其实sharding也不是什么很新的数据库功能，更不是mongodb独有的，类似的oracle中的功能叫做分区表（partition table），目前多数互联网公司的数据库应该都使用了类似的技术，口头交流中也有人喜欢叫“分库分表”。其功能是，当一个集合（或者说oracle中的表）变的很大之后，可以把这个集合的数据分到2个数据库（即shards）中，这样不会导致数据库变得大到难以维护，同时自动实现读写和负载均衡，这样在集合持续变大的过程中，只需要增大数据库的个数即可。
mongodb的sharding需要好几个服务，配置服务器（config server），路由进程（route process），片数据库（shards）。config server用来保存sharding的配置信息，在整个sharding中是单点，一般会有多个备份；route process，mongodb中叫mongos，负责与client连接并将数据分布到各个shards上；shards就是存储分片之后的数据。当然，从数据的安全和运行的业务连续性角度来看，shards和route process也应该有多个备份。
实验中，使用1个config server，1个mongos，2个shards，为了启动进程加快，试验中基本都使用了--nojournal选项。
首先建立好存储数据文件的目录，并启动config server和mongos，分别监听10000和10100端口
XXXXX@XXXXX-asus:/var/lib$ sudo mkdir -p mongodb_sharding/config XXXXX@XXXXX-asus:/var/lib$ sudo mkdir -p mongodb_sharding/shard1 XXXXX@XXXXX-asus:/var/lib$ sudo mkdir -p mongodb_sharding/shard2 XXXXX@XXXXX-asus:/var/lib$ sudo mkdir -p mongodb_sharding/config_log XXXXX@XXXXX-asus:/var/lib$ sudo mkdir -p mongodb_sharding/shard1_log XXXXX@XXXXX-asus:/var/lib$ sudo mkdir -p mongodb_sharding/shard2_log XXXXX@XXXXX-asus:/var/lib$ sudo mongod --dbpath /var/lib/mongodb_sharding/config --logpath /var/lib/mongodb_sharding/config_log/mongodb.log --port 10000 --nojournal --fork forked process: 8430 all output going to: /var/lib/mongodb_sharding/config_log/mongodb.log log file [/var/lib/mongodb_sharding/config_log/mongodb.log] exists; copied to temporary file [/var/lib/mongodb_sharding/config_log/mongodb.log.2014-03-21T08-03-50] child process started successfully, parent exiting XXXXX@XXXXX-asus:/var/lib$ sudo mkdir -p mongodb_sharding/mongos_log XXXXX@XXXXX-asus:/var/lib$ sudo mongos --bind_ip localhost --port 10100 --logpath /var/lib/mongodb_sharding/mongos_log/mongos.log --configdb localhost:10000 --fork forked process: 8709 all output going to: /var/lib/mongodb_sharding/mongos_log/mongos.log log file [/var/lib/mongodb_sharding/mongos_log/mongos.log] exists; copied to temporary file [/var/lib/mongodb_sharding/mongos_log/mongos.log.2014-03-21T08-12-03] child process started successfully, parent exiting
然后启动2个shard，分别监听20000和30000端口。
XXXXX@XXXXX-asus:/var/lib$ sudo mongod --dbpath /var/lib/mongodb_sharding/shard1 --logpath /var/lib/mongodb_sharding/shard1_log/mongodb.log --port 20000 --nojournal --fork forked process: 8834 all output going to: /var/lib/mongodb_sharding/shard1_log/mongodb.log log file [/var/lib/mongodb_sharding/shard1_log/mongodb.log] exists; copied to temporary file [/var/lib/mongodb_sharding/shard1_log/mongodb.log.2014-03-21T08-15-43] child process started successfully, parent exiting XXXXX@XXXXX-asus:/var/lib$ sudo mongod --dbpath /var/lib/mongodb_sharding/shard2 --logpath /var/lib/mongodb_sharding/shard2_log/mongodb.log --port 30000 --nojournal --fork forked process: 8846 all output going to: /var/lib/mongodb_sharding/shard2_log/mongodb.log child process started successfully, parent exiting XXXXX@XXXXX-asus:/var/lib$
然后用admin用户登陆mongos，并将2个shard加入到整个分片体系中，
