淺析MongoDB數據庫分布式存儲物理文件
作者：代震軍
今天我們將談到的是基于MongoDB數據庫的分布式存儲物理文件,包括測試的過程，希望本文能對大家有所幫助。
我們知道了如何對關系型數據數據通過auto-sharding進行分布式數據存儲，今天介紹如何對物理文件（小文件，基本小于100K）進行分布式存儲。
接著看一下要配置的測試環境(與前一篇中類似)：
模擬2個shard服務和一個config服務, 均運行在10.0.4.85機器上，只是端口不同：
Shard1：27020
Shard2：27021
Config：27022
Mongos啟動時默認使用的27017端口
在C,D,E磁盤下分別建立如下文件夾：
mongodb\bin
mongodb\db
然后用CMD命令行依次打開相應文件夾下的mongd文件:
c:\mongodb\bin\mongod --dbpath c:\mongodb\db\ --port 27020
d:\mongodb\bin\mongod --dbpath d:\mongodb\db\ --port 27021
e:\mongodb\bin\mongod --configsvr --dbpath e:\mongodb\db\ --port 27022 (注:config配置服務器)
啟動mongos時，默認開啟了27017端口
e:\mongodb\bin\mongos --configdb 10.0.4.85:27022
然后打開mongo:
E:\mongodb\bin>mongo 回車 （有時加端口會造成下面的addshard命令出問題）
- > use admin
- switched to db admin
- > db.runCommand( { addshard : "10.0.4.85:27020", allowLocal : 1, maxSize:2 , minKey:1, maxKey:10} )
- --添加sharding,maxsize單位是M,此處設置比較小的數值只為演示sharding效果
- { "shardAdded" : "shard0000", "ok" : 1 }
- > db.runCommand( { addshard : "10.0.4.85:27021", allowLocal : 1, minKey:1000} )
- { "shardAdded" : "shard0001", "ok" : 1 }
注：如果要移除sharding，可用下面寫法
- db.runCommand( { removeshard : "localhost:10000" } );
- > db.runCommand({listshards:1}); --查看shard節點列表
- > config = connect("10.0.4.85:27022")
- > configconfig = config.getSisterDB("config")
- > dnt_mongodb=db.getSisterDB("dnt_mongodb");
- dnt_mongodb
- > db.runCommand({enablesharding:"dnt_mongodb"})
- { "ok" : 1 }
- > db.printShardingStatus()
- --- Sharding Status ---
- sharding version: { "_id" : 1, "version" : 3 }
- shards:
- {
- "_id" : "shard0000",
- "host" : "10.0.4.85:27020",
- "maxSize" : NumberLong( 2 )
- }
- { "_id" : "shard0001", "host" : "10.0.4.85:27021" }
- databases:
- { "_id" : "admin", "partitioned" : false, "primary" : "config" }
- { "_id" : "dnt_mongodb", "partitioned" : true, "primary" : "shard0001" }
- > db.runCommand( { shardcollection : "dnt_mongodb.attach_gfstream.chunks", key : { files_id : 1 } } )
- --此處與之前的數據存儲方式有些不同,目前shard似乎僅支持files_id
- { "collectionsharded" : "dnt_mongodb.attach_gfstream.chunks", "ok" : 1 }
注：運行上面命令之前需要設置files_id為唯一索引[unique index]。
創建完sharding和設置相應信息后，我們加載一下測試數據，我用下面代碼來讀取要本地文件，然后批量向mongodb中添加（通過循環修改文件名來添加相同大小的文件）。
- /// <summary>
- /// 上傳文件到mongodb
- /// </summary>
- /// <param name="uploadDir">要上傳文件所在路徑</param>
- /// <param name="fileName">要上傳的文件名</param>
- /// <returns></returns>
- public bool UploadFile(string uploadDir, string fileName)
- {
- for (int i = 1; i < 10000; i++)
- {
- try
- {
- Mongo mongo = mongoDB;
- mongo.Connect();
- IMongoDatabase DB = mongo["dnt_mongodb"];
- using (FileStream fileStream = new FileStream(uploadDir + fileName, FileMode.Open))
- {
- int nFileLen = (int)fileStream.Length;
- byte[] myData = new Byte[nFileLen];
- fileStream.Read(myData, 0, nFileLen);
- GridFile fs = new GridFile(DB, "attach_gfstream");
- using (GridFileStream gfs = fs.Create(fileName + i))
- {
- gfs.Write(myData, 0, nFileLen);
- }
- }
- mongo.Disconnect();
