我们知道mongodb是分布式文档存储数据库,而mongodb部署架构分为单机,可复制集和分片群集。其中,分片群集是比较复杂,而且运维难度比较高的。很多人对于mongodb分片集群都不是很理解,这篇文章就给大家介绍一下MongoDB分片集群的问题。
1、什么是分片?为什么要分片?
我们知道数据库服务器一般出现瓶颈是在磁盘io上,或者高并发网络io,又或者单台server的cpu、内存等等一系列原因。于是,为了解决这些瓶颈问题,我们就必须扩展服务器性能;通常扩展服务器有向上扩展和向外扩展,所谓向上扩展就是给服务器加更大的磁盘,使用更大更好的内存,更换更好的cpu。这种扩展在一定程度上是可以解决性能瓶颈问题,但随着数据量大增大,瓶颈会再次出现;所以通常这种向上扩展的方式不推荐。
向外扩展是指一台服务器不够加两台,两台不够加三台,以这种方式扩展,只要出现瓶颈我们就可以使用增加服务器来解决;这样一来服务器性能解决了,但用户的读写怎么分散到多个服务器上去呢?所以我们还要想办法把数据切分成多块,让每个服务器只保存整个数据集的部分数据,这样一来使得原来一个很大的数据集就通过切片的方式,把它切分成多分,分散的存放在多个服务器上,这就是分片。
分片是可以有效解决用户写操作性能瓶颈,虽然解决了服务器性能问题和用户写性能问题,同时也带来了一个新问题,就是用户的查询。我们把整个数据集分散到多个server上,那么用户怎么查询数据呢?比如用户要查询年龄大于30的用户,该怎么查询呢?而年龄大于30的用户的数据,可能server1上有一部分数据,server2上有部分数据,我们怎么才能够把所有满足条件的数据全部查询到呢?
这个场景有点类似我们之前说的mogilefs的架构,用户上传图片到mogilefs首先要把图片的元数据写进tracker,然后在把数据存放在对应的data节点,这样一来用户来查询,首先找tracker节点,tracker会把用户的请求文件的元数据告诉客户端,然后客户端在到对应的data节点取数据,最后拼凑成一张图片。而在mongodb上也是很类似,不同的的是在mogilefs上,客户端需要自己去和后端的data节点交互,取出数据。在mongdb上客户端不需要直接和后端的data节点交互,而是通过mongodb专有的客户端代理去代客户端交互,最后把数据统一由代理返回给客户端,这样一来就可以解决用户的查询问题。简单讲所谓分片就是把一个大的数据集通过切分的方式切分成多分,分散的存放在多个服务器上,分片的目的是为了解决数据量过大而导致的性能问题。
2、数据集分片示意图
提示:我们通过分片,可以将原本1T的数据集,平均分成4分,每个节点存储原有数据集的1/4,使得原来用一台服务器处理1T的数据,现在可以用4台服务器来处理,这样一来就有效的提高了数据处理过程;这也是分布式系统的意义;在mongodb中我们把这种共同处理一个数据集的部分数据的节点叫shard,我们把使用这种分片机制的mongodb集群就叫做mongodb分片集群;
3、mongodb分片集群架构
提示:在mongodb分片集群中,通常有三类角色,第一类是router角色,router角色主要用来接收客户端的读写请求,主要运行mongos这个服务;为了使得router角色的高可用,通常会用多个节点来组成router高可用集群;第二类是config server,这类角色主要用来保存mongodb分片集群中的数据和集群的元数据信息,有点类似mogilefs中的tracker的作用;为了保证config server的高可用性,通常config server也会将其运行为一个副本集;第三类是shard角色,这类角色主要用来存放数据,类似mogilefs的数据节点,为了保证数据的高可用和完整性,通常每个shard是一个副本集;
4、mongodb分片集群工作过程
首先用户将请求发送给router,router接收到用户请求,然后去找config server拿对应请求的元数据信息,router拿到元数据信息后,然后再向对应的shard请求数据,最后将数据整合后响应给用户;在这个过程中router 就相当于mongodb的一个客户端代理;而config server用来存放数据的元数据信息,这些信息主要包含了那些shard上存放了那些数据,对应的那些数据存放在那些shard上,和mogilefs上的tracker非常类似,主要存放了两张表,一个是以数据为中心的一张表,一个是以shard节点为中心的一张表;
5、mongodb是怎么分片的?