XXXXX@XXXXX-asus:/var/lib$ mongo localhost:10100 MongoDB shell version: 2.2.4 connecting to: localhost:10100/test mongos> use admin switched to db admin mongos> db.printShardingStatus() --- Sharding Status --- sharding version: { "_id" : 1, "version" : 3 } shards: databases: { "_id" : "admin", "partitioned" : false, "primary" : "config" } mongos> db.runCommand({addshard:"localhost:20000", allowLocal:true}) { "shardAdded" : "shard0000", "ok" : 1 } mongos> db.runCommand({addshard:"localhost:30000", allowLocal:true}) { "shardAdded" : "shard0001", "ok" : 1 } mongos> db.printShardingStatus() --- Sharding Status --- sharding version: { "_id" : 1, "version" : 3 } shards: { "_id" : "shard0000", "host" : "localhost:20000" } { "_id" : "shard0001", "host" : "localhost:30000" } databases: { "_id" : "admin", "partitioned" : false, "primary" : "config" } mongos>
一般情况下，如果要做sharding的话，会对某个表的“_id”这个列来做片键（shard key），因为这个是每个集合都有的元素。当然，也可以用其他的列来做片键，不过要对这个列新建索引，否则对集合做shardcollection的时候会报错。本实验中仍然是用"_id"来做片键，实验用的数据库是test_sharding，集合是test_sharding.test。
mongos> use admin switched to db admin mongos> db.runCommand({"enablesharding":"test_sharding"}) { "ok" : 1 } mongos> db.runCommand({"shardcollection":"test_sharding.test","key":{"_id":1}}) { "collectionsharded" : "test_sharding.test", "ok" : 1 } mongos>
在没有insert任何数据的情况下，test集合的统计信息如下
mongos> use test_sharding switched to db test_sharding mongos> db.test.stats() { "sharded" : true, "ns" : "test_sharding.test", "count" : 0, "numExtents" : 1, "size" : 0, "storageSize" : 8192, "totalIndexSize" : 8176, "indexSizes" : { "_id_" : 8176 }, "avgObjSize" : 0, "nindexes" : 1, "nchunks" : 1, "shards" : { "shard0000" : { "ns" : "test_sharding.test", "count" : 0, "size" : 0, "storageSize" : 8192, "numExtents" : 1, "nindexes" : 1, "lastExtentSize" : 8192, "paddingFactor" : 1, "systemFlags" : 1, "userFlags" : 0, "totalIndexSize" : 8176, "indexSizes" : { "_id_" : 8176 }, "ok" : 1 } }, "ok" : 1 } mongos>
可以看到sharding已经被启用，但是此时只有一个shard在使用中，然后用脚本插入很多条数据，我是用ruby
#!/usr/bin/env ruby # 20140321, sharding_test.rb ### # test sharding ### require "rubygems" require "mongo" class MongoConnection def initialize(host, port) @mongoconn = Mongo::MongoClient.new(host, port) @db = @mongoconn.db("test_sharding") end def test() @coll = @db.collection("test") @coll.insert({"num"=>0}) (1..600000).each { |i| @coll.insert({ num:i, num_org:i, create_at:Time.now, txt:"The quick brown fox jumps over the lazy dog." } ) } puts @coll.find({num:5000}).explain end end mongo_conn = MongoConnection.new("localhost", 10100) puts "======TEST WITHOUT INDEX========================" mongo_conn.test()
插入600000条数据之后，集合的统计信息发生了明显的变化
