deployment-architecture.md 19 KB

 

一、介绍

根据不同的场景,apolloconfig部署的架构会有很多种,这里不讨论细节,仅从部署架构的宏观角度,来介绍各种部署的方案

1.1 flowchart

用flowchart来表达部署方式,这里先介绍一些基本的概念

1.1.1 依赖关系

依赖关系用

graph LR
	1 --> 2

表示1依赖2,也就是2必须存在,1才可以正常工作,例如

flowchart LR
	应用 --> MySQL

表示应用需要使用MySQL才可以正常工作

依赖关系可能会比较复杂,以及存在多层级的依赖,例如

flowchart LR
	服务A --> 注册中心
	服务A --> 服务B --> MySQL
	服务A --> Redis

服务A需要注册中心,服务B,Redis

并且服务B需要MySQL

1.1.2 包含关系

包含关系用

graph
	subgraph a
		b
	end

表示a包含b,也就是b是a的一部分,包含关系可能会出现嵌套的情况,例如

flowchart LR
	subgraph Linux-Server
		subgraph JVM1
			Thread1.1
			Thread1.2
		end
		subgraph JVM2
			Thread2.1
		end
		MySQL
		Redis
	end

表示在一台Linux服务器上,运行着MySQL,Redis,2个JVM,JVM里分别又存在Thread

二、单机

单机部署的场景通常是新手学习,或者公司内部对性能要求不高的测试环境,不适用于生产环境

2.1 单机,单环境 All In One

这是最简单,部署起来最方便的单机部署方式

需要:

  • 1台Linux服务器:有JRE
  • 2个database:1个PortalDB和ConfigDB

如下图,所有模块部署在同一台Linux机器上,总共有3个JVM进程

flowchart LR
	m[Meta Server]
	e[Eureka]
	c[Config Service]
	a[Admin Service]
	p[Portal]
	
	configdb[(ConfigDB)]
	portaldb[(PortalDB)]
	
	subgraph Linux Server
		subgraph JVM8080
			m
			e
			c
		end
		subgraph JVM8090
			a
		end
		subgraph JVM8070
			p
		end
	end
	
	c --> configdb
	a --> configdb
	p --> portaldb

JVM8080:对外暴露的网络端口是8080,里面有Meta Server,Eureka,Config Service,其中Config Service又使用了ConfigDB

JVM8090:对外暴露的网络端口是8090,里面有Admin Service,并且Admin Service使用了ConfigDB

JVM8070:对外暴露的网络端口是8070,里面有Portal,并且Portal使用了PortalDB

如果加入模块之间的依赖,flowchart会变成

flowchart LR
	m[Meta Server]
	e[Eureka]
	c[Config Service]
	a[Admin Service]
	p[Portal]
	
	configdb[(ConfigDB)]
	portaldb[(PortalDB)]
	
	subgraph Linux Server
		subgraph JVM8080
			m
			e
			c
		end
		subgraph JVM8090
			a
		end
		subgraph JVM8070
			p
		end
	end
	
	c --> configdb
	a --> configdb
	p --> portaldb

    m --> e
    c --> e
    a --> e
    
	p --> m
	p --> a

Config Service和Admin Service会把自己注册到Eureka上

Portal通过Meta Server服务发现Admin Service

为了flowchart看起来更加简洁,可以只表示进程之间的依赖关系

flowchart LR
	m[Meta Server]
	e[Eureka]
	c[Config Service]
	a[Admin Service]
	p[Portal]
	
	configdb[(ConfigDB)]
	portaldb[(PortalDB)]
	
	subgraph Linux Server
		subgraph JVM8080
			m
			e
			c
		end
		subgraph JVM8090
			a
		end
		subgraph JVM8070
			p
		end
	end
	