在mongodb的分片集群中,分片是按照collection字段来分的,我们把指定的字段叫shard key;根据shard key的取值不同和应用场景,我们可以基于shard key取值范围来分片,也可以基于shard key做hash分片;分好片以后将结果保存在config server上;在configserver 上保存了每一个分片对应的数据集;比如我们基于shardkey的范围来分片,在configserver上就记录了一个连续范围的shardkey的值都保存在一个分片上;如下图
上图主要描述了基于范围的分片,从shardkey最小值到最大值进行分片,把最小值到-75这个范围值的数据块保存在第一个分片上,把-75到25这个范围值的数据块保存在第二个分片上,依次类推;这种基于范围的分片,很容易导致某个分片上的数据过大,而有的分片上的数据又很小,造成分片数据不均匀;所以除了基与shard key的值的范围分片,也可以基于shard key的值做hash分片,如下图
基于hash分片,主要是对shardkey做hash计算后,然后根据最后的结果落在哪个分片上就把对应的数据块保存在对应的分片上;比如我们把shandkey做hash计算,然后对分片数量进行取模计算,如果得到的结果是0,那么就把对应的数据块保存在第一个分片上,如果取得到结果是1就保存在第二个分片上依次类推;这种基于hash分片,就有效的降低分片数据不均衡的情况,因为hash计算的值是散列的;
除了上述两种切片的方式以外,我们还可以根据区域切片,也叫基于列表切片,如下图
上图主要描述了基于区域分片,这种分片一般是针对shardkey的取值范围不是一个顺序的集合,而是一个离散的集合,比如我们可用这种方式对全国省份这个字段做切片,把流量特别大的省份单独切一个片,把流量小的几个省份组合切分一片,把国外的访问或不是国内省份的切分为一片;这种切片有点类似给shardkey做分类;不管用什么方式去做分片,我们尽可能的遵循写操作要越分散越好,读操作要越集中越好;
6、mongodb分片集群搭建
环境说明
主机名 | 角色 | ip地址 |
node01 | router | 192.168.0.41 |
node02/node03/node04 | config server replication set | 192.168.0.42 192.168.0.43 192.168.0.44 |
node05/node06/node07 | shard1 replication set | 192.168.0.45 192.168.0.46 192.168.0.47 |
node08/node09/node10 | shard2 replication set | 192.168.0.48 192.168.0.49 192.168.0.50 |
基础环境,各server做时间同步,关闭防火墙,关闭selinux,ssh互信,主机名解析
主机名解析
[root@node01 ~]# cat /etc/hosts 127.0.0.1 localhost localhost.localdomain localhost4 localhost4.localdomain4 ::1 localhost localhost.localdomain localhost6 localhost6.localdomain6 192.168.0.99 time.test.org time-node 192.168.0.41 node01.test.org node01 192.168.0.42 node02.test.org node02 192.168.0.43 node03.test.org node03 192.168.0.44 node04.test.org node04 192.168.0.45 node05.test.org node05 192.168.0.46 node06.test.org node06 192.168.0.47 node07.test.org node07 192.168.0.48 node08.test.org node08 192.168.0.49 node09.test.org node09 192.168.0.50 node10.test.org node10 192.168.0.51 node11.test.org node11 192.168.0.52 node12.test.org node12 [root@node01 ~]#
准备好基础环境以后,配置mongodb yum源
[root@node01 ~]# cat /etc/yum.repos.d/mongodb.repo [mongodb-org] name = MongoDB Repository baseurl = https://mirrors.aliyun.com/mongodb/yum/redhat/7/mongodb-org/4.4/x86_64/ gpgcheck = 1 enabled = 1 gpgkey = https://www.mongodb.org/static/pgp/server-4.4.asc [root@node01 ~]#
将mongodb yum源复制给其他节点
[root@node01 ~]# for i in {02..10} ; do scp /etc/yum.repos.d/mongodb.repo node$i:/etc/yum.repos.d/; done mongodb.repo 100% 206 247.2KB/s 00:00 mongodb.repo 100% 206 222.3KB/s 00:00 mongodb.repo 100% 206 118.7KB/s 00:00 mongodb.repo 100% 206 164.0KB/s 00:00 mongodb.repo 100% 206 145.2KB/s 00:00 mongodb.repo 100% 206 119.9KB/s 00:00 mongodb.repo 100% 206 219.2KB/s 00:00 mongodb.repo 100% 206 302.1KB/s 00:00 mongodb.repo 100% 206 289.3KB/s 00:00 [root@node01 ~]#
在每个节点上安装mongodb-org这个包
for i in {01..10} ; do ssh node$i ' yum -y install mongodb-org '; done
在config server 和shard节点上创建数据目录和日志目录,并将其属主和属组更改为mongod
