build: 添加 all-in-one 部署方案及相关配置脚本

新增内容包括:
- all-in-one 镜像构建脚本与 Dockerfile,整合前后端及 Nginx
- K3s 部署清单与最小化版本
- 健康检查脚本、Nginx 代理配置
- 生产环境数据库连接池优化配置
- 序列同步 SQL 脚本与清单校验工具
- .dockerignore、.gitignore 及部署文档
main
kangwenjing 2026-10-29 15:21:42 +08:00
parent 7f911421b8
commit cf5e84da0a
15 changed files with 1281 additions and 8 deletions

32
.dockerignore 100644
View File

@ -0,0 +1,32 @@
# 仓库根目录作为构建上下文时使用(见 deploy/all-in-one/Dockerfile)
.git
.idea
.vscode
.claude
.workbuddy
.qoder
docs
doc
dist
sql
tmp
uploads
deploy/*.tar
node_modules
frontend/node_modules
frontend1/node_modules
frontend/dist
frontend1/dist
frontend/.cert
frontend1/.cert
backend/src
backend/target/classes
backend/target/generated-sources
backend/target/generated-test-sources
backend/target/maven-archiver
backend/target/maven-status
backend/target/surefire-reports
backend/target/test-classes
backend/target/unis-crm-backend-1.0.0-SNAPSHOT.jar.original
**/*.log
**/.DS_Store

3
.gitignore vendored
View File

@ -17,4 +17,5 @@ frontend1/.cert/
*.log
tmp/
docs/
docs/
deploy/*.tar

View File

@ -17,16 +17,36 @@ spring:
max-file-size: 500MB
max-request-size: 600MB
datasource:
url: jdbc:postgresql://192.168.124.202:5432/unis_crm
username: postgres
# 生产 DB / Redis 走 K8s NodePort,空闲连接会被静默回收,需要显式超时兜底
url: jdbc:postgresql://10.100.51.227:30532/unis_scc?ApplicationName=unis-crm-prod&connectTimeout=10&socketTimeout=120&tcpKeepAlive=true
username: root
password: unis@123
driver-class-name: org.postgresql.Driver
hikari:
pool-name: crm-prod
# 实例 max_connections=100 且与 rda/unis 库共用,池子保持保守
maximum-pool-size: 10
minimum-idle: 2
# 拿不到连接时 10s 快速失败,避免请求线程堆积
connection-timeout: 10000
# 连接最长存活 5min,短于 NodePort 空闲连接回收时间
max-lifetime: 300000
keepalive-time: 120000
# 2min 未归还判定为泄漏;阈值留出大文件上传等正常慢操作余量
leak-detection-threshold: 120000
# 单条 SQL 60s 上限;事务内空闲 2min 判定为泄漏,避免长事务占锁并阻塞 vacuum
connection-init-sql: "SET statement_timeout = 60000; SET idle_in_transaction_session_timeout = 120000;"
data:
redis:
host: 192.168.124.202
port: 6379
password: zghz@123
database: 12
host: 10.100.51.227
port: 30379
password: unis@123
database: 10
# 命令与建连各 5s 超时,Redis 抖动时快速失败而不是挂住线程
timeout: 5s
connect-timeout: 5s
lettuce:
shutdown-timeout: 500ms
mybatis-plus:
mapper-locations: classpath*:/mapper/**/*.xml
@ -84,7 +104,7 @@ unisbase:
access-token-safety-seconds: 120
oms:
enabled: ${OMS_ENABLED:true}
base-url: ${OMS_BASE_URL:http://192.168.124.202:28081}
base-url: ${OMS_BASE_URL:https://oms.unisspace.com/prod-api}
api-key: ${OMS_API_KEY:c7f858d0-30b8-4b7f-9ea1-0ccf5ceb1c54}
api-key-header: ${OMS_API_KEY_HEADER:apiKey}
user-info-path: ${OMS_USER_INFO_PATH:/api/v1/user/info}