	JVM8080 --> configdb
	JVM8090 --> configdb
	JVM8070 --> portaldb
    
	JVM8090 --> JVM8080
	JVM8070 --> JVM8090

进程JVM8070依赖进程JVM8090和PortalDB

进程JVM8090依赖进程JVM8080和ConfigDB

进程JVM8080依赖ConfigDB

2.2 单机,单环境 分开部署

2.2.1 单机,单环境 分开部署 3台Linux服务器

3个JVM进程也可以分散到3台Linux机器上

需要:

  • 3台Linux服务器:分别部署3个进程
  • 2个database

    flowchart LR
    	m[Meta Server]
    	e[Eureka]
    	c[Config Service]
    	a[Admin Service]
    	p[Portal]
    	
    	configdb[(ConfigDB)]
    	portaldb[(PortalDB)]
    
    	subgraph Linux Server 1
    		subgraph JVM8080
    			m
    			e
    			c
    		end
    	end
    	
    	subgraph Linux Server 2
    		subgraph JVM8090
    			a
    		end
    	end
    	
    	subgraph Linux Server 3
    		subgraph JVM8070
    			p
    		end
    	end
    	
    	JVM8080 --> configdb
    	JVM8090 --> configdb
    	JVM8070 --> portaldb
        
    	JVM8090 --> JVM8080
    	JVM8070 --> JVM8090
    

2.2.2 单机,单环境 分开部署 2台Linux服务器

不过通常我们会把Config Service和Admin Service部署在一台Linux服务器上

需要:

  • 2台Linux服务器:1台部署Portal,另一台部署Config Service和Admin Service
  • 2个database

    flowchart LR
    	m[Meta Server]
    	e[Eureka]
    	c[Config Service]
    	a[Admin Service]
    	p[Portal]
    	
    	configdb[(ConfigDB)]
    	portaldb[(PortalDB)]
    
    	subgraph Linux Server 1
    		subgraph JVM8080
    			m
    			e
    			c
    		end
    		subgraph JVM8090
    			a
    		end
    	end
    	
    	subgraph Linux Server 2
    		subgraph JVM8070
    			p
    		end
    	end
    	
    	JVM8080 --> configdb
    	JVM8090 --> configdb
    	JVM8070 --> portaldb
        
    	JVM8090 --> JVM8080
    	JVM8070 --> JVM8090
    

后续为了flowchart更简洁,将JVM8080里的内容进行简化,只显示Config Service,里面的Meta Server和Eureka不再显示

flowchart LR
    subgraph JVM8080
        m[Meta Server]
        e[Eureka]
        c[Config Service]
    end

    subgraph new-JVM8080[JVM8080]
    	new-c[Config Service]
    end

    JVM8080 --> |simplify| new-JVM8080

所以部署架构可以简化表示成

flowchart LR
	c[Config Service]
	a[Admin Service]
	p[Portal]
	
	configdb[(ConfigDB)]
	portaldb[(PortalDB)]

	subgraph Linux Server 1
		subgraph JVM8080
			c
		end
		subgraph JVM8090
			a
		end
	end
	
	subgraph Linux Server 2
		subgraph JVM8070
			p
		end
	end
	
	JVM8080 --> configdb
	JVM8090 --> configdb
	JVM8070 --> portaldb
    
	JVM8090 --> JVM8080
	JVM8070 --> JVM8090

2.3 单机,双环境

单个环境基本没法满足实际的应用场景,例如公司里有SIT测试环境和UAT测试环境,此时需要部署2个环境提供配置服务

很容易想到的部署架构如下,把单机,单环境的部署架构重复2次即可

需要:

