本文分两部分介绍mongodb3.2.1分片部署配置及故障模拟验证。
第一部分 安装配置
一、实验环境
两组副本集做分片
版本3.2.1
副本集1:192.168.115.11:27017,192.168.115.12:27017,192.168.115.11:47017(arbiter)
副本集2:192.168.115.11:37017,192.168.115.12:37017,192.168.115.12:47017(arbiter)
configserver:192.168.115.11:10000,192.168.115.11:10001,192.168.115.12:10000
mongos:192.168.115.11:20000
二、分片介绍
1.逻辑图
片(shard):每一个分片一个副本集
配置服务器(config server):存储集群的配置信息,3.2以上版本支持副本集模式部署
路由进程(mongos):路由所有请求,然后将结果聚合。它不保存存储数据或配置信息,配置信息从配置服务器上加载到内存中。
副本集方式部署confiserver
一、部署条件
1.集群中不能有仲裁节点
2.集群中不能有延迟节点
3.每个成员必须可以创建索引
二、configserver安装配置
1.修改配置文件(其他两个节点配置文件类似,主要修改监听端口,及数据路径,如果一台机器上运行多个实例,注意配置文件名称要不一样)
cat config.conf fork = true quiet = true port = 10000 dbpath = /data/config logpath = /usr/local/mongodb/logs/config.log logappend = true directoryperdb = true configsvr = true replSet = hnrconfig/192.168.115.11:10000,192.168.115.11:10001,192.168.115.12:10000
2.服务启动和停止
/usr/local/mongodb/bin/mongod -f /usr/local/mongodb/etc/config.conf
/usr/local/mongodb/bin/mongod --shutdown --port 10000 --dbpath=/data/config
3.配置副本集
连接任意一个节点进行配置
> show dbs
2016-11-17T09:06:08.088+0800 E QUERY [thread1] Error: listDatabases failed:{ "ok" : 0, "errmsg" : "not master and slaveOk=false", "code" : 13435 } :
_getErrorWithCode@src/mongo/shell/utils.js:23:13
Mongo.prototype.getDBs@src/mongo/shell/mongo.js:53:1
shellHelper.show@src/mongo/shell/utils.js:700:19
shellHelper@src/mongo/shell/utils.js:594:15
@(shellhelp2):1:1
出现以上错误,需要执行
> rs.slaveOk()
> use admin
> db.runCommand({"replSetInitiate" : { "_id" : "hnrconfig" ,"members" : [ { "_id" : 1, "host" : "192.168.115.11:10000"},{ "_id" : 2, "host" : "192.168.115.12:10000"},{"_id" : 3, "host" : "192.168.115.11:10001"}]}})
{ "ok" : 1 }
三、mongos配置
1.配置文件
cat mongos.conf fork = true quiet = true port = 20000 logpath = /usr/local/mongodb/logs/mongos.log logappend = true configdb = 192.168.115.11:10000,192.168.115.11:10001,192.168.115.12:10000
2.启动mongos服务
/usr/local/mongodb/bin/mongos -f /usr/local/mongodb/etc/mongos.conf
四、往集群中添加分片
连接mongos
mongos> sh.addShard("hnrtest1/192.168.115.11:27017")
{ "shardAdded" : "hnrtest1", "ok" : 1 }
mongos>
mongos> sh.addShard("hnrtest2/192.168.115.12:37017")
{ "shardAdded" : "hnrtest2", "ok" : 1 }
mongos>
五、开启分片
1.先对数据库启用分片功能
mongos> sh.enableSharding("shardtest")
{ "ok" : 1 }
mongos>
2.对集合开启分片(自动分片建)
mongos> sh.shardCollection("shardtest.student",{"cre_id":1})
{ "collectionsharded" : "shardtest.student", "ok" : 1 }
mongos>
3.修改默认chunk大小(默认为64M),自动分片测试效果不好,需要插入大量数据,将其修改为1M
mongos> use config
mongos> db.settings.save({ "_id" : "chunksize", "value" : NumberLong(1) })
修改后对student2集合进行分片
mongos> sh.shardCollection("shardtest.student2",{"cre_id":1})
插入5万条数据
直接在后端分片副本集上查询
hnrtest2:PRIMARY> db.student2.find().count()
27081
hnrtest2:PRIMARY>
hnrtest1:PRIMARY> db.student2.find().count()
22918
hnrtest1:PRIMARY>
4.采用哈希分片
修改chunk为默认值64M
mongos> db.settings.save({ "_id" : "chunksize", "value" : NumberLong(64) })
student3集合在cre_id字段使用哈希分片
mongos> sh.shardCollection("shardtest.student3",{"cre_id":"hashed"})
{ "collectionsharded" : "shardtest.student3", "ok" : 1 }
mongos> sh.status()
shardtest.student3
shard key: { "cre_id" : "hashed" }