113
deploy/README.md 100644
View File

@ -0,0 +1,113 @@
# unis_scc 镜像构建与 k3s 部署
## 两种打包形态
| 形态 | 镜像 | 清单文件 | 说明 |
| --- | --- | --- | --- |
| **all-in-one(默认)** | `public/unis_scc` | `k3s-unis-scc-allinone.yaml` | 一个镜像一个 Pod:nginx 托管两个前端 + 同容器跑 Spring Boot |
| 分体 | `public/unis_scc-backend` / `-frontend` / `-system` | `k3s-unis-scc.yaml` | 三个镜像三个 Deployment,可独立升级回滚 |
all-in-one 容器内端口:`80` CRM 前端、`8090` 系统管理前端、`8080` 后端 API。任一子进程退出,容器整体退出,由 k8s 重启。
## 构建并推送
```bash
docker login docker.oa.unissense.tech -u admin
./deploy/build-and-push.sh # all-in-one,镜像名 unis_scc
./deploy/build-and-push.sh app
./deploy/build-and-push.sh backend frontend system # 分体三镜像
TAG=202610291500 ./deploy/build-and-push.sh # 指定 tag(默认取当前时间 + latest)
PUSH=0 ./deploy/build-and-push.sh # 本地验证,不推送
```
脚本会先用 JDK 17 执行 `mvn -DskipTests package`。后端依赖 `com.unisbase`,是只装在本地 `~/.m2` 的内部包,所以 jar 必须在宿主机打好,再由镜像 COPY。Apple Silicon 上走 amd64 模拟,前端构建约需几分钟。
all-in-one 的构建上下文是仓库根目录(`-f deploy/all-in-one/Dockerfile .`),根 `.dockerignore` 控制上下文体积。
## k3s 部署
```bash
# 节点可直连 Harbor:public 项目允许匿名拉取
kubectl apply -f deploy/k3s-unis-scc-allinone.yaml
kubectl -n unis-scc rollout status deploy/unis-scc
# 节点访问不了 Harbor:离线导入
docker save -o /tmp/unis_scc.tar docker.oa.unissense.tech/public/unis_scc:latest
scp /tmp/unis_scc.tar node:/tmp/ && ssh node 'k3s ctr images import /tmp/unis_scc.tar'
```
访问入口(NodePort):CRM 前端 `:30082`、系统管理前端 `:30083`、后端 API `:30081`。
## 配置说明
数据库与 Redis 走镜像内置的 `application-prod.yml`(PG `10.100.51.227:30532/unis_scc`、Redis `10.100.51.227:30379` db10)。需要覆盖时在 Deployment 的 `env` 里加 `SPRING_DATASOURCE_URL`、`SPRING_DATA_REDIS_HOST`、`WECOM_*` 等变量,`JAVA_OPTS` 可改 JVM 参数。
分体形态下两个前端容器靠 `BACKEND_HOST` / `BACKEND_PORT` 渲染 nginx 反代配置;all-in-one 反代目标是 `127.0.0.1:8080`,不需要这俩变量。
项目没有引入 `spring-boot-actuator`,探活不能用 `/actuator/health`。all-in-one 用镜像内的 `/usr/local/bin/healthcheck.sh` 同时检查三个端口,分体形态的后端用 TCP 探活。
## 可选配置
### Harbor 私有项目的拉取凭证
`public` 项目允许匿名拉取,不需要下面这个。若把镜像挪到私有项目,再取消注释并打开 Deployment 里的 `imagePullSecrets`:
```yaml
apiVersion: v1
kind: Secret
metadata:
name: harbor-regcred
namespace: unis-scc
type: kubernetes.io/dockerconfigjson
stringData:
.dockerconfigjson: '{"auths":{"docker.oa.unissense.tech":{"username":"admin","password":"unix@1234"}}}'
```
### Ingress(k3s 自带 Traefik)
两个前端是各自独立的 SPA、都按 `/` 为根构建,所以按域名拆比按路径拆省事:
```yaml
apiVersion: networking.k8s.io/v1
kind: Ingress
metadata:
name: unis-scc
namespace: unis-scc
spec:
rules:
- host: crm.unissense.top
http:
paths:
- { path: /, pathType: Prefix, backend: { service: { name: unis-scc, port: { name: crm } } } }
- host: scc-admin.unissense.top
http:
paths:
- { path: /, pathType: Prefix, backend: { service: { name: unis-scc, port: { name: sys } } } }
```
## 导入报错排查
**`TypeError: Cannot read properties of null (reading 'metadata')`**
原因:清单里存在"空文档"——`---` 之后整段都是注释,YAML 解析出一个 `null`,按数组逐个取 `metadata` 就炸。kubectl 会忽略空文档,所以命令行正常、GUI 导入失败。现已把所有示例注释移出 YAML(改放本 README),并加了校验:
```bash
./deploy/check-manifests.py # 检查同目录全部清单,报出空文档/缺字段的文档序号
```
仍失败就用无注释等价版 `k3s-unis-scc-allinone.min.yaml`(与主文件解析结果逐字段一致,绕开注释相关的解析差异)。
**离线导入后 ErrImagePull**:`imagePullPolicy` 必须是 `IfNotPresent`,否则 kubelet 仍会去 Harbor 拉取。
**改完 YAML 记得重新生成 min 版**:
```bash
python3 - <<'PY'
import yaml
src, dst = 'k3s-unis-scc-allinone.yaml', 'k3s-unis-scc-allinone.min.yaml'
docs = [d for d in yaml.safe_load_all(open(src, encoding='utf-8')) if d is not None]
open(dst, 'w', encoding='utf-8').write(
"---\n".join(yaml.dump(d, sort_keys=False, default_flow_style=False, allow_unicode=True) for d in docs))
PY
```

View File

@ -0,0 +1,59 @@
# =============================================================================
# unis_scc all-in-one 镜像(linux/amd64)
# 构建上下文 = 仓库根目录:
# docker buildx build --platform linux/amd64 \
# -f deploy/all-in-one/Dockerfile -t <registry>/public/unis_scc:<tag> .
# 后端 jar 需先在宿主机执行 mvn -DskipTests package(com.unisbase 是本地 Maven 依赖)
# =============================================================================
# ---- 构建 CRM 前端 ----
FROM node:20-alpine AS crm-web
WORKDIR /build
COPY frontend/package.json frontend/package-lock.json* ./
RUN npm ci --registry=https://registry.npmmirror.com
COPY frontend/ ./
RUN npm run build
# ---- 构建系统管理前端 ----
FROM node:20-alpine AS sys-web
WORKDIR /build
COPY frontend1/package.json frontend1/package-lock.json* ./
RUN npm ci --registry=https://registry.npmmirror.com
COPY frontend1/ ./
RUN npm run build
# ---- 运行阶段:JRE 17 + nginx ----
FROM eclipse-temurin:17-jre
ENV DEBIAN_FRONTEND=noninteractive
RUN apt-get update \
&& apt-get install -y --no-install-recommends nginx \
&& rm -rf /var/lib/apt/lists/* \
# Debian 默认站点会抢占 80 端口的 default_server
&& rm -f /etc/nginx/sites-enabled/default \
&& mkdir -p /usr/share/nginx/html/crm /usr/share/nginx/html/sys /app/logs
COPY deploy/all-in-one/nginx/unis-scc.conf /etc/nginx/conf.d/unis-scc.conf
COPY deploy/all-in-one/nginx/proxy-backend.conf /etc/nginx/snippets/proxy-backend.conf
COPY deploy/all-in-one/start.sh /usr/local/bin/start.sh
COPY deploy/all-in-one/healthcheck.sh /usr/local/bin/healthcheck.sh
COPY --from=crm-web /build/dist /usr/share/nginx/html/crm
COPY --from=sys-web /build/dist /usr/share/nginx/html/sys
COPY backend/target/unis-crm-backend-1.0.0-SNAPSHOT.jar /app/app.jar
RUN chmod +x /usr/local/bin/start.sh /usr/local/bin/healthcheck.sh \
&& nginx -t
WORKDIR /app
ENV TZ=Asia/Shanghai
ENV SPRING_PROFILES_ACTIVE=prod
# 80 CRM 前端 / 8090 系统管理前端 / 8080 后端 API
EXPOSE 80 8090 8080
HEALTHCHECK --interval=30s --timeout=5s --start-period=90s --retries=5 \
CMD ["/usr/local/bin/healthcheck.sh"]
ENTRYPOINT ["/usr/local/bin/start.sh"]