  • 2台Linux服务器
  • 4个database

    flowchart LR
    	subgraph SIT
        c1[SIT Config Service]
        a1[SIT Admin Service]
        p1[SIT Portal]
    
        configdb1[(SIT ConfigDB)]
        portaldb1[(SIT PortalDB)]
    
        subgraph SIT Linux Server
            subgraph sit-jvm-8080[SIT JVM8080]
                c1
            end
            subgraph sit-jvm-8090[SIT JVM8090]
                a1
            end
            subgraph sit-jvm-8070[SIT JVM8070]
                p1
            end
        end
    
        sit-jvm-8080 --> configdb1
        sit-jvm-8090 --> configdb1
        sit-jvm-8070 --> portaldb1
            
        sit-jvm-8090 --> sit-jvm-8080
        sit-jvm-8070 --> sit-jvm-8090
    	end
    
    	subgraph UAT
        c2[UAT Config Service]
        a2[UAT Admin Service]
        p2[UAT Portal]
    
        configdb2[(UAT ConfigDB)]
        portaldb2[(UAT PortalDB)]
    
        subgraph UAT Linux Server
            subgraph uat-jvm-8080[UAT JVM8080]
                c2
            end
            subgraph uat-jvm-8090[UAT JVM8090]
                a2
            end
            subgraph uat-jvm-8070[UAT JVM8070]
                p2
            end
        end
    
        uat-jvm-8080 --> configdb2
        uat-jvm-8090 --> configdb2
        uat-jvm-8070 --> portaldb2
            
        uat-jvm-8090 --> uat-jvm-8080
        uat-jvm-8070 --> uat-jvm-8090
    	end
    

但是这种方案,会存在2个Portal界面,没法1个界面管理2个环境,使用体验不是很好,Portal实际上可以只部署1套,推荐的部署架构如下

  • 3台Linux服务器:
    • Portal Linux Server单独部署Portal
    • SIT Linux Server部署SIT的Config Service和Admin Service
    • UAT Linux Server部署UAT的Config Service和Admin Service
  • 3个database:1个PortalDB + 1个SIT的ConfigDB + 1个UAT的ConfigDB

    flowchart LR
    	p[Portal]
    	portaldb[PortalDB]
    	p --> portaldb
    	
    	subgraph Portal Linux Server
    		subgraph JVM8070
    			p
    		end
    	end
    
    	subgraph SIT
        c1[SIT Config Service]
        a1[SIT Admin Service]
    
        configdb1[(SIT ConfigDB)]
    
        subgraph SIT Linux Server
            subgraph sit-jvm-8080[SIT JVM8080]
                c1
            end
            subgraph sit-jvm-8090[SIT JVM8090]
                a1
            end
        end
    
        sit-jvm-8080 --> configdb1
        sit-jvm-8090 --> configdb1
            
        sit-jvm-8090 --> sit-jvm-8080
    	end
    
    	subgraph UAT
        c2[UAT Config Service]
        a2[UAT Admin Service]
    
        configdb2[(UAT ConfigDB)]
    
        subgraph UAT Linux Server
            subgraph uat-jvm-8080[UAT JVM8080]
                c2
            end
            subgraph uat-jvm-8090[UAT JVM8090]
                a2
            end
        end
    
        uat-jvm-8080 --> configdb2
        uat-jvm-8090 --> configdb2
            
        uat-jvm-8090 --> uat-jvm-8080
    	end
    
    	JVM8070 --> sit-jvm-8090
    	JVM8070 --> uat-jvm-8090
    

2.4 单机,三个环境

假设现在需要满足SIT、UAT、PP这3个环境的使用场景,

在之前双环境的基础之上,再多加1台PP环境的Linux服务和ConfigDB即可,Portal通过修改配置的方式,来管理这3个环境

flowchart LR
	p[Portal]
	portaldb[PortalDB]
	p --> portaldb
	
	subgraph Portal Linux Server
		subgraph JVM8070
			p
		end
	end

	subgraph SIT
        c1[SIT Config Service]
        a1[SIT Admin Service]

        configdb1[(SIT ConfigDB)]

        subgraph SIT Linux Server
            subgraph sit-jvm-8080[SIT JVM8080]
                c1
            end
            subgraph sit-jvm-8090[SIT JVM8090]
                a1
            end
        end

        sit-jvm-8080 --> configdb1
        sit-jvm-8090 --> configdb1
        
        sit-jvm-8090 --> sit-jvm-8080
	end

	subgraph UAT
        c2[UAT Config Service]
        a2[UAT Admin Service]

