高可用的MongoDB集群-实战篇

1.概述

  最近有同学和网友私信我,问我MongoDB方面的问题;这里我整理一篇博客来赘述下MongoDB供大家学习参考,博客的目录内容如下:

  • 基本操作
  • CRUD
  • MapReduce

  本篇文章是基于MongoDB集群(Sharding+Replica Sets)上演示的,故操作的内容都是集群层面的,所以有些命令和单独的使用MongoDB库有异样。具体集群搭建可以参考我写的《高可用的MongoDB集群》。

2.基本操作

  常用的 Shell 命令如下所示:

db.help()    # 数据库帮助
db.collections.help()    # 集合帮助
rs.help()    # help on replica set
show dbs    # 展示数据库名
show collections    # 展示collections在当前库
use db_name    # 选择数据库

  查看集合基本信息,内容如下所示:

#查看帮助  
db.yourColl.help(); 

#查询当前集合的数据条数  
db.yourColl.count(); 

#查看数据空间大小 
db.userInfo.dataSize(); 

#得到当前聚集集合所在的
db db.userInfo.getDB(); 

#得到当前聚集的状态 
db.userInfo.stats(); 

#得到聚集集合总大小 
db.userInfo.totalSize(); 

#聚集集合储存空间大小 
db.userInfo.storageSize(); 

#Shard版本信息  
db.userInfo.getShardVersion() 

#聚集集合重命名,将userInfo重命名为users
db.userInfo.renameCollection("users"); 
 
#删除当前聚集集合 
db.userInfo.drop();

3.CRUD

3.1创建

  在集群中,我们增加一个 friends 库,命令如下所示:

db.runCommand({enablesharding:"friends"});

  在库新建后,我们在该库下创建一个user分片,命令如下:

db.runCommand( { shardcollection : "friends. user"});

3.2新增

  在MongoDB中,save和insert都能达到新增的效果。但是这两者是有区别的,在save函数中,如果原来的对象不存在,那他们都可以向collection里插入数据;如果已经存在,save会调用update更新里面的记录,而insert则会忽略操作。

  另外,在insert中可以一次性插叙一个列表,而不用遍历,效率高,save则需要遍历列表,一个个插入,下面我们可以看下两个函数的原型,通过函数原型我们可以看出,对于远程调用来说,是一次性将整个列表post过来让MongoDB去处理,效率会高些。

  Save函数原型如下所示:

技术分享

  Insert函数原型(部分代码)如下所示:

技术分享

3.3查询

3.3.1查询所有记录

db. user.find();

  默认每页显示20条记录,当显示不下的情况下,可以用it迭代命令查询下一页数据。注意:键入it命令不能带“;” 但是你可以设置每页显示数据的大小,用DBQuery.shellBatchSize= 50;这样每页就显示50条记录了。

3.3.2查询去掉后的当前聚集集合中的某列的重复数据

db. user.distinct("name"); 

#会过滤掉name中的相同数据 相当于:
select distict name from user;

3.3.3查询等于条件数据

db.user.find({"age": 24}); 
#相当于:
select * from user where age = 24;

3.3.4查询大于条件数据

db.user.find({age: {$gt: 24}}); 

# 相当于:
select * from user where age >24;

3.3.5查询小于条件数据

db.user.find({age: {$lt: 24}}); 
#相当于:
select * from user where age < 24;

3.3.6查询大于等于条件数据

db.user.find({age: {$gte: 24}}); 
#相当于:
select * from user where age >= 24;

3.3.7查询小于等于条件数据

db.user.find({age: {$lte: 24}}); 
#相当于:
select * from user where age <= 24;

3.3.8查询AND和OR条件数据

  • AND
db.user.find({age: {$gte: 23, $lte: 26}});

#相当于
select * from user where age >=23 and age <= 26;
  • OR

db.user.find({$or: [{age: 22}, {age: 25}]}); 

#相当于:
select * from user where age = 22 or age = 25;

 

3.3.9模糊查询

db.user.find({name: /mongo/}); 

#相当于%% 
select * from user where name like %mongo%;

3.3.10开头匹配

db.user.find({name: /^mongo/}); 
# 与SQL中得like语法类似
select * from user where name like mongo%;

3.3.11指定列查询

db.user.find({}, {name: 1, age: 1}); 