View File

@ -0,0 +1,7 @@
#!/usr/bin/env bash
# 同时确认 nginx(:80) 与 Spring Boot(:8080) 都在监听
set -uo pipefail
bash -c ': </dev/tcp/127.0.0.1/80' || { echo "nginx :80 未就绪"; exit 1; }
bash -c ': </dev/tcp/127.0.0.1/8090' || { echo "nginx :8090 未就绪"; exit 1; }
bash -c ': </dev/tcp/127.0.0.1/8080' || { echo "backend :8080 未就绪"; exit 1; }
exit 0

View File

@ -0,0 +1,10 @@
proxy_pass http://unis_backend;
proxy_http_version 1.1;
proxy_set_header Connection "";
proxy_set_header Host $host;
proxy_set_header X-Real-IP $remote_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-Proto $scheme;
proxy_connect_timeout 10s;
proxy_read_timeout 360s;
proxy_send_timeout 360s;

View File

@ -0,0 +1,94 @@
# unis_scc all-in-one:同容器内 nginx 托管两个前端站点,反代到 127.0.0.1:8080 的 Spring Boot
upstream unis_backend {
server 127.0.0.1:8080;
keepalive 32;
}
# ---- CRM 前端(原 9002 端口)----
server {
listen 80;
server_name _;
root /usr/share/nginx/html/crm;
index index.html;
# 与后端 multipart 上限(500MB/600MB)对齐,否则大文件上传会被 nginx 默认 1m 拦掉
client_max_body_size 600m;
location / {
try_files $uri $uri/ /index.html;
}
# 带 hash 的构建产物可以放心长缓存;index.html 必须每次回源,否则发版后拿到旧包
location /assets/ {
expires 30d;
add_header Cache-Control "public, immutable";
}
location = /index.html {
add_header Cache-Control "no-cache";
}
location = /healthz {
access_log off;
return 200 "ok\n";
}
# 系统管理接口:前端以 /api/sys/* 发起,后端真实路径是 /sys/*
location /api/sys/ {
rewrite ^/api/(.*)$ /$1 break;
include /etc/nginx/snippets/proxy-backend.conf;
}
# CRM 业务接口本身就带 /api 前缀
location /api/ {
include /etc/nginx/snippets/proxy-backend.conf;
}
location /sys/ {
include /etc/nginx/snippets/proxy-backend.conf;
}
}
# ---- 系统管理前端(原 9003 端口)----
server {
listen 8090;
server_name _;
root /usr/share/nginx/html/sys;
index index.html;
client_max_body_size 600m;
location / {
try_files $uri $uri/ /index.html;
}
location /assets/ {
expires 30d;
add_header Cache-Control "public, immutable";
}
location = /index.html {
add_header Cache-Control "no-cache";
}
location = /healthz {
access_log off;
return 200 "ok\n";
}
location /api/sys/ {
rewrite ^/api/(.*)$ /$1 break;
include /etc/nginx/snippets/proxy-backend.conf;
}
location /api/ {
include /etc/nginx/snippets/proxy-backend.conf;
}
# 系统前端直接用 /sys/* 调后端
location /sys/ {
include /etc/nginx/snippets/proxy-backend.conf;
}
}

View File

@ -0,0 +1,44 @@
#!/usr/bin/env bash
# 同容器内拉起 Spring Boot 后端与 nginx,任一进程退出即整体退出,交给 k8s 重启。
set -uo pipefail
: "${SPRING_PROFILES_ACTIVE:=prod}"
export SPRING_PROFILES_ACTIVE
: "${JAVA_OPTS:=-XX:MaxRAMPercentage=70.0 -XX:+UseG1GC}"
APP_PID=""
NGX_PID=""
cleanup() {
[ -n "$NGX_PID" ] && kill -TERM "$NGX_PID" 2>/dev/null
[ -n "$APP_PID" ] && kill -TERM "$APP_PID" 2>/dev/null
return 0
}
trap 'cleanup; exit 0' TERM INT
echo "[start] java $JAVA_OPTS -jar /app/app.jar (profile=$SPRING_PROFILES_ACTIVE)"
# shellcheck disable=SC2086
java $JAVA_OPTS -jar /app/app.jar &
APP_PID=$!
echo "[start] nginx (crm :80 / sys :8090)"
nginx -g 'daemon off;' &
NGX_PID=$!
code=0
while true; do
if ! kill -0 "$APP_PID" 2>/dev/null; then
wait "$APP_PID"; code=$?
echo "[start] 后端进程退出 code=$code,容器随之退出"
break
fi
if ! kill -0 "$NGX_PID" 2>/dev/null; then
wait "$NGX_PID"; code=$?
echo "[start] nginx 进程退出 code=$code,容器随之退出"
break
fi
sleep 2
done
cleanup
exit "$code"

