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
mirror-to-gitee / mirror (push) Successful in 9s

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)
This commit is contained in:
MiFaZhan
2026-11-03 10:40:48 +08:00
parent 70749bea7a
commit e98f8561f7
5 changed files with 255 additions and 18 deletions
+175
View File
@@ -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