        configdb2[(UAT ConfigDB)]

        subgraph UAT Linux Server
            subgraph uat-jvm-8080[UAT JVM8080]
                c2
            end
            subgraph uat-jvm-8090[UAT JVM8090]
                a2
            end
        end

        uat-jvm-8080 --> configdb2
        uat-jvm-8090 --> configdb2
        
        uat-jvm-8090 --> uat-jvm-8080
	end
	
	subgraph PP
        c3[PP Config Service]
        a3[PP Admin Service]

        configdb3[(PP ConfigDB)]

        subgraph PP Linux Server
            subgraph pp-jvm-8080[PP JVM8080]
                c3
            end
            subgraph pp-jvm-8090[PP JVM8090]
                a3
            end
        end

        pp-jvm-8080 --> configdb3
        pp-jvm-8090 --> configdb3
        
        pp-jvm-8090 --> pp-jvm-8080
	end

	JVM8070 --> sit-jvm-8090
	JVM8070 --> uat-jvm-8090
	JVM8070 --> pp-jvm-8090

2.5 单机,多个环境

原理同上,每个环境1台Linux服务器+1个ConfigDB

然后Portal添加新环境的信息即可

三、高可用

1个环境只有1个Config Service进程,无法满足高可用,为了避免单点宕机后影响系统的可用性,需要多实例部署,也就是部署多个Java进程在不同的Linux服务器上

3.1 最简高可用,单环境

回到常见的非高可用部署方式,

flowchart LR
	c[Config Service]
	a[Admin Service]
	p[Portal]
	
	configdb[(ConfigDB)]
	portaldb[(PortalDB)]

	subgraph Linux Server 1
		subgraph JVM8080
			c
		end
		subgraph JVM8090
			a
		end
	end
	
	subgraph Linux Server 2
		subgraph JVM8070
			p
		end
	end
	
	JVM8080 --> configdb
	JVM8090 --> configdb
	JVM8070 --> portaldb
    
	JVM8090 --> JVM8080
	JVM8070 --> JVM8090

当Linux Server 1宕机时,client就只能读取本地磁盘上的config-cache了,如果需要防止单台Linux宕机导致Config Service不可用,可以尝试再新增1台Linux机器

需要

  • 3台Linux服务器:1台部署Portal,另外2台分别部署Config Service和Admin Service
  • 2个database

    flowchart LR
    	c-1[Config Service]
    	c-2[Config Service]
    	a-1[Admin Service]
    	a-2[Admin Service]
    	p[Portal]
    	
    	configdb[(ConfigDB)]
    	portaldb[(PortalDB)]
    	
    	JVM8080-1[JVM8080]
    	JVM8080-2[JVM8080]
    	
    	JVM8090-1[JVM8090]
    	JVM8090-2[JVM8090]
    
    	subgraph Linux Server 1.1
    		subgraph JVM8080-1[JVM8080]
    			c-1
    		end
    		subgraph JVM8090-1[JVM8090]
    			a-1
    		end
    	end
    	subgraph Linux Server 1.2
    		subgraph JVM8080-2[JVM8080]
    			c-2
    		end
    		subgraph JVM8090-2[JVM8090]
    			a-2
    		end
    	end
    	
    	subgraph Linux Server 2
    		subgraph JVM8070
    			p
    		end
    	end
    	
    	JVM8080-1 --> configdb
    	JVM8090-1 --> configdb
    	JVM8080-2 --> configdb
    	JVM8090-2 --> configdb
    	
    	JVM8070 --> portaldb
        
    	JVM8090-1 --> JVM8080-1
    	JVM8090-2 --> JVM8080-2
    	
    	JVM8070 --> JVM8090-1
    	JVM8070 --> JVM8090-2
    

这种部署方式下,Linux Server 1.1 或者 Linux Server 1.2宕机,系统仍旧可用,

3.2 高可用,单环境

在上述的基础上,如果client的数量有很多(例如上万个Java进程),可以横向扩展Config Service,引入Linux Server 1.3, Linux Server 1.4, ...