View File

@ -0,0 +1,71 @@
#!/usr/bin/env bash
# 构建 unis_scc 镜像(linux/amd64)并推送到 Harbor。
#
# ./deploy/build-and-push.sh # 默认:all-in-one 单镜像 unis_scc
# ./deploy/build-and-push.sh app # 同上,显式指定
# ./deploy/build-and-push.sh backend frontend system # 分别打三个镜像
# TAG=202610291500 ./deploy/build-and-push.sh
# PUSH=0 ./deploy/build-and-push.sh # 只构建不推送
set -euo pipefail
REGISTRY="${REGISTRY:-docker.oa.unissense.tech}"
PROJECT="${PROJECT:-public}"
BASE_NAME="${BASE_NAME:-unis_scc}"
TAG="${TAG:-$(date +%Y%m%d%H%M)}"
PLATFORM="${PLATFORM:-linux/amd64}"
PUSH="${PUSH:-1}"
ROOT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
BACKEND_JAR="$ROOT_DIR/backend/target/unis-crm-backend-1.0.0-SNAPSHOT.jar"
# 组件 -> "构建上下文|Dockerfile|镜像后缀"
spec() {
case "$1" in
app) echo "$ROOT_DIR|$ROOT_DIR/deploy/all-in-one/Dockerfile|" ;;
backend) echo "$ROOT_DIR/backend|$ROOT_DIR/backend/Dockerfile|-backend" ;;
frontend) echo "$ROOT_DIR/frontend|$ROOT_DIR/frontend/Dockerfile|-frontend" ;;
system) echo "$ROOT_DIR/frontend1|$ROOT_DIR/frontend1/Dockerfile|-system" ;;
*) return 1 ;;
esac
}
targets=("$@")
if [ ${#targets[@]} -eq 0 ]; then
targets=(app)
fi
need_jar=0
for t in "${targets[@]}"; do
case "$t" in app|backend) need_jar=1 ;; esac
done
if [ "$need_jar" = 1 ]; then
echo "==> 重新打包后端 jar(JDK 17)"
JAVA_HOME="${JAVA_HOME_17:-$(/usr/libexec/java_home -v 17)}" \
mvn -B -q -DskipTests -f "$ROOT_DIR/backend/pom.xml" clean package
[ -f "$BACKEND_JAR" ] || { echo "缺少 $BACKEND_JAR"; exit 1; }
fi
for comp in "${targets[@]}"; do
IFS='|' read -r ctx dockerfile suffix < <(spec "$comp") || { echo "未知组件: $comp"; exit 1; }
image="$REGISTRY/$PROJECT/$BASE_NAME$suffix"
echo "==> 构建 $image:$TAG ($PLATFORM)"
docker buildx build --platform "$PLATFORM" \
-f "$dockerfile" \
-t "$image:$TAG" -t "$image:latest" \
--load "$ctx"
done
if [ "$PUSH" != 1 ]; then
echo "==> PUSH=0,跳过推送"; exit 0
fi
for comp in "${targets[@]}"; do
IFS='|' read -r _ _ suffix < <(spec "$comp")
image="$REGISTRY/$PROJECT/$BASE_NAME$suffix"
echo "==> 推送 $image"
docker push "$image:$TAG"
docker push "$image:latest"
done
echo "==> 完成,TAG=$TAG"

View File

@ -0,0 +1,42 @@
#!/usr/bin/env python3
"""校验 k8s 清单,专防"注释成段被注释掉后留下空文档"这类导入器炸点。
用法: ./deploy/check-manifests.py [文件...] # 不传参则检查同目录下所有 yaml
"""
import sys
from pathlib import Path
import yaml
REQUIRED = {"apiVersion", "kind", "metadata"}
def main(argv: list[str]) -> int:
here = Path(__file__).resolve().parent
files = [Path(a) for a in argv] if argv else sorted(here.glob("*.yaml"))
failed = False
for f in files:
docs = list(yaml.safe_load_all(f.read_text(encoding="utf-8")))
problems = []
for i, d in enumerate(docs):
if d is None:
problems.append(f"doc#{i} 是空文档(--- 之后只有注释),导入器读 metadata 会得到 null")
elif not isinstance(d, dict):
problems.append(f"doc#{i} 不是映射对象: {type(d).__name__}")
elif missing := REQUIRED - set(d):
problems.append(f"doc#{i} 缺少字段 {sorted(missing)}")
elif not d["metadata"].get("name"):
problems.append(f"doc#{i} metadata.name 为空")
if problems:
failed = True
print(f"FAIL {f.name}")
for p in problems:
print(f" - {p}")
else:
kinds = ", ".join(f"{d['kind']}/{d['metadata']['name']}" for d in docs)
print(f"OK {f.name} ({len(docs)} 个资源) {kinds}")
return 1 if failed else 0
if __name__ == "__main__":
raise SystemExit(main(sys.argv[1:]))

