LangChain-Chatchat 开发与应用(十) 企业级部署方案-Docker-Compose-K8s集群化实践

企业级部署方案Docker Compose / K8s 集群化实践标签Docker | Kubernetes | 高可用 | 生产部署 | DevOps一、从单机玩具到生产系统前面九篇咱们从 0 到 1 搭起了 Chatchat做了二次开发优化了性能。但到目前为止所有操作都是单机的。要真正上线服务企业用户还需要解决服务挂了怎么办怎么扩容配置怎么管理数据怎么备份怎么灰度发布今天这篇咱们把 Chatchat 从单机部署升级到企业级集群部署。二、部署方案选型2.1 三种部署方式对比方案复杂度适用场景扩展性Docker Compose低中小团队、单/多机手动扩容Kubernetes高大规模、云原生自动扩容云服务商方案中快速上线、不想运维按需扩展建议团队 10 人用户 1000 → Docker Compose团队 10 人需要自动运维 → Kubernetes想最快上线 → 云服务商方案阿里云、腾讯云等2.2 本文覆盖这篇重点讲Docker Compose和Kubernetes两种方案。三、Docker Compose 部署3.1 整体架构┌─────────────────────────────────────────────────┐ │ Nginx (反向代理) │ │ 端口 80/443SSL 终止 │ └─────────────────────────────────────────────────┘ │ ┌───────────────┼───────────────┐ ↓ ↓ ↓ ┌──────────────┐ ┌──────────────┐ ┌──────────────┐ │ Chatchat │ │ Chatchat │ │ Xinference │ │ API x2 │ │ WebUI │ │ (LLM Emb) │ │ (负载均衡) │ │ │ │ │ └──────────────┘ └──────────────┘ └──────────────┘ │ │ │ └───────────────┼───────────────┘ ↓ ┌──────────────────┐ │ Milvus / FAISS │ │ 向量数据库 │ └──────────────────┘ │ ┌──────────────────┐ │ PostgreSQL │ │ (元数据存储) │ └──────────────────┘3.2 docker-compose.ymlversion:3.8services:# 反向代理 nginx:image:nginx:alpineports:-80:80-443:443volumes:-./nginx.conf:/etc/nginx/nginx.conf:ro-./ssl:/etc/nginx/ssl:rodepends_on:-chatchat-api-1-chatchat-api-2-chatchat-webuirestart:unless-stopped# Chatchat API 实例 1 chatchat-api-1:image:chatchat/chatchat:latestenvironment:-CHATCHAT_ROOT/data/chatchat-MODEL_PLATFORMS__0__API_BASE_URLhttp://xinference:9997/v1volumes:-chatchat-data:/data/chatchat-./configs:/app/configs:rodepends_on:-postgres-milvus-xinferencerestart:unless-stoppeddeploy:resources:limits:memory:4G# Chatchat API 实例 2 chatchat-api-2:image:chatchat/chatchat:latestenvironment:-CHATCHAT_ROOT/data/chatchat-MODEL_PLATFORMS__0__API_BASE_URLhttp://xinference:9997/v1volumes:-chatchat-data:/data/chatchat-./configs:/app/configs:rodepends_on:-postgres-milvus-xinferencerestart:unless-stoppeddeploy:resources:limits:memory:4G# Chatchat WebUI chatchat-webui:image:chatchat/chatchat-webui:latestenvironment:-API_BASE_URLhttp://nginx:80/apidepends_on:-chatchat-api-1-chatchat-api-2restart:unless-stopped# Xinference (模型服务) xinference:image:xprobe/xinference:latestports:-9997:9997environment:-XINFERENCE_HOME/data/xinferencevolumes:-xinference-data:/data/xinference-./models:/models:ro# 预下载的模型deploy:resources:reservations:devices:-driver:nvidiacount:1capabilities:[gpu]restart:unless-stopped# Milvus 向量数据库 