#相当于:
select name, age from user;

  当然name也可以用true或false,当用ture的情况下和name:1效果一样,如果用false就是排除name,显示name以外的列信息。

3.3.12指定列查询+条件查询

db.user.find({age: {$gt: 25}}, {name: 1, age: 1}); 

#相当于:
select name, age from user where age > 25;

 db.user.find({name: ‘zhangsan‘, age: 22});

 #相当于:

 select * from user where name = ‘zhangsan‘ and age = 22;

3.3.13排序

#升序:
db.user.find().sort({age: 1});
#降序:
db.
user.find().sort({age: -1});

3.3.14查询5条数据

db.user.find().limit(5); 

#相当于:
select * from user limit 5;

3.3.15N条以后数据

db.user.find().skip(10); 

#相当于:
select * from user where id not in ( select * from user limit 5 );

3.3.16在一定区域内查询记录

#查询在5~10之间的数据
db.user.find().limit(10).skip(5);

  可用于分页,limit是pageSize,skip是第几页*pageSize。

3.3.17COUNT

db.user.find({age: {$gte: 25}}).count(); 

#相当于:
select count(*) from user where age >= 20;

3.3.18安装结果集排序

db.userInfo.find({sex: {$exists: true}}).sort(); 

3.3.19不等于NULL

db.user.find({sex: {$ne: null}}) 

#相当于:
select * from user where sex not null;

3.4索引

  创建索引,并指定主键字段,命令内容如下所示:

db.epd_favorites_folder.ensureIndex({"id":1},{"unique":true,"dropDups":true})
db.epd_focus.ensureIndex({"id":1},{"unique":true,"dropDups":true})

3.5更新

  update命令格式,如下所示:

db.collection.update(criteria,objNew,upsert,multi) 

  参数说明: criteria:

  查询条件 objNew:update对象和一些更新操作符

  upsert:如果不存在update的记录,是否插入objNew这个新的文档,true为插入,默认为false,不插入。

  multi:默认是false,只更新找到的第一条记录。如果为true,把按条件查询出来的记录全部更新。

  下面给出一个示例,更新id为 1 中 price 的值,内容如下所示:

db. user.update({id: 1},{$set:{price:2}});  

#相当于:
update user set price=2 where id=1;

3.6删除

3.6.1删除指定记录

db. user. remove( { id:1 } );  

#相当于:
delete from user where id=1;

3.6.2删除所有记录

db. user. remove( { } );  

#相当于:
delete from user;

3.6.3DROP

db. user. drop();  

#相当于:
drop table user;

4.MapReduce

  MongoDB中的 MapReduce 是编写JavaScript脚本,然后由MongoDB去解析执行对应的脚本,下面给出 Java API 操作MR。代码如下所示:

  MongdbManager类,用来初始化MongoDB:

package cn.mongo.util;

import org.slf4j.Logger;
import org.slf4j.LoggerFactory;

import com.mongodb.DB;
import com.mongodb.Mongo;
import com.mongodb.MongoOptions;

/**
 * @Date Mar 3, 2015
 * 
 * @author dengjie
 * 
 * @Note mongodb manager
 */
public class MongdbManager {

    private static final Logger logger = LoggerFactory.getLogger(MongdbManager.class);
    private static Mongo mongo = null;
    private static String tag = SystemConfig.getProperty("dev.tag");

    private MongdbManager() {
    }

    static {
        initClient();
    }

    // get DB object
    public static DB getDB(String dbName) {
        return mongo.getDB(dbName);
    }

    // get DB object without param
    public static DB getDB() {
        String dbName = SystemConfig.getProperty(String.format("%s.mongodb.dbname", tag));
        return mongo.getDB(dbName);
    }

    // init mongodb pool
    private static void initClient() {
        try {
            String[] hosts = SystemConfig.getProperty(String.format("%s.mongodb.host", tag)).split(",");
            for (int i = 0; i < hosts.length; i++) {
                try {
                    String host = hosts[i].split(":")[0];
                    int port = Integer.parseInt(hosts[i].split(":")[1]);
                    mongo = new Mongo(host, port);
                    if (mongo.getDatabaseNames().size() > 0) {
                        logger.info(String.format("connection success,host=[%s],port=[%d]", host, port));
                        break;
                    }
                } catch (Exception ex) {
                    ex.printStackTrace();
                    logger.error(String.format("create connection has error,msg is %s", ex.getMessage()));
                }
            }