- }
- catch { }
- }
- return true;
- }
在批量添加約10000次(約10000個文件)之后，mongodb開始把sharding出來的chunk從shard0000分布到shard0001上，我們可以用下面指令來進行驗證：
> db.printShardingStatus()
- --- Sharding Status ---
- sharding version: { "_id" : 1, "version" : 3 }
- shards:
- {
- "_id" : "shard0000",
- "host" : "10.0.4.85:27020",
- "maxSize" : NumberLong( 2 )
- }
- { "_id" : "shard0001", "host" : "10.0.4.85:27021" }
- databases:
- { "_id" : "admin", "partitioned" : false, "primary" : "config" }
- { "_id" : "dnt_mongodb", "partitioned" : true, "primary" : "shard0000" }
- dnt_mongodb.attach_gfstream.chunks chunks:
- { "files_id" : { $minKey : 1 } } -->> { "files_id" : ObjectId("4c85fd02145a9b1534010d89") } on : shard0001 { "t" : 2000, "i" : 0 }
- { "files_id" : ObjectId("4c85fd02145a9b1534010d89") } -->> { "files_id" : ObjectId("4c85fdec145a9b0b340005a7") } on : shard0000 { "t" :3000, "i" : 1 }
- { "files_id" : ObjectId("4c85fdec145a9b0b340005a7") } -->> { "files_id" : ObjectId("4c85fe08145a9b0b34000aaf") } on : shard0001 { "t" :3000, "i" : 4 }
- { "files_id" : ObjectId("4c85fe08145a9b0b34000aaf") } -->> { "files_id" : ObjectId("4c85fe27145a9b0b34000fb7") } on : shard0001 { "t" :4000, "i" : 1 }
- { "files_id" : ObjectId("4c85fe27145a9b0b34000fb7") } -->> { "files_id" : ObjectId("4c85fe43145a9b0b340014bf") } on : shard0000 { "t" :4000, "i" : 7 }
- { "files_id" : ObjectId("4c85fe43145a9b0b340014bf") } -->> { "files_id" : ObjectId("4c85fe61145a9b0b340019c7") } on : shard0000 { "t" :4000, "i" : 8 }
- { "files_id" : ObjectId("4c85fe61145a9b0b340019c7") } -->> { "files_id" : ObjectId("4c85fe7b145a9b0b34001ecf") } on : shard0000 { "t" :5000, "i" : 1 }
- { "files_id" : ObjectId("4c85fe7b145a9b0b34001ecf") } -->> { "files_id" : ObjectId("4c85fe9a145a9b0b340023d7") } on : shard0001 { "t" :5000, "i" : 4 }
- { "files_id" : ObjectId("4c85fe9a145a9b0b340023d7") } -->> { "files_id" : ObjectId("4c85feb7145a9b0b340028df") } on : shard0001 { "t" :6000, "i" : 1 }
- { "files_id" : ObjectId("4c85feb7145a9b0b340028df") } -->> { "files_id" : ObjectId("4c85feea145a9b0b340032ef") } on : shard0000 { "t" :6000, "i" : 4 }
- { "files_id" : ObjectId("4c85feea145a9b0b340032ef") } -->> { "files_id" : ObjectId("4c85ff25145a9b0b34003cff") } on : shard0000 { "t" :7000, "i" : 1 }
- { "files_id" : ObjectId("4c85ff25145a9b0b34003cff") } -->> { "files_id" : ObjectId("4c85ff57145a9b0b3400470f") } on : shard0001 { "t" :7000, "i" : 4 }
- { "files_id" : ObjectId("4c85ff57145a9b0b3400470f") } -->> { "files_id" : ObjectId("4c85ff87145a9b0b3400511f") } on : shard0001 { "t" :8000, "i" : 1 }
- { "files_id" : ObjectId("4c85ff87145a9b0b3400511f") } -->> { "files_id" : ObjectId("4c85ffcd145a9b0b34005b2f") } on : shard0000 { "t" :8000, "i" : 16 }
- { "files_id" : ObjectId("4c85ffcd145a9b0b34005b2f") } -->> { "files_id" : ObjectId("4c85fff7145a9b0b3400653f") } on : shard0000 { "t" :8000, "i" : 17 }