milvus-standalone:image:milvusdb/milvus:latestports:-19530:19530environment:-ETCD_ENDPOINTSetcd:2379-MINIO_ADDRESSminio:9000volumes:-milvus-data:/var/lib/milvusdepends_on:-etcd-miniorestart:unless-stoppedetcd:image:quay.io/coreos/etcd:v3.5.5environment:-ETCD_AUTO_COMPACTION_MODErevision-ETCD_AUTO_COMPACTION_RETENTION1000volumes:-etcd-data:/etcdrestart:unless-stoppedminio:image:minio/minio:latestenvironment:-MINIO_ACCESS_KEYminioadmin-MINIO_SECRET_KEYminioadminvolumes:-minio-data:/minio_datacommand:minio server /minio_datarestart:unless-stopped# PostgreSQL postgres:image:postgres:15-alpineenvironment:-POSTGRES_USERchatchat-POSTGRES_PASSWORDyour-strong-password-POSTGRES_DBchatchatvolumes:-postgres-data:/var/lib/postgresql/data-./init.sql:/docker-entrypoint-initdb.d/init.sql:rorestart:unless-stopped# 数据卷 volumes:chatchat-data:xinference-data:milvus-data:etcd-data:minio-data:postgres-data:3.3 Nginx 配置# nginx.conf upstream chatchat_api { least_conn; # 最少连接负载均衡 server chatchat-api-1:7861; server chatchat-api-2:7861; } server { listen 80; server_name chatchat.yourcompany.com; # 强制 HTTPS return 301 https://$server_name$request_uri; } server { listen 443 ssl http2; server_name chatchat.yourcompany.com; ssl_certificate /etc/nginx/ssl/cert.pem; ssl_certificate_key /etc/nginx/ssl/key.pem; # API 代理 location /api/ { proxy_pass http://chatchat_api/; proxy_set_header Host $host; proxy_set_header X-Real-IP $remote_addr; proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for; # SSE 流式支持 proxy_buffering off; proxy_cache off; # 超时设置 proxy_read_timeout 300s; proxy_send_timeout 300s; } # WebUI location / { proxy_pass http://chatchat-webui:8501; proxy_set_header Host $host; proxy_set_header X-Real-IP $remote_addr; } # Xinference 管理界面可选限制 IP location /xinference/ { allow 10.0.0.0/8; # 仅内网访问 deny all; proxy_pass http://xinference:9997/; } }3.4 启动和运维# 启动所有服务docker-composeup-d# 查看状态docker-composeps# 查看日志docker-composelogs-fchatchat-api-1# 扩容 API 实例docker-composeup-d--scalechatchat-api-13# 重启单个服务docker-composerestart xinference# 备份数据docker-composeexecpostgres pg_dump-Uchatchat chatchatbackup.sql四、Kubernetes 部署4.1 架构设计┌─────────────────────────────────────────┐ │ Ingress Controller │ │ (Nginx / Traefik / ALB) │ └─────────────────────────────────────────┘ │ ┌───────────────┼───────────────┐ ↓ ↓ ↓ ┌────────┐ ┌────────┐ ┌────────────┐ │Service │ │Service │ │ Service │ │API │ │WebUI │ │Xinference │ └────┬───┘ └────┬───┘ └─────┬──────┘ │ │ │ ┌────┴───┐ ┌────┴───┐ ┌────┴──────┐ │Pod x3 │ │Pod x2 │ │Pod x1(GPU)│ │(API) │ │(WebUI) │ │(模型服务) │ └────────┘ └────────┘ └───────────┘4.2 Namespace 和 ConfigMap# namespace.yamlapiVersion:v1kind:Namespacemetadata:name:chatchat---# configmap.yamlapiVersion:v1kind:ConfigMapmetadata:name:chatchat-confignamespace:chatchatdata:basic_settings.yaml:|# 