[root@node01 ~]# for i in {02..10} ; do ssh node$i 'mkdir -p /mongodb/{data,log} && chown -R mongod.mongod /mongodb/ && ls -ld /mongodb'; done drwxr-xr-x 4 mongod mongod 29 Nov 11 22:47 /mongodb drwxr-xr-x 4 mongod mongod 29 Nov 11 22:45 /mongodb drwxr-xr-x 4 mongod mongod 29 Nov 11 22:45 /mongodb drwxr-xr-x 4 mongod mongod 29 Nov 11 22:45 /mongodb drwxr-xr-x 4 mongod mongod 29 Nov 11 22:45 /mongodb drwxr-xr-x 4 mongod mongod 29 Nov 11 22:45 /mongodb drwxr-xr-x 4 mongod mongod 29 Nov 11 22:45 /mongodb drwxr-xr-x 4 mongod mongod 29 Nov 11 22:45 /mongodb drwxr-xr-x 4 mongod mongod 29 Nov 11 22:45 /mongodb [root@node01 ~]#
配置shard1 replication set
[root@node05 ~]# cat /etc/mongod.conf systemLog: destination: file logAppend: true path: /mongodb/log/mongod.log storage: dbPath: /mongodb/data/ journal: enabled: true processManagement: fork: true pidFilePath: /var/run/mongodb/mongod.pid timeZoneInfo: /usr/share/zoneinfo net: bindIp: 0.0.0.0 sharding: clusterRole: shardsvr replication: replSetName: shard1_replset [root@node05 ~]# scp /etc/mongod.conf node06:/etc/ mongod.conf 100% 360 394.5KB/s 00:00 [root@node05 ~]# scp /etc/mongod.conf node07:/etc/ mongod.conf 100% 360 351.7KB/s 00:00 [root@node05 ~]#
配置shard2 replication set
[root@node08 ~]# cat /etc/mongod.conf systemLog: destination: file logAppend: true path: /mongodb/log/mongod.log storage: dbPath: /mongodb/data/ journal: enabled: true processManagement: fork: true pidFilePath: /var/run/mongodb/mongod.pid timeZoneInfo: /usr/share/zoneinfo net: bindIp: 0.0.0.0 sharding: clusterRole: shardsvr replication: replSetName: shard2_replset [root@node08 ~]# scp /etc/mongod.conf node09:/etc/ mongod.conf 100% 360 330.9KB/s 00:00 [root@node08 ~]# scp /etc/mongod.conf node10:/etc/ mongod.conf 100% 360 385.9KB/s 00:00 [root@node08 ~]#
启动shard1 replication set和shard2 replication set
[root@node05 ~]# systemctl start mongod.service [root@node05 ~]# ss -tnl State Recv-Q Send-Q Local Address:Port Peer Address:Port LISTEN 0 128 *:22 *:* LISTEN 0 100 127.0.0.1:25 *:* LISTEN 0 128 *:27018 *:* LISTEN 0 128 :::22 :::* LISTEN 0 100 ::1:25 :::* [root@node05 ~]#for i in {06..10} ; do ssh node$i 'systemctl start mongod.service && ss -tnl';done State Recv-Q Send-Q Local Address:Port Peer Address:Port LISTEN 0 128 *:22 *:* LISTEN 0 100 127.0.0.1:25 *:* LISTEN 0 128 *:27018 *:* LISTEN 0 128 :::22 :::* LISTEN 0 100 ::1:25 :::* State Recv-Q Send-Q Local Address:Port Peer Address:Port LISTEN 0 128 *:22 *:* LISTEN 0 100 127.0.0.1:25 *:* LISTEN 0 128 *:27018 *:* LISTEN 0 128 :::22 :::* LISTEN 0 100 ::1:25 :::* State Recv-Q Send-Q Local Address:Port Peer Address:Port LISTEN 0 128 *:22 *:* LISTEN 0 100 127.0.0.1:25 *:* LISTEN 0 128 *:27018 *:* LISTEN 0 128 :::22 :::* LISTEN 0 100 ::1:25 :::* State Recv-Q Send-Q Local Address:Port Peer Address:Port LISTEN 0 128 *:22 *:* LISTEN 0 100 127.0.0.1:25 *:* LISTEN 0 128 *:27018 *:* LISTEN 0 128 :::22 :::* LISTEN 0 100 ::1:25 :::* State Recv-Q Send-Q Local Address:Port Peer Address:Port LISTEN 0 128 *:22 *:* LISTEN 0 100 127.0.0.1:25 *:* LISTEN 0 128 *:27018 *:* LISTEN 0 128 :::22 :::* LISTEN 0 100 ::1:25 :::* [root@node05 ~]#
提示:默认不指定shard监听端口,它默认就监听在27018端口,所以启动shard节点后,请确保27018端口正常监听即可;
连接node05的mongodb 初始化shard1_replset副本集