- { "files_id" : ObjectId("4c85fff7145a9b0b3400653f") } -->> { "files_id" : ObjectId("4c860021145a9b0b34006f4f") } on : shard0000 { "t" :8000, "i" : 18 }
- { "files_id" : ObjectId("4c860021145a9b0b34006f4f") } -->> { "files_id" : ObjectId("4c86004f145a9b0b3400795f") } on : shard0000 { "t" :8000, "i" : 19 }
- { "files_id" : ObjectId("4c86004f145a9b0b3400795f") } -->> { "files_id" : ObjectId("4c860080145a9b0b3400836f") } on : shard0000 { "t" :9000, "i" : 1 }
- { "files_id" : ObjectId("4c860080145a9b0b3400836f") } -->> { "files_id" : ObjectId("4c8600b5145a9b0b34008d7f") } on : shard0001 { "t" :9000, "i" : 7 }
- { "files_id" : ObjectId("4c8600b5145a9b0b34008d7f") } -->> { "files_id" : ObjectId("4c860115145a9b0b3400a183") } on : shard0001 { "t" :9000, "i" : 8 }
- { "files_id" : ObjectId("4c860115145a9b0b3400a183") } -->> { "files_id" : ObjectId("4c860198145a9b0b3400b587") } on : shard0001 { "t" :10000, "i" : 1 }
- { "files_id" : ObjectId("4c860198145a9b0b3400b587") } -->> { "files_id" : ObjectId("4c8601fc145a9b0b3400c98b") } on : shard0000 { "t" :10000, "i" : 11 }
- { "files_id" : ObjectId("4c8601fc145a9b0b3400c98b") } -->> { "files_id" : ObjectId("4c86025b145a9b0b3400dd8f") } on : shard0000 { "t" :10000, "i" : 12 }
- { "files_id" : ObjectId("4c86025b145a9b0b3400dd8f") } -->> { "files_id" : ObjectId("4c8602ca145a9b0b3400f193") } on : shard0000 { "t" :10000, "i" : 13 }
- { "files_id" : ObjectId("4c8602ca145a9b0b3400f193") } -->> { "files_id" : ObjectId("4c860330145a9b0b34010597") } on : shard0000 { "t" :10000, "i" : 14 }
- { "files_id" : ObjectId("4c860330145a9b0b34010597") } -->> { "files_id" : { $maxKey : 1 } } on : shard0000 { "t" : 10000, "i" : 15 }
當前，綜合比較，發現還是chunks的值要遠大于files集合所占用的磁盤空間（前者存儲文件二進制流信息，后者存儲結構化數據信息（如文件名稱大小等）：
下面是一個測試，用于讀寫shard0001（注意不是shard0000）上的圖片數據，因為mongos可以很好的管理sharding下各分區下的數據chunk，所以我們只要告訴它要取的文件名稱即可:)
比如要獲取"2010\09\07\2\2856090617370.gif6243"這個文件(帶日期型文件路徑只是一種格式，因為我們的產品會將上傳的附件放到相應磁盤目錄下，這種帶路徑的命名方式會方便與磁盤路徑進行對應)，其目前位于shard0001中，我們只要通過下面html代碼即可獲取圖文件信息：
- <img src="getfile.aspx?filename=2010\09\07\2\2856090617370.gif6243" width="30" />
相應的getfile.aspx.cs 代碼參見如下：
- 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();
- }
- protected void Page_Load(object sender, EventArgs e)
- {
- if (!string.IsNullOrEmpty(Request.QueryString["filename"]))
- {
- string filename = Request.QueryString["filename"];
- Init();
- String filesystem = "attach_gfstream";
- GridFile fs = new GridFile(DB, filesystem);
- GridFileStream gfs = fs.OpenRead(filename);
- Byte[] buffer = new Byte[gfs.Length];
- //下面的Expires和Cache-Control設置主要用于squid反向加速
- 更多內容參見 http://www.cnblogs.com/daizhj/archive/2010/08/19/1803454.html
- HttpContext.Current.Response.AddHeader("Expires", DateTime.Now.AddDays(20).ToString("r"));
- HttpContext.Current.Response.AddHeader("Cache-Control", "public");
- // 需要讀的數據長度
- long dataToRead = gfs.Length;
- int length;
- while (dataToRead > 0)
- {
- // 檢查客戶端是否還處于連接狀態
- if (HttpContext.Current.Response.IsClientConnected)