基础配置...model_settings.yaml:|# 模型配置...kb_settings.yaml:|# 知识库配置...4.3 Chatchat API Deployment# chatchat-api.yamlapiVersion:apps/v1kind:Deploymentmetadata:name:chatchat-apinamespace:chatchatspec:replicas:3# 3 个副本selector:matchLabels:app:chatchat-apitemplate:metadata:labels:app:chatchat-apispec:containers:-name:apiimage:chatchat/chatchat:latestports:-containerPort:7861env:-name:CHATCHAT_ROOTvalue:/data/chatchat-name:MODEL_PLATFORMS__0__API_BASE_URLvalue:http://xinference:9997/v1volumeMounts:-name:configmountPath:/app/configsreadOnly:true-name:datamountPath:/data/chatchatresources:requests:memory:2Gicpu:500mlimits:memory:4Gicpu:2000mlivenessProbe:httpGet:path:/healthport:7861initialDelaySeconds:30periodSeconds:10readinessProbe:httpGet:path:/readyport:7861initialDelaySeconds:5periodSeconds:5volumes:-name:configconfigMap:name:chatchat-config-name:datapersistentVolumeClaim:claimName:chatchat-data-pvc---# ServiceapiVersion:v1kind:Servicemetadata:name:chatchat-apinamespace:chatchatspec:selector:app:chatchat-apiports:-port:7861targetPort:7861type:ClusterIP---# HPA 自动扩缩容apiVersion:autoscaling/v2kind:HorizontalPodAutoscalermetadata:name:chatchat-api-hpanamespace:chatchatspec:scaleTargetRef:apiVersion:apps/v1kind:Deploymentname:chatchat-apiminReplicas:3maxReplicas:10metrics:-type:Resourceresource:name:cputarget:type:UtilizationaverageUtilization:70-type:Resourceresource:name:memorytarget:type:UtilizationaverageUtilization:804.4 Xinference GPU Deployment# xinference.yamlapiVersion:apps/v1kind:Deploymentmetadata:name:xinferencenamespace:chatchatspec:replicas:1selector:matchLabels:app:xinferencetemplate:metadata:labels:app:xinferencespec:nodeSelector:accelerator:nvidia# 调度到 GPU 节点containers:-name:xinferenceimage:xprobe/xinference:latestports:-containerPort:9997env:-name:XINFERENCE_HOMEvalue:/data/xinferencevolumeMounts:-name:model-cachemountPath:/data/xinferenceresources:requests:memory:8Gicpu:2000mnvidia.com/gpu:1# 请求 1 张 GPUlimits:memory:16Gicpu:4000mnvidia.com/gpu:1---apiVersion:v1kind:Servicemetadata:name:xinferencenamespace:chatchatspec:selector:app:xinferenceports:-port:9997targetPort:99974.5 Ingress 配置# ingress.yamlapiVersion:networking.k8s.io/v1kind:Ingressmetadata:name:chatchat-ingressnamespace:chatchatannotations:nginx.ingress.kubernetes.io/ssl-redirect:truenginx.ingress.kubernetes.io/proxy-body-size:100mnginx.ingress.kubernetes.io/proxy-read-timeout:300nginx.ingress.kubernetes.io/proxy-send-timeout:300cert-manager.io/cluster-issuer:letsencryptspec:tls:-hosts:-chatchat.yourcompany.comsecretName:chatchat-tlsrules:-host:chatchat.yourcompany.comhttp:paths:-path:/apipathType:Prefixbackend:service:name:chatchat-apiport:number:7861-path:/pathType:Prefixbackend:service:name:chatchat-webuiport:number:85014.6 