Admin Service由于只有Portal访问,在数量上可以比Config Service少很多

具体如何评定Config Service的数量,请参考 Apollo性能测试报告

3.3 高可用,双环境

2.3 单机,双环境种,如果想让SIT和UAT都变成高可用,只需要分别在环境中再添加机器即可,如下图,每个环境中各有2台Linux Server,如果有性能上需求,可以再在每个环境中,使用更多的机器来部署Config Service即可

flowchart LR
	p[Portal]
	portaldb[(PortalDB)]
	p --> portaldb
	
	subgraph Portal Linux Server
		subgraph JVM8070
			p
		end
	end

	subgraph SIT
        sit-c1[SIT Config Service]
        sit-a1[SIT Admin Service]
        sit-c2[SIT Config Service]
        sit-a2[SIT Admin Service]
        
        sit-configdb[(SIT ConfigDB)]

        subgraph SIT Linux Server 2.1
            subgraph sit-c1-jvm-8080[SIT JVM8080]
                sit-c1
            end
            subgraph sit-c1-jvm-8090[SIT JVM8090]
                sit-a1
            end
        end
        
        subgraph SIT Linux Server 2.2
            subgraph sit-c2-jvm-8080[SIT JVM8080]
                sit-c2
            end
            subgraph sit-c2-jvm-8090[SIT JVM8090]
                sit-a2
            end
        end
        
        sit-c1-jvm-8080 --> sit-configdb
        sit-c1-jvm-8090 --> sit-configdb
        sit-c2-jvm-8080 --> sit-configdb
        sit-c2-jvm-8090 --> sit-configdb
        
        sit-c1-jvm-8090 --> sit-c1-jvm-8080
        sit-c2-jvm-8090 --> sit-c2-jvm-8080
	end

	subgraph UAT
        uat-c1[UAT Config Service]
        uat-a1[UAT Admin Service]
        uat-c2[UAT Config Service]
        uat-a2[UAT Admin Service]
        
        uat-configdb[(UAT ConfigDB)]

        subgraph UAT Linux Server 2.1
            subgraph uat-c1-jvm-8080[UAT JVM8080]
                uat-c1
            end
            subgraph uat-c1-jvm-8090[UAT JVM8090]
                uat-a1
            end
        end
        
        subgraph UAT Linux Server 2.2
            subgraph uat-c2-jvm-8080[UAT JVM8080]
                uat-c2
            end
            subgraph uat-c2-jvm-8090[UAT JVM8090]
                uat-a2
            end
        end
        
        uat-c1-jvm-8080 --> uat-configdb
        uat-c1-jvm-8090 --> uat-configdb
        uat-c2-jvm-8080 --> uat-configdb
        uat-c2-jvm-8090 --> uat-configdb
        
        uat-c1-jvm-8090 --> uat-c1-jvm-8080
        uat-c2-jvm-8090 --> uat-c2-jvm-8080
	end

	JVM8070 --> sit-c1-jvm-8090
	JVM8070 --> sit-c2-jvm-8090
	
	JVM8070 --> uat-c1-jvm-8090
	JVM8070 --> uat-c2-jvm-8090

3.4 高可用,多个环境

在上述的基础上,如果要添加一个环境,例如BETA环境,需要新增2台及以上的Linux服务器+1个ConfigDB

Portal添加新环境的信息,指向BETA环境的apollo.meta

3.5 高可用,单环境,单机房

实际生产环境中,很多公司和测试环境进行了隔离,所以生产环境属于单环境,只有一个PRO环境

在只有1个机房时,参考 3.2 高可用,单环境

3.6 高可用,单环境,双机房

如果有2个机房,通常机房之间存在网络隔离,如果是同城机房,idc1和idc2,可以采用如下的部署方式

flowchart LR
	idc1-p[idc1 Portal]
	idc2-p[idc2 Portal]
	portaldb[(PortalDB)]
	idc1-p --> portaldb
	idc2-p --> portaldb