> rs.initiate( ... { ... _id : "shard1_replset", ... members: [ ... { _id : 0, host : "node05:27018" }, ... { _id : 1, host : "node06:27018" }, ... { _id : 2, host : "node07:27018" } ... ] ... } ... ) { "ok" : 1, "$clusterTime" : { "clusterTime" : Timestamp(1605107401, 1), "signature" : { "hash" : BinData(0,"AAAAAAAAAAAAAAAAAAAAAAAAAAA="), "keyId" : NumberLong(0) } }, "operationTime" : Timestamp(1605107401, 1) } shard1_replset:SECONDARY>
连接node08的mongodb 初始化shard2_replset副本集
> rs.initiate( ... { ... _id : "shard2_replset", ... members: [ ... { _id : 0, host : "node08:27018" }, ... { _id : 1, host : "node09:27018" }, ... { _id : 2, host : "node10:27018" } ... ] ... } ... ) { "ok" : 1, "$clusterTime" : { "clusterTime" : Timestamp(1605107644, 1), "signature" : { "hash" : BinData(0,"AAAAAAAAAAAAAAAAAAAAAAAAAAA="), "keyId" : NumberLong(0) } }, "operationTime" : Timestamp(1605107644, 1) } shard2_replset:OTHER>
配置configserver replication set
[root@node02 ~]# cat /etc/mongod.conf systemLog: destination: file logAppend: true path: /mongodb/log/mongod.log storage: dbPath: /mongodb/data/ journal: enabled: true processManagement: fork: true pidFilePath: /var/run/mongodb/mongod.pid timeZoneInfo: /usr/share/zoneinfo net: bindIp: 0.0.0.0 sharding: clusterRole: configsvr replication: replSetName: cfg_replset [root@node02 ~]# scp /etc/mongod.conf node03:/etc/mongod.conf mongod.conf 100% 358 398.9KB/s 00:00 [root@node02 ~]# scp /etc/mongod.conf node04:/etc/mongod.conf mongod.conf 100% 358 270.7KB/s 00:00 [root@node02 ~]#
启动config server
[root@node02 ~]# systemctl start mongod.service [root@node02 ~]# ss -tnl State Recv-Q Send-Q Local Address:Port Peer Address:Port LISTEN 0 128 *:27019 *:* LISTEN 0 128 *:22 *:* LISTEN 0 100 127.0.0.1:25 *:* LISTEN 0 128 :::22 :::* LISTEN 0 100 ::1:25 :::* [root@node02 ~]# ssh node03 'systemctl start mongod.service && ss -tnl' State Recv-Q Send-Q Local Address:Port Peer Address:Port LISTEN 0 128 *:27019 *:* LISTEN 0 128 *:22 *:* LISTEN 0 100 127.0.0.1:25 *:* LISTEN 0 128 :::22 :::* LISTEN 0 100 ::1:25 :::* [root@node02 ~]# ssh node04 'systemctl start mongod.service && ss -tnl' State Recv-Q Send-Q Local Address:Port Peer Address:Port LISTEN 0 128 *:27019 *:* LISTEN 0 128 *:22 *:* LISTEN 0 100 127.0.0.1:25 *:* LISTEN 0 128 :::22 :::* LISTEN 0 100 ::1:25 :::* [root@node02 ~]#
提示:config server 默认在不指定端口的情况监听在27019这个端口,启动后,请确保该端口处于正常监听;
连接node02的mongodb,初始化cfg_replset 副本集
> rs.initiate( ... { ... _id: "cfg_replset", ... configsvr: true, ... members: [ ... { _id : 0, host : "node02:27019" }, ... { _id : 1, host : "node03:27019" }, ... { _id : 2, host : "node04:27019" } ... ] ... } ... ) { "ok" : 1, "$gleStats" : { "lastOpTime" : Timestamp(1605108177, 1), "electionId" : ObjectId("000000000000000000000000") }, "lastCommittedOpTime" : Timestamp(0, 0), "$clusterTime" : { "clusterTime" : Timestamp(1605108177, 1), "signature" : { "hash" : BinData(0,"AAAAAAAAAAAAAAAAAAAAAAAAAAA="), "keyId" : NumberLong(0) } }, "operationTime" : Timestamp(1605108177, 1) } cfg_replset:SECONDARY>
配置router
[root@node01 ~]# cat /etc/mongos.conf systemLog: destination: file path: /var/log/mongodb/mongos.log logAppend: true processManagement: fork: true net: bindIp: 0.0.0.0 sharding: configDB: "cfg_replset/node02:27019,node03:27019,node04:27019" [root@node01 ~]#