K8s 运维命令# 部署kubectl apply-fnamespace.yaml kubectl apply-fconfigmap.yaml kubectl apply-fchatchat-api.yaml kubectl apply-fxinference.yaml kubectl apply-fingress.yaml# 查看状态kubectl get pods-nchatchat kubectl get svc-nchatchat kubectl get hpa-nchatchat# 查看日志kubectl logs-fdeployment/chatchat-api-nchatchat# 手动扩容kubectl scale deployment chatchat-api--replicas5-nchatchat# 进入容器排查kubectlexec-itdeployment/chatchat-api-nchatchat -- /bin/bash# 更新镜像滚动更新kubectlsetimage deployment/chatchat-apiapichatchat/chatchat:v2.0-nchatchat五、数据持久化和备份5.1 需要持久化的数据数据存储位置备份策略知识库文件PVC / 对象存储每日增量向量索引Milvus / FAISS每周全量对话记录PostgreSQL每日全量配置文件ConfigMap / Git版本控制模型文件PVC / 对象存储冷备5.2 备份脚本#!/bin/bash# backup.shBACKUP_DIR/backup/$(date%Y%m%d)mkdir-p$BACKUP_DIR# 1. 备份 PostgreSQLkubectlexec-nchatchat deployment/postgres --\pg_dump-Uchatchat chatchat$BACKUP_DIR/db.sql# 2. 备份知识库文件kubectlcp-nchatchat chatchat-api-xxx:/data/chatchat/knowledge_base\$BACKUP_DIR/knowledge_base# 3. 备份 Milvus调用 Milvus 备份 APIcurl-XPOST http://milvus:19530/api/v1/backup\-d{\backup_name\:\backup_$(date%Y%m%d)\}# 4. 上传到对象存储aws s3sync$BACKUP_DIRs3://chatchat-backup/# 5. 清理 7 天前的备份find/backup-typed-mtime7-execrm-rf{}\;六、高可用和灾备6.1 高可用设计┌─────────────────────────────────────────┐ │ 多可用区部署 │ │ │ │ 可用区 A 可用区 B │ │ ┌─────┐ ┌─────┐ │ │ │ API │◄────────►│ API │ │ │ └─────┘ └─────┘ │ │ │ │ │ │ ┌─────┐ ┌─────┐ │ │ │ GPU │ │ GPU │ │ │ └─────┘ └─────┘ │ │ │ │ 共享存储NAS / 对象存储 │ │ 共享数据库RDS / 自建主从 │ └─────────────────────────────────────────┘6.2 灾备策略级别RTORPO方案单点故障分钟级0多副本 健康检查单机故障分钟级0K8s 自动迁移机房故障小时级分钟级多可用区数据误删小时级天级定期备份灾难恢复天级天级异地备份七、CI/CD 流水线7.1 GitLab CI 示例# .gitlab-ci.ymlstages:-build-test-deployvariables:DOCKER_REGISTRY:registry.yourcompany.comIMAGE_NAME:chatchatbuild:stage:buildscript:-docker build-t $DOCKER_REGISTRY/$IMAGE_NAME:$CI_COMMIT_SHA .-docker push $DOCKER_REGISTRY/$IMAGE_NAME:$CI_COMMIT_SHAtest:stage:testscript:-pip install-r requirements-test.txt-pytest tests/deploy-staging:stage:deployscript:-kubectl set image deployment/chatchat-api api$DOCKER_REGISTRY/$IMAGE_NAME:$CI_COMMIT_SHA-n chatchat-stagingonly:-developdeploy-production:stage:deployscript:-kubectl set image deployment/chatchat-api api$DOCKER_REGISTRY/$IMAGE_NAME:$CI_COMMIT_SHA-n chatchatonly:-mainwhen:manual# 需要手动触发八、小结这篇咱们完成了企业级部署✅ Docker Compose 方案多服务编排、负载均衡、数据持久化✅ Kubernetes 方案Deployment、Service、HPA、Ingress✅ 数据备份知识库、向量库、数据库的备份策略✅ 高可用多副本、健康检查、多可用区✅ CI/CDGitLab CI 流水线示例生产部署 checklist□ SSL 证书配置 □ 反向代理和负载均衡 □ 服务健康检查 □ 自动扩缩容 □ 数据持久化 □ 定期备份 □ 日志收集 □ 监控告警 □ 灰度发布 □ 灾备演练你们团队目前用什么部署方案Docker Compose 还是 K8s遇到过什么坑欢迎交流