View File

@ -0,0 +1,121 @@
apiVersion: v1
kind: Namespace
metadata:
name: unis-scc
labels:
app.kubernetes.io/part-of: unis-scc
---
apiVersion: apps/v1
kind: Deployment
metadata:
name: unis-scc
namespace: unis-scc
labels:
app.kubernetes.io/name: unis-scc
spec:
replicas: 1
revisionHistoryLimit: 5
strategy:
type: RollingUpdate
rollingUpdate:
maxSurge: 1
maxUnavailable: 0
selector:
matchLabels:
app.kubernetes.io/name: unis-scc
template:
metadata:
labels:
app.kubernetes.io/name: unis-scc
spec:
terminationGracePeriodSeconds: 45
containers:
- name: unis-scc
image: docker.oa.unissense.tech/public/unis_scc:latest
imagePullPolicy: IfNotPresent
ports:
- name: crm
containerPort: 80
- name: sys
containerPort: 8090
- name: api
containerPort: 8080
env:
- name: TZ
value: Asia/Shanghai
- name: SPRING_PROFILES_ACTIVE
value: prod
resources:
requests:
cpu: 300m
memory: 1Gi
limits:
cpu: '2'
memory: 2Gi
startupProbe:
exec:
command:
- /usr/local/bin/healthcheck.sh
periodSeconds: 5
failureThreshold: 60
readinessProbe:
exec:
command:
- /usr/local/bin/healthcheck.sh
periodSeconds: 10
timeoutSeconds: 5
livenessProbe:
exec:
command:
- /usr/local/bin/healthcheck.sh
initialDelaySeconds: 30
periodSeconds: 20
timeoutSeconds: 5
failureThreshold: 5
volumeMounts:
- name: app-logs
mountPath: /app/logs
volumes:
- name: app-logs
emptyDir: {}
---
apiVersion: v1
kind: Service
metadata:
name: unis-scc
namespace: unis-scc
labels:
app.kubernetes.io/name: unis-scc
spec:
type: NodePort
selector:
app.kubernetes.io/name: unis-scc
ports:
- name: crm
port: 80
targetPort: crm
nodePort: 30082
- name: sys
port: 8090
targetPort: sys
nodePort: 30083
- name: api
port: 8080
targetPort: api
nodePort: 30081
---
apiVersion: v1
kind: Service
metadata:
name: unis-scc-api
namespace: unis-scc
labels:
app.kubernetes.io/name: unis-scc
spec:
type: ClusterIP
selector:
app.kubernetes.io/name: unis-scc
ports:
- name: api
port: 8080
targetPort: api

View File

@ -0,0 +1,171 @@
# ============================================================================
# unis_scc(UNIS CRM)—— k3s 单镜像 all-in-one 部署
#
# 镜像:docker.oa.unissense.tech/public/unis_scc:latest (linux/amd64)
# 一个容器里同时跑:
# nginx :80 -> CRM 前端静态站点 + /api、/sys 反代
# nginx :8090 -> 系统管理前端静态站点 + /api、/sys 反代
# java :8080 -> Spring Boot 后端(profile=prod)
# 任一子进程退出,容器整体退出,由 k8s 重启,不会出现"半死不活"的 Pod。
#
# 在线部署(public 项目允许匿名拉取):
# kubectl apply -f k3s-unis-scc-allinone.yaml
# kubectl -n unis-scc rollout status deploy/unis-scc
#
# 离线导入(节点访问不了 Harbor):
# 离线包已生成:deploy/unis_scc-202610291459.tar(163MB)
# scp deploy/unis_scc-202610291459.tar node:/tmp/ && ssh node 'k3s ctr images import /tmp/unis_scc-202610291459.tar'
# kubectl apply -f k3s-unis-scc-allinone.yaml
#
# 访问入口(NodePort):
# CRM 前端 http://<节点IP>:30082
# 系统管理前端 http://<节点IP>:30083
# 后端 API http://<节点IP>:30081 (调试用;正常流量走容器内 nginx)
#
# 数据库/Redis 用镜像内置的 application-prod.yml:
# PG 10.100.51.227:30532/unis_scc Redis 10.100.51.227:30379 db10
# 可选的 Harbor 拉取凭证、Ingress 示例见 deploy/README.md
# ============================================================================
apiVersion: v1
kind: Namespace
metadata:
name: unis-scc
labels:
app.kubernetes.io/part-of: unis-scc
---
apiVersion: apps/v1
kind: Deployment
metadata:
name: unis-scc
namespace: unis-scc
labels:
app.kubernetes.io/name: unis-scc
spec:
replicas: 1
revisionHistoryLimit: 5
strategy:
type: RollingUpdate
rollingUpdate:
maxSurge: 1
maxUnavailable: 0
selector:
matchLabels:
app.kubernetes.io/name: unis-scc
template:
metadata:
labels:
app.kubernetes.io/name: unis-scc
spec:
# imagePullSecrets:
# - name: harbor-regcred
# 后端是 Spring Boot,优雅停机需要点时间
terminationGracePeriodSeconds: 45
containers:
- name: unis-scc
image: docker.oa.unissense.tech/public/unis_scc:latest
# 离线导入场景必须 IfNotPresent,否则仍会去 Harbor 拉取并 ErrImagePull;
# 想每次强制拉最新,改成 Always 或把 image 换成时间戳 tag
imagePullPolicy: IfNotPresent
ports:
- name: crm
containerPort: 80
- name: sys
containerPort: 8090
- name: api
containerPort: 8080
env:
- name: TZ
value: Asia/Shanghai
- name: SPRING_PROFILES_ACTIVE
value: prod
# 容器入口脚本会原样拼到 java 命令上,默认 -XX:MaxRAMPercentage=70.0
# - name: JAVA_OPTS
# value: "-Xms512m -Xmx1536m"
# 需要覆盖镜像内配置时打开:
# - name: SPRING_DATASOURCE_URL
# value: "jdbc:postgresql://10.100.51.227:30532/unis_scc?connectTimeout=10&socketTimeout=120&tcpKeepAlive=true"
# - name: WECOM_ENABLED
# value: "true"
# 企业微信 SSO 回调域名等
# - name: WECOM_REDIRECT_URI
# value: "https://crm.unissense.top/api/wecom/sso/callback"
resources:
requests:
cpu: 300m
memory: 1Gi
limits:
cpu: "2"
memory: 2Gi
# 镜像里的 healthcheck.sh 会同时确认 :80 / :8090 / :8080 都在监听
startupProbe:
exec:
command: ["/usr/local/bin/healthcheck.sh"]
periodSeconds: 5
failureThreshold: 60
readinessProbe:
exec:
command: ["/usr/local/bin/healthcheck.sh"]
periodSeconds: 10
timeoutSeconds: 5
livenessProbe:
exec:
command: ["/usr/local/bin/healthcheck.sh"]
initialDelaySeconds: 30
periodSeconds: 20
timeoutSeconds: 5
failureThreshold: 5
volumeMounts:
- name: app-logs
mountPath: /app/logs
volumes:
# 日志主要走 stdout(kubectl logs deploy/unis-scc),这里只是给应用留可写目录
- name: app-logs
emptyDir: {}
---
# 统一入口:一个 Service 暴露三个端口
apiVersion: v1
kind: Service
metadata:
name: unis-scc
namespace: unis-scc
labels:
app.kubernetes.io/name: unis-scc
spec:
type: NodePort
selector:
app.kubernetes.io/name: unis-scc
ports:
- name: crm
port: 80
targetPort: crm
nodePort: 30082
- name: sys
port: 8090
targetPort: sys
nodePort: 30083
- name: api
port: 8080
targetPort: api
nodePort: 30081
---
# 集群内网调用后端用的 ClusterIP,省得依赖 NodePort
apiVersion: v1
kind: Service
metadata:
name: unis-scc-api
namespace: unis-scc
labels:
app.kubernetes.io/name: unis-scc
spec:
type: ClusterIP
selector:
app.kubernetes.io/name: unis-scc
ports:
- name: api
port: 8080
targetPort: api