mongos> db.test.stats() { "sharded" : true, "ns" : "test_sharding.test", "count" : 600001, "numExtents" : 18, "size" : 72000032, "storageSize" : 116883456, "totalIndexSize" : 19491584, "indexSizes" : { "_id_" : 19491584 }, "avgObjSize" : 119.99985333357778, "nindexes" : 1, "nchunks" : 6, "shards" : { "shard0000" : { "ns" : "test_sharding.test", "count" : 319678, "size" : 38361360, "avgObjSize" : 120, "storageSize" : 58441728, "numExtents" : 9, "nindexes" : 1, "lastExtentSize" : 20643840, "paddingFactor" : 1, "systemFlags" : 1, "userFlags" : 0, "totalIndexSize" : 10383520, "indexSizes" : { "_id_" : 10383520 }, "ok" : 1 }, "shard0001" : { "ns" : "test_sharding.test", "count" : 280323, "size" : 33638672, "avgObjSize" : 119.99968607641898, "storageSize" : 58441728, "numExtents" : 9, "nindexes" : 1, "lastExtentSize" : 20643840, "paddingFactor" : 1, "systemFlags" : 1, "userFlags" : 0, "totalIndexSize" : 9108064, "indexSizes" : { "_id_" : 9108064 }, "ok" : 1 } }, "ok" : 1 } mongos>
可见，2个shard上都被分配到了数据，分配的还算近似均匀。试验中，如果数据少的话，基本看不到sharding的效果，数据都放在shard0000上，但是数据多了之后，sharding的效果就有了。引用别人博客上的一段话，不知道对不对，来自这里（http://blog.sina.com.cn/s/blog_502c8cc40100pdiv.html）
为了测试Sharding的balance效果，我陆续插入了大约200M的数据，插入过程中使用db.stats() 查询数据分布情况。发现在数据量较小，30M以下时，所有trunk都存储在了shard0000上，但继续插入后，数据开始平均分布，并且mongos会对多个shard之间的数据进行rebalance 。在插入数据达到200M，刚插入结束时，shard0000上大约有135M数据，而shard0001上大约有65M数据，但过一段时间之后，shard0000上的数据量减少到了115M，shard0001上的数据量达到了85M。
实验中发现，mongodb对shard的选择有某种记忆性，比如，写入10000条数据，最后第10000条数据是写在shard0001上的，此时用remove删除所有数据，再次写数据的时候，发现mongodb是从shard0001开始写，而不是shard0000。还有一个比较恶心的地方，mongodb似乎不允许更改shard key，即时是空集合也不行，至少我没有找到相关的方法。
insert数据之后可以发现sharding status发生变化
mongos> db.printShardingStatus() --- Sharding Status --- sharding version: { "_id" : 1, "version" : 3 } shards: { "_id" : "shard0000", "host" : "localhost:20000" } { "_id" : "shard0001", "host" : "localhost:30000" } databases: { "_id" : "admin", "partitioned" : false, "primary" : "config" } { "_id" : "test", "partitioned" : false, "primary" : "shard0000" } { "_id" : "test_sharding", "partitioned" : true, "primary" : "shard0000" } test_sharding.test chunks: shard0000 3 shard0001 3 { "_id" : { $minKey : 1 } } -->> { "_id" : ObjectId("532c3064cf7c7c2f53000001") } on : shard0000 Timestamp(5000, 0) { "_id" : ObjectId("532c3064cf7c7c2f53000001") } -->> { "_id" : ObjectId("532c306acf7c7c2f530012d9") } on : shard0001 Timestamp(5000, 1) { "_id" : ObjectId("532c306acf7c7c2f530012d9") } -->> { "_id" : ObjectId("532c310acf7c7c2f530290e0") } on : shard0000 Timestamp(4000, 1) { "_id" : ObjectId("532c310acf7c7c2f530290e0") } -->> { "_id" : ObjectId("532c31a1cf7c7c2f5304f397") } on : shard0000 Timestamp(3000, 2) { "_id" : ObjectId("532c31a1cf7c7c2f5304f397") } -->> { "_id" : ObjectId("532c3234cf7c7c2f53075090") } on : shard0001 Timestamp(4000, 2) { "_id" : ObjectId("532c3234cf7c7c2f53075090") } -->> { "_id" : { $maxKey : 1 } } on : shard0001 Timestamp(4000, 3) mongos>
此时，还可以再增加一个shard，操作和之前类似
XXXXX@XXXXX-asus:/var/lib$ sudo mkdir -p mongodb_sharding/shard3 XXXXX@XXXXX-asus:/var/lib$ sudo mkdir -p mongodb_sharding/shard3_log XXXXX@XXXXX-asus:/var/lib$ sudo mongod --dbpath /var/lib/mongodb_sharding/shard3 --logpath /var/lib/mongodb_sharding/shard3_log/mongodb.log --port 40000 --nojournal --fork forked process: 12525 all output going to: /var/lib/mongodb_sharding/shard3_log/mongodb.log child process started successfully, parent exiting XXXXX@XXXXX-asus:/var/lib$ mongo localhost:10100 MongoDB shell version: 2.2.4 connecting to: localhost:10100/test mongos> use admin switched to db admin mongos> db.runCommand({addshard:"localhost:40000", allowLocal:true}) { "shardAdded" : "shard0002", "ok" : 1 } mongos>