	configdb[(ConfigDB)]

	idc1-c1-jvm-8080 --> configdb
	idc1-c1-jvm-8090 --> configdb
	idc1-c2-jvm-8080 --> configdb
	idc1-c2-jvm-8090 --> configdb

	idc2-c1-jvm-8080 --> configdb
	idc2-c1-jvm-8090 --> configdb
	idc2-c2-jvm-8080 --> configdb
	idc2-c2-jvm-8090 --> configdb

	subgraph idc1
		subgraph idc1 Portal Linux Server
			subgraph idc1-JVM8070
				idc1-p
			end
		end
	
		idc1-c1[idc1 Config Service]
		idc1-a1[idc1 Admin Service]
		idc1-c2[idc1 Config Service]
		idc1-a2[idc1 Admin Service]
		
		

		subgraph idc1 Linux Server 2.1
			subgraph idc1-c1-jvm-8080[idc1 JVM8080]
				idc1-c1
			end
			subgraph idc1-c1-jvm-8090[idc1 JVM8090]
				idc1-a1
			end
		end
		
		subgraph idc1 Linux Server 2.2
			subgraph idc1-c2-jvm-8080[idc1 JVM8080]
				idc1-c2
			end
			subgraph idc1-c2-jvm-8090[idc1 JVM8090]
				idc1-a2
			end
		end
		
		idc1-c1-jvm-8090 --> idc1-c1-jvm-8080
		idc1-c2-jvm-8090 --> idc1-c2-jvm-8080
	end

	subgraph idc2
		subgraph idc2 Portal Linux Server
			subgraph idc2-JVM8070
				idc2-p
			end
		end
		idc2-c1[idc2 Config Service]
		idc2-a1[idc2 Admin Service]
		idc2-c2[idc2 Config Service]
		idc2-a2[idc2 Admin Service]

		subgraph idc2 Linux Server 2.1
			subgraph idc2-c1-jvm-8080[idc2 JVM8080]
				idc2-c1
			end
			subgraph idc2-c1-jvm-8090[idc2 JVM8090]
				idc2-a1
			end
		end
		
		subgraph idc2 Linux Server 2.2
			subgraph idc2-c2-jvm-8080[idc2 JVM8080]
				idc2-c2
			end
			subgraph idc2-c2-jvm-8090[idc2 JVM8090]
				idc2-a2
			end
		end
		
		idc2-c1-jvm-8090 --> idc2-c1-jvm-8080
		idc2-c2-jvm-8090 --> idc2-c2-jvm-8080
	end


	idc1-JVM8070 --> idc1-c1-jvm-8090
	idc1-JVM8070 --> idc1-c2-jvm-8090
	
	idc2-JVM8070 --> idc2-c1-jvm-8090
	idc2-JVM8070 --> idc2-c2-jvm-8090

每个机房有自己的一套Portal, Config Service, Admin Service

对于ConfigDB,在同城双机房下,连接的ConfigDB是同一个,不存在2个不同的ConfigDB,对于PortalDB也是如此,需要连接同一个

ConfigDB和PortalDB在图中没有放入idc1或者idc2,需要自行选用合适的MySQL架构以及部署方式

四、部署图

4.1 ctrip

以ctrip为例,我们的部署策略如下: Deployment

  • Portal部署在生产环境的机房,通过它来直接管理FAT、UAT、PRO等环境的配置
  • Meta Server、Config Service和Admin Service在每个环境都单独部署,使用独立的数据库
  • Meta Server、Config Service和Admin Service在生产环境部署在两个机房,实现双活
  • Meta Server和Config Service部署在同一个JVM进程内,Admin Service部署在同一台服务器的另一个JVM进程内

4.2 样例部署图

@lyliyongblue 贡献的样例部署图(建议右键新窗口打开看大图):

Deployment