory: 192M
+
+  docs:
+    deploy:
+      resources:
+        limits:
+          memory: 64M
+
+  # ---------- profile 可选服务 ----------
+  bot:
+    deploy:
+      resources:
+        limits:
+          memory: 192M
+
+  onelink-mock:
+    deploy:
+      resources:
+        limits:
+          memory: 192M
+
+  nginx:
+    deploy:
+      resources:
+        limits:
+          memory: 128M
+
+  # 单端口入口（with-gateway）；之前没有任何上限
+  gateway:
+    deploy:
+      resources:
+        limits:
+          memory: 256M

deploy/config/admin-ui/default.conf

@@ -1,3 +1,9 @@
+# DNS 解析：nginx 只在启动时解析一次 upstream 主机名，admin-svr 重建换 IP 后
+# /api/ 会一直 502。127.0.0.11 是 Docker 内置 DNS，resolve + zone 让运行期自动跟随。
+resolver 127.0.0.11 valid=10s ipv6=off;
+
+upstream gw_admin_svr { zone gw_admin_svr 64k; server admin-svr:8008 resolve; }
+
 server {
     listen 80;
     root /usr/share/nginx/html;
@@ -40,7 +46,7 @@ server {
         proxy_set_header Upgrade $http_upgrade;
         proxy_http_version 1.1;
         chunked_transfer_encoding off;
-        proxy_pass http://admin-svr:8008/api/v1/;
+        proxy_pass http://gw_admin_svr/api/v1/;
         proxy_redirect off;
         proxy_set_header Host $host;
         proxy_set_header X-Real-IP $remote_addr;

deploy/config/gateway/nginx.conf

@@ -16,9 +16,14 @@
 # 小程序 / H5 端的静态资源是根路径部署的，不能挂在同一域名的子路径下，
 # 需要单独域名或单独端口（见 config/nginx/iotlcm.conf 的 app 段）。
 
-upstream gw_admin_ui  { server admin-ui:80; }
+# DNS 解析：nginx 默认只在启动时把 upstream 主机名解析一次，容器重建/重启换了 IP
-upstream gw_openapi   { server openapi:8000; }
+# 就一直往旧地址打（2026-11-03 踩到：整站 /openapi 全 502，重启网关才恢复）。
-upstream gw_docs      { server docs:80; }
+# 127.0.0.11 是 Docker 内置 DNS；下面的 resolve + zone 让它在运行期自动跟随。
+resolver 127.0.0.11 valid=10s ipv6=off;
+
+upstream gw_admin_ui  { zone gw_admin_ui 64k;  server admin-ui:80  resolve; }
+upstream gw_openapi   { zone gw_openapi 64k;   server openapi:8000  resolve; }
+upstream gw_docs      { zone gw_docs 64k;      server docs:80      resolve; }
 
 server {
     listen 80;

deploy/config/nginx/iotlcm.conf

@@ -8,28 +8,42 @@
 # 把 your-domain.example.com 替换为你的实际域名。
 # ============================================================
 
+# ---------- 后端地址解析 ----------
+# 这些 upstream 写的是 compose 服务名。nginx 默认只在**启动时**解析一次，
+# 容器重建后 IP 变了就会持续 502（需重启 nginx）。resolve + zone 让它在运行期自动跟随。
+# 注意：127.0.0.11 是 Docker 内置 DNS，仅在容器内有效：
+#   ・在容器内用本文件   → 保持下面的 resolver 不动
+#   ・把本文件并入宿主机 nginx → 把 resolver 换成宿主机 DNS（或删掉 resolver 与各 upstream 的 resolve 参数）
+resolver 127.0.0.11 valid=10s ipv6=off;
+
 upstream iotlcm_admin_api {
-    server admin-svr:8008;
+    zone iotlcm_admin_api 64k;
+    server admin-svr:8008 resolve;
 }
 
 upstream iotlcm_admin_ui {
-    server admin-ui:80;
+    zone iotlcm_admin_ui 64k;
+    server admin-ui:80 resolve;
 }
 
 upstream iotlcm_openapi_api {
-    server openapi:8000;
+    zone iotlcm_openapi_api 64k;
+    server openapi:8000 resolve;
 }
 
 upstream iotlcm_docs {
-    server docs:80;
+    zone iotlcm_docs 64k;
+    server docs:80 resolve;
 }
 
 upstream iotlcm_app_web {
-    server app-web:80;
+    zone iotlcm_app_web 64k;
+    server app-web:80 resolve;
 }
 
 upstream iotlcm_app_server {
-    server app-server:8010;
+    zone iotlcm_app_server 64k;
+    server app-server:8010 resolve;
 }
 