            // 设置连接池的信息
            MongoOptions opt = mongo.getMongoOptions();
            opt.connectionsPerHost = SystemConfig.getIntProperty(String.format("%s.mongodb.poolsize", tag));// poolsize
            opt.threadsAllowedToBlockForConnectionMultiplier = SystemConfig.getIntProperty(String.format(
                    "%s.mongodb.blocksize", tag));// blocksize
            opt.socketKeepAlive = true;
            opt.autoConnectRetry = true;
        } catch (Exception e) {
            e.printStackTrace();
        }
    }
}

  MongoDBFactory类,用来封装操作业务代码,具体内容如下所示:

package cn.mongo.util;

import java.util.ArrayList;
import java.util.List;

import org.slf4j.Logger;
import org.slf4j.LoggerFactory;

import cn.diexun.domain.MGDCustomerSchema;

import com.mongodb.BasicDBList;
import com.mongodb.DB;
import com.mongodb.DBCollection;
import com.mongodb.DBObject;
import com.mongodb.util.JSON;

/**
 * @Date Mar 3, 2015
 *
 * @Author dengjie
 */
public class MongoDBFactory {

    private static Logger logger = LoggerFactory.getLogger(MongoDBFactory.class);

    // save data to mongodb
    public static void save(MGDCustomerSchema mgs, String collName) {
        DB db = null;
        try {
            db = MongdbManager.getDB();
            DBCollection coll = db.getCollection(collName);
            DBObject dbo = (DBObject) JSON.parse(mgs.toString());
            coll.insert(dbo);
        } catch (Exception ex) {
            ex.printStackTrace();
            logger.error(String.format("save object to mongodb has error,msg is %s", ex.getMessage()));
        } finally {
            if (db != null) {
                db.requestDone();
                db = null;
            }
        }
    }

    // batch insert
    public static void save(List<?> mgsList, String collName) {
        DB db = null;
        try {
            db = MongdbManager.getDB();
            DBCollection coll = db.getCollection(collName);
            BasicDBList data = (BasicDBList) JSON.parse(mgsList.toString());
            List<DBObject> list = new ArrayList<DBObject>();
            int commitSize = SystemConfig.getIntProperty("mongo.commit.size");
            int rowCount = 0;
            long start = System.currentTimeMillis();
            for (Object dbo : data) {
                rowCount++;
                list.add((DBObject) dbo);
                if (rowCount % commitSize == 0) {
                    try {
                        coll.insert(list);
                        list.clear();
                        logger.info(String.format("current commit rowCount = [%d],commit spent time = [%s]s", rowCount,
                                (System.currentTimeMillis() - start) / 1000.0));
                    } catch (Exception ex) {
                        ex.printStackTrace();
                        logger.error(String.format("batch commit data to mongodb has error,msg is %s", ex.getMessage()));
                    }
                }
            }
            if (rowCount % commitSize != 0) {
                try {
                    coll.insert(list);
                    logger.info(String.format("insert data to mongo has spent total time = [%s]s",
                            (System.currentTimeMillis() - start) / 1000.0));
                } catch (Exception ex) {
                    ex.printStackTrace();
                    logger.error(String.format("commit end has error,msg is %s", ex.getMessage()));
                }
            }
        } catch (Exception ex) {
            ex.printStackTrace();
            logger.error(String.format("save object list to mongodb has error,msg is %s", ex.getMessage()));
        } finally {
            if (db != null) {
                db.requestDone();
                db = null;
            }
        }
    }
}

  LoginerAmountMR类,这是一个统计登录用户数的MapReduce计算类,代码如下:

package cn.mongo.mapreduce;

import java.sql.Timestamp;
import java.util.ArrayList;
import java.util.Date;
import java.util.List;

import org.bson.BSONObject;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;

import cn.diexun.conf.ConfigureAPI.MR;
import cn.diexun.conf.ConfigureAPI.PRECISION;
import cn.diexun.domain.Kpi;
import cn.diexun.util.CalendarUtil;
import cn.diexun.util.MongdbManager;
import cn.diexun.util.MysqlFactory;

import com.mongodb.DB;
import com.mongodb.DBCollection;
import com.mongodb.DBCursor;
import com.mongodb.DBObject;
import com.mongodb.MapReduceOutput;
import com.mongodb.ReadPreference;