unique: false
balancing: true
chunks:
hnrtest1 2
hnrtest2 2
{ "cre_id" : { "$minKey" : 1 } } -->> { "cre_id" : NumberLong("-4611686018427387902") } on : hnrtest1 Timestamp(2, 2)
{ "cre_id" : NumberLong("-4611686018427387902") } -->> { "cre_id" : NumberLong(0) } on : hnrtest1 Timestamp(2, 3)
{ "cre_id" : NumberLong(0) } -->> { "cre_id" : NumberLong("4611686018427387902") } on : hnrtest2 Timestamp(2, 4)
{ "cre_id" : NumberLong("4611686018427387902") } -->> { "cre_id" : { "$maxKey" : 1 } } on : hnrtest2 Timestamp(2, 5)
往student3插入1万条数据,在每个分片上查询
hnrtest1:PRIMARY> db.student3.find().count()
4952
hnrtest1:PRIMARY>
hnrtest2:PRIMARY> db.student3.find().count()
5047
hnrtest2:PRIMARY>
第二部分 故障模拟验证
一、模拟config服务副本集primary节点宕机
1.关闭服务
/usr/local/mongodb/bin/mongod --shutdown --port 10000 --dbpath=/data/config
2.副本集重新选举一个primary节点
3.读取数据,所有数据均正常返回
mongos> use shardtest
switched to db shardtest
mongos>
mongos> db.student.find().count()
99999
mongos> db.student2.find().count()
49999
mongos> db.student3.find().count()
9999
mongos>
4.对新的集合进行分片,插入5千条数据
mongos> sh.shardCollection("shardtest.student4",{"cre_id":"hashed"})
{ "collectionsharded" : "shardtest.student4", "ok" : 1 }
mongos>
在每个分片上查询数据
hnrtest2:PRIMARY> db.student4.find().count()
2525
hnrtest2:PRIMARY>
hnrtest1:PRIMARY> db.student4.find().count()
2474
hnrtest1:PRIMARY>
二、config服务数据备份恢复
1.数据备份
/usr/local/mongodb/bin/mongodump -h 192.168.115.11:10001 -o configdata
2.关闭所有config服务节点
/usr/local/mongodb/bin/mongod --shutdown --port 10000 --dbpath=/data/config
/usr/local/mongodb/bin/mongod --shutdown --port 10001 --dbpath=/data/config1
3.数据读取操作
由于mongos是将config的配置信息全部加载到内存中运行,因此此时通过mongos查询数据一切正常,但是不能对新的集合进行分片操作
mongos> db.student.find().count()
99999
mongos> db.student2.find().count()
49999
mongos> db.student3.find().count()
9999
mongos> db.student4.find().count()
4999
mongos>
4.对集合进行分片操作,无法完成
mongos> sh.shardCollection("shardtest.student5",{"cre_id":"hashed"})
{
"ok" : 0,
"errmsg" : "None of the hosts for replica set hnrconfig could be contacted.",
"code" : 71
}
mongos>
5.关闭mongos服务,删除config节点所有数据
6.重新启动三个config服务
7.重新初始化副本集
> rs.slaveOk()
> use admin
> db.runCommand({"replSetInitiate" : { "_id" : "hnrconfig" ,"members" : [ { "_id" : 1, "host" : "192.168.115.11:10000"},{ "_id" : 2, "host" : "192.168.115.12:10000"},{"_id" : 3, "host" : "192.168.115.11:10001"}]}})
8.启动mongos服务,此时没有任何数据
9.导入备份的config数据
/usr/local/mongodb/bin/mongorestore -h 192.168.115.11:10000 -d config configdata/config/
在mongos查询,但是查询数据会出现超时,数据无法查询
10.在mongos执行如下命令
mongos> sh.enableSharding("shardtest")
{ "ok" : 0, "errmsg" : "Operation timed out", "code" : 50 }
mongos日志
2016-11-17T14:46:21.197+0800 I SHARDING [Balancer] about to log metadata event into actionlog: { _id: "node1.hnr.com-2016-11-17T14:46:21.197+0800-582d523ded1c4b679a84877b", server: "node1.hnr.com", clientAddr: "", time: new Date(1479365181197), what: "balancer.round", ns: "", details: { executionTimeMillis: 30007, errorOccured: true, errmsg: "could not get updated shard list from config server due to ExceededTimeLimit Operation timed out" } }
官网上说是bug,恢复失败
https://jira.mongodb.org/browse/SERVER-22392