加入新的shard之后，数据会自动飘移到新的shard上，可以查看集合的统计信息
mongos> db.test.stats() { "sharded" : true, "ns" : "test_sharding.test", "count" : 600001, "numExtents" : 22, "size" : 72000032, "storageSize" : 117579776, "totalIndexSize" : 19499760, "indexSizes" : { "_id_" : 19499760 }, "avgObjSize" : 119.99985333357778, "nindexes" : 1, "nchunks" : 6, "shards" : { "shard0000" : { "ns" : "test_sharding.test", "count" : 319678, "size" : 38361360, "avgObjSize" : 120, "storageSize" : 58441728, "numExtents" : 9, "nindexes" : 1, "lastExtentSize" : 20643840, "paddingFactor" : 1, "systemFlags" : 1, "userFlags" : 0, "totalIndexSize" : 10383520, "indexSizes" : { "_id_" : 10383520 }, "ok" : 1 }, "shard0001" : { "ns" : "test_sharding.test", "count" : 275499, "size" : 33059880, "avgObjSize" : 120, "storageSize" : 58441728, "numExtents" : 9, "nindexes" : 1, "lastExtentSize" : 20643840, "paddingFactor" : 1, "systemFlags" : 1, "userFlags" : 0, "totalIndexSize" : 8952720, "indexSizes" : { "_id_" : 8952720 }, "ok" : 1 }, "shard0002" : { "ns" : "test_sharding.test", "count" : 4824, "size" : 578792, "avgObjSize" : 119.98175787728026, "storageSize" : 696320, "numExtents" : 4, "nindexes" : 1, "lastExtentSize" : 524288, "paddingFactor" : 1, "systemFlags" : 1, "userFlags" : 0, "totalIndexSize" : 163520, "indexSizes" : { "_id_" : 163520 }, "ok" : 1 } }, "ok" : 1 } mongos>
用admin账户可以看新的sharding status
mongos> db.printShardingStatus() --- Sharding Status --- sharding version: { "_id" : 1, "version" : 3 } shards: { "_id" : "shard0000", "host" : "localhost:20000" } { "_id" : "shard0001", "host" : "localhost:30000" } { "_id" : "shard0002", "host" : "localhost:40000" } databases: { "_id" : "admin", "partitioned" : false, "primary" : "config" } { "_id" : "test", "partitioned" : false, "primary" : "shard0000" } { "_id" : "test_sharding", "partitioned" : true, "primary" : "shard0000" } test_sharding.test chunks: shard0002 2 shard0000 2 shard0001 2 { "_id" : { $minKey : 1 } } -->> { "_id" : ObjectId("532c3064cf7c7c2f53000001") } on : shard0002 Timestamp(6000, 0) { "_id" : ObjectId("532c3064cf7c7c2f53000001") } -->> { "_id" : ObjectId("532c306acf7c7c2f530012d9") } on : shard0002 Timestamp(7000, 0) { "_id" : ObjectId("532c306acf7c7c2f530012d9") } -->> { "_id" : ObjectId("532c310acf7c7c2f530290e0") } on : shard0000 Timestamp(6000, 1) { "_id" : ObjectId("532c310acf7c7c2f530290e0") } -->> { "_id" : ObjectId("532c31a1cf7c7c2f5304f397") } on : shard0000 Timestamp(3000, 2) { "_id" : ObjectId("532c31a1cf7c7c2f5304f397") } -->> { "_id" : ObjectId("532c3234cf7c7c2f53075090") } on : shard0001 Timestamp(7000, 1) { "_id" : ObjectId("532c3234cf7c7c2f53075090") } -->> { "_id" : { $maxKey : 1 } } on : shard0001 Timestamp(4000, 3) mongos>
可以看到，新增的shard没有很好的去平均所有的数据，而是只有很少的一部分数据飘移到了新的shard中，当然，这可能也是因为数据不足够多造成的。