/**
 * @Date Mar 13, 2015
 * 
 * @Author dengjie
 * 
 * @Note use mr jobs stats user login amount
 */
public class LoginerAmountMR {
    private static Logger logger = LoggerFactory.getLogger(LoginerAmountMR.class);

   // map 函数JS字符串拼接
private static String map() { String map = "function(){"; map += "if(this.userName != \"\"){"; map += "emit({" + "kpi_code:‘login_times‘,username:this.userName," + "district_id:this.districtId,product_style:this.product_style," + "customer_property:this.customer_property},{count:1});"; map += "}"; map += "}"; return map; }
  
private static String reduce() { String reduce = "function(key,values){"; reduce += "var total = 0;"; reduce += "for(var i=0;i<values.length;i++){"; reduce += "total += values[i].count;}"; reduce += "return {count:total};"; reduce += "}"; return reduce; }
  // reduce 函数字符串拼接
public static void main(String[] args) { loginNumbers("t_login_20150312"); } /** * login user amount * * @param collName */ public static void loginNumbers(String collName) { DB db = null; try { db = MongdbManager.getDB(); db.setReadPreference(ReadPreference.secondaryPreferred()); DBCollection coll = db.getCollection(collName); String result = MR.COLLNAME_TMP; long start = System.currentTimeMillis(); MapReduceOutput mapRed = coll.mapReduce(map(), reduce(), result, null); logger.info(String.format("mr run spent time=%ss", (System.currentTimeMillis() - start) / 1000.0)); start = System.currentTimeMillis(); DBCursor cursor = mapRed.getOutputCollection().find(); List<Kpi> list = new ArrayList<Kpi>(); while (cursor.hasNext()) { DBObject obj = cursor.next(); BSONObject key = (BSONObject) obj.get("_id"); BSONObject value = (BSONObject) obj.get("value"); Object kpiValue = value.get("count"); Object userName = key.get("username"); Object districtId = key.get("district_id"); Object customerProperty = key.get("customer_property"); Object productStyle = key.get("product_style"); Kpi kpi = new Kpi(); try { kpi.setUserName(userName == null ? "" : userName.toString()); kpi.setKpiCode(key.get("kpi_code").toString()); kpi.setKpiValue(Math.round(Double.parseDouble(kpiValue.toString()))); kpi.setCustomerProperty(customerProperty == null ? "" : customerProperty.toString()); kpi.setDistrictId(districtId == "" ? 0 : Integer.parseInt(districtId.toString())); kpi.setProductStyle(productStyle == null ? "" : productStyle.toString()); kpi.setCreateDate(collName.split("_")[2]); kpi.setUpdateDate(Timestamp.valueOf(CalendarUtil.formatMap.get(PRECISION.HOUR).format(new Date()))); list.add(kpi); } catch (Exception exx) { exx.printStackTrace(); logger.error(String.format("parse type or get value has error,msg is %s", exx.getMessage())); } } MysqlFactory.insert(list); logger.info(String.format("store mysql spent time is %ss", (System.currentTimeMillis() - start) / 1000.0)); } catch (Exception ex) { ex.printStackTrace(); logger.error(String.format("run map-reduce jobs has error,msg is %s", ex.getMessage())); } finally { if (db != null) { db.requestDone(); db = null; } } } }

5.总结

  在计算 MongoDB 的MapReduce计算的时候,拼接JavaScript字符串时需要谨慎小心,很容易出错,上面给出的代码只是一部分代码,供参考学习使用;另外,若是要做MapReduce任务计算,推荐使用Hadoop的MapReduce计算框架,MongoDB的MapReduce框架这里仅做介绍学习了解。

6.结束语

  这篇博客就和大家分享到这里,若是大家在研究学习的过程当中有什么问题,可以加群进行讨论或发送邮件给我,我会尽我所能为您解答,与君共勉!

郑重声明:本站内容如果来自互联网及其他传播媒体,其版权均属原媒体及文章作者所有。转载目的在于传递更多信息及用于网络分享,并不代表本站赞同其观点和对其真实性负责,也不构成任何其他建议。