- {
- length = gfs.Read(buffer, 0, 10000);
- HttpContext.Current.Response.OutputStream.Write(buffer, 0, length);
- HttpContext.Current.Response.Flush();
- buffer = new Byte[10000];
- dataToRead = dataToRead - length;
- }
- else
- {
- // 如果不再連接則跳出死循環
- dataToRead = -1;
- }
- }
- gfs.Dispose();
- this.Mongo.Disconnect();
- HttpContext.Current.Response.End();
- }
- }
- }
當然，上面只是對chunks進行sharding，如果要對files集合分片時，可以用下面命令行：
- > db.runCommand( { shardcollection : "dnt_mongodb.attach_gfstream.files", key : { _id : 1 } } )
- { "collectionsharded" : "dnt_mongodb.attach_gfstream.files", "ok" : 1 }
在我添加了近50萬記錄后，mongos開始將新的文件信息保存到shard0001上，如下圖：
可以使用如下命令行來查看 shard上的信息：
- > db.printShardingStatus()
- .../省略之前files_id的shard信息
- { "filename" : { $minKey : 1 } } -->> { "filename" : "2010\\09\\08\\2\\1393993713076.gif1" } on : shard0000 { "t" : 1000, "i" : 6 }
- { "filename" : "2010\\09\\08\\2\\1393993713076.gif1" } -->> { "filename" : "2010\\09\\08\\2\\2396571814760.gif9999" } on : shard0000 { "t" : 1000, "i" : 7 }
- { "filename" : "2010\\09\\08\\2\\2396571814760.gif9999"} -->> { "filename" : "2010\\09\\08\\2\\2819270318096.gif25366" } on : shard0000 { "t" : 2000, "i" : 2 }
- { "filename" : "2010\\09\\08\\2\\2819270318096.gif25366" } -->> { "filename" : "2010\\09\\08\\2\\3100748419355.gif999" } on : shard0000{ "t" : 2000, "i" : 3 }
- { "filename" : "2010\\09\\08\\2\\3100748419355.gif999" } -->> { "filename" : { $maxKey : 1 } } on : shard0001 { "t" : 2000, "i" : 0 }
下面是mongos上進行sharding時的信息：
- Wed Sep 08 17:25:44 [conn5] ns: dnt_mongodb.attach_gfstream.files ClusteredCursor::query ShardConnection had to change attempt: 0
- Wed Sep 08 17:32:34 [conn6] ns: dnt_mongodb.attach_gfstream.files ClusteredCursor::query ShardConnection had to change attempt: 0
- Wed Sep 08 17:38:49 [conn55] autosplitting dnt_mongodb.attach_gfstream.chunks size: 188884488 shard: ns:dnt_mongodb.attach_gfstream.chunks at: shard0001:10.0.4.85:27021 lastmod: 11|3 min: { files_id: ObjectId('4c8755b3145a9b16d41d5dc9') } m
- ax: { files_id: MaxKey } on: { files_id: ObjectId('4c8759a5145a9b16d42300d7') }(splitThreshold 188743680)
- Wed Sep 08 17:38:49 [conn55] config change: { _id: "4_85-2010-09-08T09:38:49-10", server: "4_85", time: new Date(1283938729648), what: "split", ns: "dnt_mongodb.attach_gfstream.chunks", details: { before: { min: { files_id: ObjectId('4c8755
- b3145a9b16d41d5dc9') }, max: { files_id: MaxKey } }, left: { min: { files_id: ObjectId('4c8755b3145a9b16d41d5dc9') }, max: { files_id: ObjectId('4c8759a5145a9b16d42300d7') } }, right: { min: { files_id: ObjectId('4c8759a5145a9b16d42300d7')
- }, max: { files_id: MaxKey } } } }
- Wed Sep 08 17:38:49 [conn98] ns: dnt_mongodb.attach_gfstream.chunks ClusteredCursor::query ShardConnection had to change attempt: 0
好了，今天的文章就先到這里了。
原文標題：基于Mongodb分布式存儲物理文件
鏈接：http://www.cnblogs.com/daizhj/archive/2010/09/08/1821481.html
【編輯推薦】
責任編輯：彭凡
來源：
博客園