View File

@ -0,0 +1,342 @@
# ============================================================================
# unis_scc (UNIS CRM) —— k3s 部署清单
#
# 镜像(linux/amd64,Harbor: docker.oa.unissense.tech,project = public 公共项目)
# docker.oa.unissense.tech/public/unis_scc-backend:202610291447 Spring Boot 后端 :8080
# docker.oa.unissense.tech/public/unis_scc-frontend:202610291447 CRM 前端 nginx :80
# docker.oa.unissense.tech/public/unis_scc-system:202610291447 系统管理前端 nginx :80
#
# 在线部署(k3s 节点能直连 Harbor,公共项目可匿名拉取):
# kubectl apply -f k3s-unis-scc.yaml
# kubectl -n unis-scc rollout status deployment/unis-scc-backend
# kubectl -n unis-scc rollout status deployment/unis-scc-frontend
# kubectl -n unis-scc rollout status deployment/unis-scc-system
#
# 离线导入(节点访问不了 Harbor 时):
# k3s ctr images import unis_scc-images-202610291447.tar # 离线包已在 deploy/ 下生成
# kubectl apply -f k3s-unis-scc.yaml
#
# 访问入口(NodePort):
# CRM 前端 http://<节点IP>:30082
# 系统管理前端 http://<节点IP>:30083 (/sys/api/* 走同一后端)
# 后端 API http://<节点IP>:30081 (仅调试用,正常流量走前端 nginx 反代)
#
# 数据库/Redis 不再走 ConfigMap,已烘进镜像里的 application-prod.yml:
# PG 10.100.51.227:30532/unis_scc
# Redis 10.100.51.227:30379 db10
# 如需临时覆盖,见下方 backend Deployment 里注释掉的 env 段。
# 可选的 Harbor 拉取凭证、Ingress 示例见 deploy/README.md
# ============================================================================
apiVersion: v1
kind: Namespace
metadata:
name: unis-scc
labels:
app.kubernetes.io/part-of: unis-scc
# --- 私有项目才需要;public 项目可匿名拉取,留作备用 ---
# apiVersion: v1
# kind: Secret
# metadata:
# name: harbor-regcred
# namespace: unis-scc
# type: kubernetes.io/dockerconfigjson
# stringData:
# .dockerconfigjson: '{"auths":{"docker.oa.unissense.tech":{"username":"admin","password":"unix@1234"}}}'
---
# ===========================================================================
# 后端:Spring Boot(profile=prod),监听 8080
# ===========================================================================
apiVersion: apps/v1
kind: Deployment
metadata:
name: unis-scc-backend
namespace: unis-scc
labels:
app.kubernetes.io/name: unis-scc
app.kubernetes.io/component: backend
spec:
replicas: 1
revisionHistoryLimit: 3
strategy:
type: RollingUpdate
rollingUpdate:
maxSurge: 1
maxUnavailable: 0
selector:
matchLabels:
app.kubernetes.io/name: unis-scc
app.kubernetes.io/component: backend
template:
metadata:
labels:
app.kubernetes.io/name: unis-scc
app.kubernetes.io/component: backend
spec:
# imagePullSecrets:
# - name: harbor-regcred
terminationGracePeriodSeconds: 45
containers:
- name: backend
image: docker.oa.unissense.tech/public/unis_scc-backend:202610291447
imagePullPolicy: IfNotPresent
# 镜像 ENTRYPOINT 是 `java -jar app.jar`,这里只补 JVM 参数,不改启动方式
command: ["java"]
args:
- "-XX:MaxRAMPercentage=70.0"
- "-XX:+UseG1GC"
- "-Duser.timezone=Asia/Shanghai"
- "-jar"
- "/app/app.jar"
ports:
- name: http
containerPort: 8080
env:
- name: TZ
value: Asia/Shanghai
- name: SPRING_PROFILES_ACTIVE
value: prod
# - name: SPRING_DATASOURCE_URL
# value: "jdbc:postgresql://10.100.51.227:30532/unis_scc?connectTimeout=10&socketTimeout=120&tcpKeepAlive=true"
# - name: SPRING_DATA_REDIS_HOST
# value: "10.100.51.227"
# - name: WECOM_ENABLED
# value: "true"
resources:
requests:
cpu: 250m
memory: 768Mi
limits:
cpu: "1500m"
memory: 1800Mi
# 项目未引入 spring-boot-actuator,用 TCP 探活
startupProbe:
tcpSocket:
port: http
periodSeconds: 5
failureThreshold: 60
readinessProbe:
tcpSocket:
port: http
periodSeconds: 10
timeoutSeconds: 3
livenessProbe:
tcpSocket:
port: http
initialDelaySeconds: 15
periodSeconds: 20
timeoutSeconds: 3
failureThreshold: 5
volumeMounts:
- name: app-logs
mountPath: /app/logs
volumes:
# 日志主要走 stdout(kubectl logs 即可查看),这里只是给应用留出可写目录
- name: app-logs
emptyDir: {}
---
apiVersion: v1
kind: Service
metadata:
name: unis-scc-backend