 # ---------- OpenAPI 对外域名 ----------

deploy/docker-compose.yml

@@ -175,6 +175,14 @@ services:
         condition: service_healthy
       rabbitmq:
         condition: service_healthy
+    # 就绪探针：nginx 类容器只在启动时解析一次后端 IP，
+    # 没有它就无法用 depends_on 的 service_healthy 保证启动顺序（见 gateway）
+    healthcheck:
+      test: ["CMD", "wget", "--quiet", "--output-document=/dev/null", "http://127.0.0.1:8008/api.json"]
+      interval: 10s
+      timeout: 3s
+      retries: 5
+      start_period: 20s
 
   admin-ui:
     image: 'crpi-qflrqrxoohlkkvgc.cn-chengdu.personal.cr.aliyuncs.com/iotlcm/iotlcm:iotlcm-admin-ui-v1.0.4'
@@ -191,7 +199,14 @@ services:
     networks:
       - iotlcm-network
     depends_on:
-      - admin-svr
+      admin-svr:
+        condition: service_healthy
+    healthcheck:
+      test: ["CMD", "wget", "--quiet", "--output-document=/dev/null", "http://127.0.0.1/"]
+      interval: 10s
+      timeout: 3s
+      retries: 5
+      start_period: 5s
 
   openapi:
     image: 'crpi-qflrqrxoohlkkvgc.cn-chengdu.personal.cr.aliyuncs.com/iotlcm/iotlcm:iotlcm-openapi-v1.0.4'
@@ -213,6 +228,12 @@ services:
         condition: service_healthy
       rabbitmq:
         condition: service_healthy
+    healthcheck:
+      test: ["CMD", "wget", "--quiet", "--output-document=/dev/null", "http://127.0.0.1:8000/api.json"]
+      interval: 10s
+      timeout: 3s
+      retries: 5
+      start_period: 20s
 
   upstream-chan:
     image: 'crpi-qflrqrxoohlkkvgc.cn-chengdu.personal.cr.aliyuncs.com/iotlcm/iotlcm:iotlcm-upstream-chan-v1.0.4'
@@ -370,6 +391,12 @@ services:
       - "127.0.0.1:8019:80"
     networks:
       - iotlcm-network
+    healthcheck:
+      test: ["CMD", "wget", "--quiet", "--output-document=/dev/null", "http://127.0.0.1/docs/"]
+      interval: 10s
+      timeout: 3s
+      retries: 5
+      start_period: 10s
 
   # ---------------- 可选：运营商接口挡板 ----------------
   # 模拟中国移动 OneLink 接口，用于在没有真实运营商环境时联调整条链路：
@@ -418,10 +445,14 @@ services:
     networks:
       - iotlcm-network
     depends_on:
-      - admin-ui
+      admin-ui:
-      - openapi
+        condition: service_healthy
-      - app-web
+      openapi:
-      - docs
+        condition: service_healthy
+      docs:
+        condition: service_healthy
+      app-web:
+        condition: service_started
     profiles:
       - with-nginx
 
@@ -437,10 +468,16 @@ services:
       - ./data/logs/gateway:/var/log/nginx
     networks:
       - iotlcm-network
+    # 后端必须"健康"后才起网关：否则 nginx 会在后端拿到 IP 之前把 upstream
+    # 解析结果缓存下来，容器重启换 IP 后所有请求变成 502（2026-11-03 踩过）。
+    # 配置里另有 resolver + resolve，可在运行期自动跟随 IP 变化。
     depends_on:
-      - admin-ui
+      admin-ui:
-      - openapi
+        condition: service_healthy
-      - docs
+      openapi:
+        condition: service_healthy
+      docs:
+        condition: service_healthy
     profiles:
       - with-gateway