提示:configDB必须是副本集名称/成员监听地址:port的形式,成员至少要写一个;
启动router
[root@node01 ~]# mongos -f /etc/mongos.conf about to fork child process, waiting until server is ready for connections. forked process: 1510 child process started successfully, parent exiting [root@node01 ~]# ss -tnl State Recv-Q Send-Q Local Address:Port Peer Address:Port LISTEN 0 128 *:22 *:* LISTEN 0 100 127.0.0.1:25 *:* LISTEN 0 128 *:27017 *:* LISTEN 0 128 :::22 :::* LISTEN 0 100 ::1:25 :::* [root@node01 ~]#
连接mongos,添加shard1 replication set 和shard2 replication set
mongos> sh.addShard("shard1_replset/node05:27018,node06:27018,node07:27018") { "shardAdded" : "shard1_replset", "ok" : 1, "operationTime" : Timestamp(1605109085, 3), "$clusterTime" : { "clusterTime" : Timestamp(1605109086, 1), "signature" : { "hash" : BinData(0,"AAAAAAAAAAAAAAAAAAAAAAAAAAA="), "keyId" : NumberLong(0) } } } mongos> sh.addShard("shard2_replset/node08:27018,node09:27018,node10:27018") { "shardAdded" : "shard2_replset", "ok" : 1, "operationTime" : Timestamp(1605109118, 2), "$clusterTime" : { "clusterTime" : Timestamp(1605109118, 3), "signature" : { "hash" : BinData(0,"AAAAAAAAAAAAAAAAAAAAAAAAAAA="), "keyId" : NumberLong(0) } } } mongos>
提示:添加shard 副本集也是需要指明副本集名称/成员的格式添加;
到此分片集群就配置好了
查看sharding 集群状态
mongos> sh.status() --- Sharding Status --- sharding version: { "_id" : 1, "minCompatibleVersion" : 5, "currentVersion" : 6, "clusterId" : ObjectId("5fac01dd8d6fa3fe899662c8") } shards: { "_id" : "shard1_replset", "host" : "shard1_replset/node05:27018,node06:27018,node07:27018", "state" : 1 } { "_id" : "shard2_replset", "host" : "shard2_replset/node08:27018,node09:27018,node10:27018", "state" : 1 } active mongoses: "4.4.1" : 1 autosplit: Currently enabled: yes balancer: Currently enabled: yes Currently running: yes Collections with active migrations: config.system.sessions started at Wed Nov 11 2020 23:43:14 GMT+0800 (CST) Failed balancer rounds in last 5 attempts: 0 Migration Results for the last 24 hours: 45 : Success databases: { "_id" : "config", "primary" : "config", "partitioned" : true } config.system.sessions shard key: { "_id" : 1 } unique: false balancing: true chunks: shard1_replset 978 shard2_replset 46 too many chunks to print, use verbose if you want to force print mongos>
提示:可以看到当前分片集群中有两个shard 副本集,分别是shard1_replset和shard2_replset;以及一个config server
对testdb数据库启用sharding功能
mongos> sh.enableSharding("testdb") { "ok" : 1, "operationTime" : Timestamp(1605109993, 9), "$clusterTime" : { "clusterTime" : Timestamp(1605109993, 9), "signature" : { "hash" : BinData(0,"AAAAAAAAAAAAAAAAAAAAAAAAAAA="), "keyId" : NumberLong(0) } } } mongos> sh.status() --- Sharding Status --- sharding version: { "_id" : 1, "minCompatibleVersion" : 5, "currentVersion" : 6, "clusterId" : ObjectId("5fac01dd8d6fa3fe899662c8") } shards: { "_id" : "shard1_replset", "host" : "shard1_replset/node05:27018,node06:27018,node07:27018", "state" : 1 } { "_id" : "shard2_replset", "host" : "shard2_replset/node08:27018,node09:27018,node10:27018", "state" : 1 } active mongoses: "4.4.1" : 1 autosplit: Currently enabled: yes balancer: Currently enabled: yes Currently running: no Failed balancer rounds in last 5 attempts: 0 Migration Results for the last 24 hours: 214 : Success databases: { "_id" : "config", "primary" : "config", "partitioned" : true } config.system.sessions shard key: { "_id" : 1 } unique: false balancing: true chunks: shard1_replset 810 shard2_replset 214 too many chunks to print, use verbose if you want to force print { "_id" : "testdb", "primary" : "shard2_replset", "partitioned" : true, "version" : { "uuid" : UUID("454aad2e-b397-4c88-b5c4-c3b21d37e480"), "lastMod" : 1 } } mongos>
提示:在对某个数据库启动sharding功能后,它会给我们分片一个主shard所谓主shard是用来存放该数据库下没有做分片的colleciton;对于分片的collection会分散在各个shard上;
启用对testdb库下的peoples集合启动sharding,并指明在age字段上做基于范围的分片
mongos> sh.shardCollection("testdb.peoples",{"age":1}) { "collectionsharded" : "testdb.peoples", "collectionUUID" : UUID("ec095411-240d-4484-b45d-b541c33c3975"), "ok" : 1, "operationTime" : Timestamp(1605110694, 11), "$clusterTime" : { "clusterTime" : Timestamp(1605110694, 11), "signature" : { "hash" : BinData(0,"AAAAAAAAAAAAAAAAAAAAAAAAAAA="), "keyId" : NumberLong(0) } } } mongos> sh.status() --- Sharding Status --- sharding version: { "_id" : 1, "minCompatibleVersion" : 5, "currentVersion" : 6, "clusterId" : ObjectId("5fac01dd8d6fa3fe899662c8") } shards: { "_id" : "shard1_replset", "host" : "shard1_replset/node05:27018,node06:27018,node07:27018", "state" : 1 } { "_id" : "shard2_replset", "host" : "shard2_replset/node08:27018,node09:27018,node10:27018", "state" : 1 } active mongoses: "4.4.1" : 1 autosplit: Currently enabled: yes balancer: Currently enabled: yes Currently running: no Failed balancer rounds in last 5 attempts: 0 Migration Results for the last 24 hours: 408 : Success databases: { "_id" : "config", "primary" : "config", "partitioned" : true } config.system.sessions shard key: { "_id" : 1 } unique: false balancing: true chunks: shard1_replset 616 shard2_replset 408 too many chunks to print, use verbose if you want to force print { "_id" : "testdb", "primary" : "shard2_replset", "partitioned" : true, "version" : { "uuid" : UUID("454aad2e-b397-4c88-b5c4-c3b21d37e480"), "lastMod" : 1 } } testdb.peoples shard key: { "age" : 1 } unique: false balancing: true chunks: shard2_replset 1 { "age" : { "$minKey" : 1 } } -->> { "age" : { "$maxKey" : 1 } } on : shard2_replset Timestamp(1, 0) mongos>
提示:如果对应的collection存在,我们还需要先对collection创建shardkey索引,然后在使用sh.shardCollection()来对colleciton启用sharding功能;基于范围做分片,我们可以在多个字段上做;
基于hash做分片
mongos> sh.shardCollection("testdb.peoples1",{"name":"hashed"}) { "collectionsharded" : "testdb.peoples1", "collectionUUID" : UUID("f6213da1-7c7d-4d5e-8fb1-fc554efb9df2"), "ok" : 1, "operationTime" : Timestamp(1605111014, 2), "$clusterTime" : { "clusterTime" : Timestamp(1605111014, 2), "signature" : { "hash" : BinData(0,"AAAAAAAAAAAAAAAAAAAAAAAAAAA="), "keyId" : NumberLong(0) } } } mongos> sh.status() --- Sharding Status --- sharding version: { "_id" : 1, "minCompatibleVersion" : 5, "currentVersion" : 6, "clusterId" : ObjectId("5fac01dd8d6fa3fe899662c8") } shards: { "_id" : "shard1_replset", "host" : "shard1_replset/node05:27018,node06:27018,node07:27018", "state" : 1 } { "_id" : "shard2_replset", "host" : "shard2_replset/node08:27018,node09:27018,node10:27018", "state" : 1 } active mongoses: "4.4.1" : 1 autosplit: Currently enabled: yes balancer: Currently enabled: yes Currently running: yes Collections with active migrations: config.system.sessions started at Thu Nov 12 2020 00:10:16 GMT+0800 (CST) Failed balancer rounds in last 5 attempts: 0 Migration Results for the last 24 hours: 480 : Success databases: { "_id" : "config", "primary" : "config", "partitioned" : true } config.system.sessions shard key: { "_id" : 1 } unique: false balancing: true chunks: shard1_replset 543 shard2_replset 481 too many chunks to print, use verbose if you want to force print { "_id" : "testdb", "primary" : "shard2_replset", "partitioned" : true, "version" : { "uuid" : UUID("454aad2e-b397-4c88-b5c4-c3b21d37e480"), "lastMod" : 1 } } testdb.peoples shard key: { "age" : 1 } unique: false balancing: true chunks: shard2_replset 1 { "age" : { "$minKey" : 1 } } -->> { "age" : { "$maxKey" : 1 } } on : shard2_replset Timestamp(1, 0) testdb.peoples1 shard key: { "name" : "hashed" } unique: false balancing: true chunks: shard1_replset 2 shard2_replset 2 { "name" : { "$minKey" : 1 } } -->> { "name" : NumberLong("-4611686018427387902") } on : shard1_replset Timestamp(1, 0) { "name" : NumberLong("-4611686018427387902") } -->> { "name" : NumberLong(0) } on : shard1_replset Timestamp(1, 1) { "name" : NumberLong(0) } -->> { "name" : NumberLong("4611686018427387902") } on : shard2_replset Timestamp(1, 2) { "name" : NumberLong("4611686018427387902") } -->> { "name" : { "$maxKey" : 1 } } on : shard2_replset Timestamp(1, 3) mongos>
提示:基于hash做分片只能在一个字段上做,不能指定多个字段;从上面的状态信息可以看到testdb.peoples被分到了shard2上,peoples1一部分分到了shard1,一部分分到了shard2上;所以在peoples中插入多少条数据,它都会写到shard2上,在peoples1中插入数据会被写入到shard1和shard2上;
验证:在peoples1 集合上插入数据,看看是否将数据分片到不同的shard上呢?
在mongos上插入数据
mongos> use testdb switched to db testdb mongos> for (i=1;i<=10000;i++) db.peoples1.insert({name:"people"+i,age:(i%120),classes:(i%20)}) WriteResult({ "nInserted" : 1 }) mongos>
在shard1上查看数据
shard1_replset:PRIMARY> show dbs admin 0.000GB config 0.001GB local 0.001GB testdb 0.000GB shard1_replset:PRIMARY> use testdb switched to db testdb shard1_replset:PRIMARY> show tables peoples1 shard1_replset:PRIMARY> db.peoples1.find().count() 4966 shard1_replset:PRIMARY>
提示:在shard1上可以看到对应collection保存了4966条数据;
在shard2上查看数据
shard2_replset:PRIMARY> show dbs admin 0.000GB config 0.001GB local 0.011GB testdb 0.011GB shard2_replset:PRIMARY> use testdb switched to db testdb shard2_replset:PRIMARY> show tables peoples peoples1 shard2_replset:PRIMARY> db.peoples1.find().count() 5034 shard2_replset:PRIMARY>
提示:在shard2上可以看到有peoples集合和peoples1集合,其中peoples1集合保存了5034条数据;shard1和shard2总共就保存了我们刚才插入的10000条数据;
到此到此mongodb的分片集群就搭建,测试完毕了。现在大家对MongoDB分片集群都有所了解了吧,希望文本对大家学习和理解MongoDB分片集群有帮助。
免责声明:本站发布的内容(图片、视频和文字)以原创、转载和分享为主,文章观点不代表本网站立场,如果涉及侵权请联系站长邮箱:mmqy2019@163.com进行举报,并提供相关证据,查实之后,将立刻删除涉嫌侵权内容。
长按识别二维码并关注微信
更方便到期提醒、手机管理