namespace: unis-scc
labels:
app.kubernetes.io/name: unis-scc
app.kubernetes.io/component: backend
spec:
type: ClusterIP
selector:
app.kubernetes.io/name: unis-scc
app.kubernetes.io/component: backend
ports:
- name: http
port: 8080
targetPort: http
---
# 后端 NodePort,便于从集群外直接调 API / 健康检查
apiVersion: v1
kind: Service
metadata:
name: unis-scc-backend-external
namespace: unis-scc
labels:
app.kubernetes.io/name: unis-scc
app.kubernetes.io/component: backend
spec:
type: NodePort
selector:
app.kubernetes.io/name: unis-scc
app.kubernetes.io/component: backend
ports:
- name: http
port: 8080
targetPort: http
nodePort: 30081
---
# ===========================================================================
# CRM 前端:nginx 静态站点 + /api、/sys 反向代理到后端
# ===========================================================================
apiVersion: apps/v1
kind: Deployment
metadata:
name: unis-scc-frontend
namespace: unis-scc
labels:
app.kubernetes.io/name: unis-scc
app.kubernetes.io/component: frontend
spec:
replicas: 1
revisionHistoryLimit: 3
selector:
matchLabels:
app.kubernetes.io/name: unis-scc
app.kubernetes.io/component: frontend
template:
metadata:
labels:
app.kubernetes.io/name: unis-scc
app.kubernetes.io/component: frontend
spec:
# imagePullSecrets:
# - name: harbor-regcred
containers:
- name: frontend
image: docker.oa.unissense.tech/public/unis_scc-frontend:202610291447
imagePullPolicy: IfNotPresent
ports:
- name: http
containerPort: 80
env:
- name: TZ
value: Asia/Shanghai
# 容器启动脚本会把这两个值渲染进 nginx 配置
- name: BACKEND_HOST
value: unis-scc-backend.unis-scc.svc.cluster.local
- name: BACKEND_PORT
value: "8080"
resources:
requests:
cpu: 20m
memory: 32Mi
limits:
cpu: 300m
memory: 128Mi
readinessProbe:
httpGet:
path: /
port: http
periodSeconds: 10
livenessProbe:
httpGet:
path: /
port: http
initialDelaySeconds: 10
periodSeconds: 20
failureThreshold: 5
---
apiVersion: v1
kind: Service
metadata:
name: unis-scc-frontend
namespace: unis-scc
labels:
app.kubernetes.io/name: unis-scc
app.kubernetes.io/component: frontend
spec:
type: NodePort
selector:
app.kubernetes.io/name: unis-scc
app.kubernetes.io/component: frontend
ports:
- name: http
port: 80
targetPort: http
nodePort: 30082
---
# ===========================================================================
# 系统管理前端:nginx 静态站点,/sys/api/* 转发到后端 /sys/*
# ===========================================================================
apiVersion: apps/v1
kind: Deployment
metadata:
name: unis-scc-system
namespace: unis-scc
labels:
app.kubernetes.io/name: unis-scc
app.kubernetes.io/component: system
spec:
replicas: 1
revisionHistoryLimit: 3
selector:
matchLabels:
app.kubernetes.io/name: unis-scc
app.kubernetes.io/component: system
template:
metadata:
labels:
app.kubernetes.io/name: unis-scc
app.kubernetes.io/component: system
spec:
# imagePullSecrets:
# - name: harbor-regcred
containers:
- name: system
image: docker.oa.unissense.tech/public/unis_scc-system:202610291447
imagePullPolicy: IfNotPresent
ports:
- name: http
containerPort: 80
env:
- name: TZ
value: Asia/Shanghai
- name: BACKEND_HOST
value: unis-scc-backend.unis-scc.svc.cluster.local
- name: BACKEND_PORT
value: "8080"
resources:
requests:
cpu: 20m
memory: 32Mi
limits:
cpu: 300m
memory: 128Mi
readinessProbe:
httpGet:
path: /
port: http
periodSeconds: 10
livenessProbe:
httpGet:
path: /
port: http
initialDelaySeconds: 10
periodSeconds: 20
failureThreshold: 5
---
apiVersion: v1
kind: Service
metadata:
name: unis-scc-system
namespace: unis-scc
labels:
app.kubernetes.io/name: unis-scc
app.kubernetes.io/component: system
spec:
type: NodePort
selector:
app.kubernetes.io/name: unis-scc
app.kubernetes.io/component: system
ports:
- name: http
port: 80
targetPort: http
nodePort: 30083

