前面介绍了
如何搭建mongodb sharding 集群
，本文对shard进行大量数据拆分测试，并谈了对于片键的选择注意事项，（可能不全，希望指教）
1 激活test数据库的分片功能。
mongos> db.runCommand({"enablesharding": "test"})
{ "ok" : 1 }
2 查看整个sharding 的架构
mongos> db.runCommand({listshards:1});
{
"shards" : [
{
"_id" : "shard0000",
"host" : "10.250.7.225:27018"
},
{
"_id" : "shard0001",
"host" : "10.250.7.249:27019"
},
{
"_id" : "shard0002",
"host" : "10.250.7.241:27020"
}
],
"ok" : 1
}
mongos> printShardingStatus();
--- Sharding Status ---
sharding version: { "_id" : 1, "version" : 3 }
shards:
{ "_id" : "shard0000", "host" : "10.250.7.225:27018" }
{ "_id" : "shard0001", "host" : "10.250.7.249:27019" }
{ "_id" : "shard0002", "host" : "10.250.7.241:27020" }
databases:
{ "_id" : "admin", "partitioned" : false, "primary" : "config" }
{ "_id" : "test", "partitioned" : true, "primary" : "shard0000" }
Note：test 数据库的 partitioned 为true 意味开启了sharding 功能。
3 激活test数据库中的集合sharding功能，并指定片键为"_id"，索引唯一。
mongos> db.runCommand({shardcollection:'test.yql',key:{_id:1}, unique : true});
{ "collectionsharded" : "test.yql", "ok" : 1 }
插入17条数据，并查看yql的状况
mongos> db.yql.stats();
{
"sharded" : true,
"flags" : 1,
"ns" : "test.yql",
"count" : 17,
"numExtents" : 1,
"size" : 1616,
"storageSize" : 8192,
"totalIndexSize" : 8176,
"indexSizes" : {
"_id_" : 8176
},
"avgObjSize" : 95.05882352941177,
"nindexes" : 1,
"nchunks" : 1,
"shards" : {
"shard0000" : { --表示yql被拆分到了10.250.7.225这台机器上了
"ns" : "test.yql",
"count" : 17,
"size" : 1616,
"avgObjSize" : 95.05882352941177,
"storageSize" : 8192,
"numExtents" : 1,
"nindexes" : 1,
"lastExtentSize" : 8192,
"paddingFactor" : 1,
"flags" : 1,
"totalIndexSize" : 8176,
"indexSizes" : {
"_id_" : 8176
},
"ok" : 1
}
},
"ok" : 1
}
mongos>
4 使用 pymongodb 向test.yql插入200w的数据
[mongodb@rac4 pymongo]$ python
Python 2.4.3 (#1, Jan 21 2009, 01:11:33)
[GCC 4.1.2 20071124 (Red Hat 4.1.2-42)] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>>
>>> import pymongo
>>>conn = pymongo.Connection(host="127.0.0.1")
>>> db=conn.test
>>> for i in xrange(2000000):
... val=dict(val="yangql"+str(i))
... db.yql.insert(val)
ObjectId('4eb2aa5940643e5bb61e847b')
ObjectId('4eb2aa5940643e5bb61e847c')
ObjectId('4eb2aa5940643e5bb61e847d')
ObjectId('4eb2aa5940643e5bb61e847e')
ObjectId('4eb2aa5940643e5bb61e847f')
....省略....
5 再次查看,collection yql在整个sharding的分布状况：这次collection yql 被拆分到了三个不同的shard上面。
mongos> db.yql.stats();
{
"sharded" : true,
"flags" : 1,
"ns" : "test.yql",
"count" : 2000017,
"numExtents" : 23,
"size" : 92001336,
"storageSize" : 146833408,
"totalIndexSize" : 64917440,
"indexSizes" : {
"_id_" : 64917440
},
"avgObjSize" : 46.00027699764552,
"nindexes" : 1,
"nchunks" : 4,
"shards" : {
"shard0000" : {
"ns" : "test.yql",
"count" : 29448,
"size" : 1296204,
"avgObjSize" : 44.016707416462914,
"storageSize" : 2793472,
"numExtents" : 5,
"nindexes" : 1,
"lastExtentSize" : 2097152,
"paddingFactor" : 1,
"flags" : 1,
"totalIndexSize" : 964768,
"indexSizes" : {
"_id_" : 964768
},
"ok" : 1
},
"shard0001" : {
"ns" : "test.yql",
"count" : 1942524,
"size" : 89471128,
"avgObjSize" : 46.05921368281679,
"storageSize" : 141246464, --这里shard0001上面的数据比较多，和片键的选择有关。
"numExtents" : 13,
"nindexes" : 1,
"lastExtentSize" : 30072832,
"paddingFactor" : 1,
"flags" : 1,
"totalIndexSize" : 63036960,
"indexSizes" : {
"_id_" : 63036960
},
"ok" : 1
},
"shard0002" : {
"ns" : "test.yql",
"count" : 28045,
"size" : 1234004,
"avgObjSize" : 44.00085576751649,
