MongoDB分布式存储的MapReduce并行查询

今天要介绍的是如何使用MONGODB中提供的MapReduce功能进行查询。

  今天介绍如何基于sharding机制进行mapreduce查询。在MongoDB的官方文档中,这么一句话:

Sharded Environments
      In sharded environments, data processing of map/reduce operations runs in parallel on all shards.
即: map/reduce操作会并行运行在所有的shards上。

  下面我们就用之前这篇文章中白搭建的环境来构造mapreduce查询:

  首先要说的是,基于sharding的mapreduce与非sharding的数据在返回结构上有一些区别,我目前注意到的主要是不支持定制式的json格式的返回数据,也就是下面方式可能会出现问题:

  return { count : total };
注意:上面的情况目前出现在了我的测试环境下,如下图:

 

  就需要改成 return count;

  下面是测试代码,首先是按帖子id来查询相应数量(基于分组查询实例方式):

Code highlighting produced by Actipro CodeHighlighter (freeware)http://www.CodeHighlighter.com/-->public partial class getfile : System.Web.UI.Page
    {

        public Mongo Mongo { get; set; }

        public IMongoDatabase DB
        {
            get
            {
                return this.Mongo["dnt_mongodb"];
            }
        }

        /// <summary>
        /// Sets up the test environment.  You can either override this OnInit to add custom initialization.
        /// </summary>
        public virtual void Init()
        {
            string ConnectionString = "Server=10.0.4.85:27017;ConnectTimeout=30000;ConnectionLifetime=300000;MinimumPoolSize=512;MaximumPoolSize=51200;Pooled=true";
            if (String.IsNullOrEmpty(ConnectionString))
                throw new ArgumentNullException("Connection string not found.");
            this.Mongo = new Mongo(ConnectionString);
            this.Mongo.Connect();         
        }
        string mapfunction = "function(){\n" +
                        "  if(this._id=='548111') { emit(this._id, 1); } \n" +   
                        "};";

        string reducefunction = "function(key, current ){" +
                                "   var count = 0;" +
                                "   for(var i in current) {" +
                                "       count+=current[i];" +
                                "   }" +
                                "   return count ;\n" +
                              "};";

      
        protected void Page_Load(object sender, EventArgs e)
        {
            Init();

            var mrb = DB["posts1"].MapReduce();//attach_gfstream.files
            int groupCount = 0;
            using (var mr = mrb.Map(mapfunction).Reduce(reducefunction))
            {
                foreach (Document doc in mr.Documents)
                {
                    groupCount = int.Parse(doc["value"].ToString());
                }
            }
            this.Mongo.Disconnect();
        }     
     }

  下面是运行时的查询结果,如下:

MongoDB分布式存储的MapReduce并行查询

  接着演示一下如何把查询到的帖子信息返回并装入list集合,这里只查询ID为548110和548111两个帖子:

Code highlighting produced by Actipro CodeHighlighter (freeware)http://www.CodeHighlighter.com/-->        string mapfunction = "function(){\n" +
                        "  if(this._id=='548110'|| this._id=='548111') { emit(this, 1); } \n" +    
                        "};";

        string reducefunction = "function(doc, current ){" +
                                "   return doc;\n" +
                               "};";
      
        protected void Page_Load(object sender, EventArgs e)
        {
            Init();

            var mrb = DB["posts1"].MapReduce();//attach_gfstream.files
            List<Document> postDoc = new List<Document>();
            using (var mr = mrb.Map(mapfunction).Reduce(reducefunction))
            {
                foreach (Document doc in mr.Documents)
                {
                    postDoc.Add((Document)doc["value"]);
                }
            }
            this.Mongo.Disconnect();
        }
下面是运行时的查询结果,如下:

 

  上面的map/reduce方法还有许多写法,如果大家感兴趣可以看一下如下这些链接:

  http://cookbook.mongodb.org/patterns/unique_items_map_reduce/

  http://www.mongodb.org/display/DOCS/MapReduce

  以及之前我写的这篇文章:http://www.cnblogs.com/daizhj/archive/2010/06/10/1755761.html

  当然在mongos进行map/reduce运算时,会生成一些临时文件,如下图:

MongoDB分布式存储的MapReduce并行查询

  我猜这些临时文件可能会对再次查询系统时的性能有一些提升(但目前未观察到)。

  当然对于mongodb的gridfs系统(可使用它搭建分布式文件存储系统,我之前在这篇文章中已介绍过,我也做了测试,但遗憾的是并未成功,它经常会报一些错误,比如:

  Thu Sep 09 12:09:29 Assertion failure _grab client\parallel.cpp 461
看来mapreduce程序链接到mongodb上时,会产生一些问题,但不知道是不是其自身稳定性的原因,还是我的机器环境设置问题(内存或配置的64位系统mongos与32位的client连接问题)。








本文转自 wws5201985 51CTO博客,原文链接:http://blog.51cto.com/wws5201985/786458,如需转载请自行联系原作者