View File

@ -0,0 +1,146 @@
-- ============================================================================
-- 生产库迁移后同步序列:unis_crm(PG17@192.168.124.202) -> unis_scc(PG16@10.100.51.227)
--
-- 成因(已实测复现):Navicat 的“结构导出”先 CREATE SEQUENCE "x_id_seq"(START 1,不带
-- 源库当前值),再 CREATE TABLE ... GENERATED BY DEFAULT AS IDENTITY。规范名被占用,
-- identity 只能另建一个 *_id_seq1 并从 1 开始。新库现状:34 个 identity 列绑在
-- *_id_seq1 上,94 个序列里有 37 个已无任何列引用(其中一部分还残留着旧库数值)。
-- 所以不要按 "<表>_<列>_seq" 猜序列名,必须用 pg_get_serial_sequence() 解析;
-- 否则连 pg_dump 自带的 setval 都会打到那些孤儿序列上,不报错但完全不生效。
--
-- 首选做法:用原生 pg_dump 把结构+数据一次性恢复进空库,序列名与数值都会自动对齐,
-- 本脚本只是兜底自检(正常输出“上调 0 个,剩余落后 0 个”)。
--
-- 用法:灌完数据之后、放量之前执行。只在序列落后时上调,从不下调,可重复执行。
-- psql "host=10.100.51.227 port=30532 dbname=unis_scc user=root" -f 本文件
--
-- 验证:已在 PostgreSQL 16.15 上按上述成因搭建夹具,复现 identity / serial / 业务编码
-- 三种 duplicate key 失败,执行本脚本后全部恢复,重复执行幂等。
-- ============================================================================
begin;
-- 序列被别的会话占用时跳过并告警,不要卡住发布
set local lock_timeout = '5s';
set local statement_timeout = '10min';
-- 待检查清单:identity / serial 列由目录解析,业务编码序列按名字显式列出
create temp table if not exists seq_targets (seq text, tbl text, col text, prefix text) on commit preserve rows;
truncate seq_targets;
insert into seq_targets (seq, tbl, col, prefix)
select pg_get_serial_sequence(format('%I.%I', n.nspname, c.relname), a.attname),
format('%I.%I', n.nspname, c.relname),
a.attname,
null
from pg_attribute a
join pg_class c on c.oid = a.attrelid
join pg_namespace n on n.oid = c.relnamespace
where n.nspname = 'public'
and c.relkind = 'r'
and a.attnum > 0
and (a.attidentity <> '' or exists (
select 1 from pg_attrdef d
where d.adrelid = a.attrelid
and d.adnum = a.attnum
and pg_get_expr(d.adbin, d.adrelid) like 'nextval%'));
-- 编码形如 QD-20261029-000822,mapper 里用 nextval() 直接取号
insert into seq_targets (seq, tbl, col, prefix) values
('public.crm_channel_expansion_code_seq', 'crm_channel_expansion', 'channel_code', 'QD'),
('public.crm_customer_code_seq', 'crm_customer', 'customer_code', 'CUS'),
('public.crm_opportunity_code_seq', 'crm_opportunity', 'opportunity_code', 'OPP');
do $$
declare
r record;
v_next bigint;
v_max bigint;
v_fixed int := 0;
v_late int := 0;
begin
----------------------------------------------------------------------------
-- 第一轮:落后的上调到 max + 1
----------------------------------------------------------------------------
for r in select seq, tbl, col, prefix from seq_targets where seq is not null order by tbl, col
loop
begin
if r.prefix is null then
execute format('select coalesce(max(%I), 0) from %s', r.col, r.tbl) into v_max;
else
execute format(
$f$select coalesce(max((substring(%I from '([0-9]+)$'))::bigint), 0)
from %I where %I ~ %L$f$,
r.col, r.tbl, r.col, '^' || r.prefix || '-[0-9]{8}-[0-9]+$') into v_max;
end if;
continue when v_max = 0; -- 空表不动序列
execute format(
'select case when is_called then last_value + 1 else last_value end from %s',
r.seq) into v_next;
if v_next <= v_max then
-- is_called = true:下一次 nextval() 返回 v_max + 1
execute format('select setval(%L, %s, true)', r.seq, v_max);
v_fixed := v_fixed + 1;
raise notice '[sync] % : next % -> %', r.seq, v_next, v_max + 1;
end if;
exception when others then
raise warning '[sync] skip % : %', r.seq, sqlerrm;
end;
end loop;
----------------------------------------------------------------------------
-- 第二轮:自检,正常情况下不应再有落后项
----------------------------------------------------------------------------
for r in select seq, tbl, col, prefix from seq_targets where seq is not null order by tbl, col
loop
begin
if r.prefix is null then
execute format('select coalesce(max(%I), 0) from %s', r.col, r.tbl) into v_max;
else
execute format(
$f$select coalesce(max((substring(%I from '([0-9]+)$'))::bigint), 0)
from %I where %I ~ %L$f$,
r.col, r.tbl, r.col, '^' || r.prefix || '-[0-9]{8}-[0-9]+$') into v_max;
end if;
continue when v_max = 0;
execute format(
'select case when is_called then last_value + 1 else last_value end from %s',
r.seq) into v_next;
if v_next <= v_max then
v_late := v_late + 1;
raise warning '[check] % 仍然落后:next = %,期望 >= %', r.seq, v_next, v_max + 1;
end if;
exception when others then
raise warning '[check] skip % : %', r.seq, sqlerrm;
end;
end loop;
raise notice '序列同步完成:上调 % 个,剩余落后 % 个', v_fixed, v_late;
end
$$;
commit;
-- 参考:库里还有这些序列没有任何列或默认值引用(迁移残留,确认后可清理)
select s.relname as unused_sequence,
(select last_value from pg_sequences ps
where ps.schemaname = 'public' and ps.sequencename = s.relname) as last_value
from pg_class s
join pg_namespace n on n.oid = s.relnamespace
where s.relkind = 'S'
and n.nspname = 'public'
and not exists (
select 1 from pg_attrdef d
where pg_get_expr(d.adbin, d.adrelid) like '%' || s.relname || '%')
and not exists (
select 1 from seq_targets t
where t.seq = 'public.' || s.relname or t.seq = s.relname)
and s.relname not in (
select (string_to_array(seq, '.'))[2] from seq_targets where seq is not null)
order by 1;