"storageSize" : 2793472,
"numExtents" : 5,
"nindexes" : 1,
"lastExtentSize" : 2097152,
"paddingFactor" : 1,
"flags" : 1,
"totalIndexSize" : 915712,
"indexSizes" : {
"_id_" : 915712
},
"ok" : 1
}
},
"ok" : 1
}
从下面的查询结果可以看出，集合yql 根据片键的分布：
shard0000 _id:{$minKey:1} -->>"4eb298b3adbd9673afee95e3"
shard0000 _id:4eb298b3adbd9673afee95e3"-->>"4eb2a64640643e5bb60072f7"
shard0002 _id: "4eb2a64640643e5bb60072f7-->>"4eb2a65340643e5bb600e084"
shard0001 _id:4eb2a65340643e5bb600e084"-->>{ $maxKey:1}
mongos> printShardingStatus();
--- Sharding Status ---
sharding version: { "_id" : 1, "version" : 3 }
shards:
{ "_id" : "shard0000", "host" : "10.250.7.225:27018" }
{ "_id" : "shard0001", "host" : "10.250.7.249:27019" }
{ "_id" : "shard0002", "host" : "10.250.7.241:27020" }
databases:
{ "_id" : "admin", "partitioned" : false, "primary" : "config" }
{ "_id" : "test", "partitioned" : true, "primary" : "shard0000" }
test.yql chunks:
shard0000 2
shard0002 1
shard0001 1
{ "_id" : { $minKey : 1 } } -->> { "_id" : ObjectId("4eb298b3adbd9673afee95e3") } on : shard0000 { "t" : 2000, "i" : 1 }
{ "_id" : ObjectId("4eb298b3adbd9673afee95e3") } -->> { "_id" : ObjectId("4eb2a64640643e5bb60072f7") } on : shard0000 { "t" : 1000, "i" : 3 }
{ "_id" : ObjectId("4eb2a64640643e5bb60072f7") } -->> { "_id" : ObjectId("4eb2a65340643e5bb600e084") } on : shard0002 { "t" : 3000, "i" : 1 }
{ "_id" : ObjectId("4eb2a65340643e5bb600e084") } -->> { "_id" : { $maxKey : 1 } } on : shard0001 { "t" : 3000, "i" : 0 }
mongos>
在配置的时候，我们只是指定了片键，对集合的拆分则是由mongos来进行的，即集群自动进行数据拆分，并进行负载均衡！对于片键的选择会影响到集合在shard的分布，对于本文中的例子：
选择_id 为片键，分了四个区间：
[{$minKey:1},"4eb298b3adbd9673afee95e3" ]
["4eb298b3adbd9673afee95e3","4eb2a64640643e5bb60072f7"]
["4eb2a64640643e5bb60072f7","4eb2a65340643e5bb600e084"]
["4eb2a65340643e5bb600e084",{$maxKey:1}]
随着数据的继续插入，再进行拆分数据插入的数据依然会插入到shard0001服务器上的，对于上面的例子再次插入200w的数据：
查看yql的分片分布：shard0000 和shard0002上面存储的数据没有变化，所有插入的数据全部落入到shard0001上面。
mongos> db.yql.stats();
{
"sharded" : true,
"flags" : 1,
"ns" : "test.yql",
"count" : 4000017,
"numExtents" : 26,
"size" : 184000936,
"storageSize" : 278208512,
"totalIndexSize" : 129802176,
"indexSizes" : {
"_id_" : 129802176
},
"avgObjSize" : 46.00003849983638,
"nindexes" : 1,
"nchunks" : 4,
"shards" : {
"shard0000" : {
"ns" : "test.yql",
"count" : 29448,
"size" : 1296204,
"avgObjSize" : 44.016707416462914,
"storageSize" : 2793472,
"numExtents" : 5,
"nindexes" : 1,
"lastExtentSize" : 2097152,
"paddingFactor" : 1,
"flags" : 1,
"totalIndexSize" : 964768,
"indexSizes" : {
"_id_" : 964768
},
"ok" : 1
},
"shard0001" : {
"ns" : "test.yql",
"count" : 3942524,
"size" : 181470728,
"avgObjSize" : 46.02907376086994,
"storageSize" : 272621568,
"numExtents" : 16,
"nindexes" : 1,
"lastExtentSize" : 51974144,
"paddingFactor" : 1,
"flags" : 1,
"totalIndexSize" : 127921696,
"indexSizes" : {
"_id_" : 127921696
},
"ok" : 1
},
"shard0002" : {
"ns" : "test.yql",
"count" : 28045,
"size" : 1234004,
"avgObjSize" : 44.00085576751649,
"storageSize" : 2793472,
"numExtents" : 5,
"nindexes" : 1,
"lastExtentSize" : 2097152,
"paddingFactor" : 1,
"flags" : 1,
"totalIndexSize" : 915712,
"indexSizes" : {
"_id_" : 915712
},
"ok" : 1
}
},
"ok" : 1
}
mongos>
这样不符合sharding的初衷：负载均衡，最终也会导致shard0001 上面的磁盘空间和内存不足。
Note：如果写入负载比较高，要分散负载，此时必须考虑选择均匀分布的片键。不能在只有几个值的片键，比如对于status 有四个值，A,B,C,D.mongos 不会创建多于4个块。选择片键时，除了均匀分布也要有个增长上线即最大值，否则就会出现上面的情况，违背了sharding的初衷。
