不同的适用场景下应该选择不同的扩容方式，本文主要介绍了四种主要的扩容实践与代码样例。

方案一

环境准备

建立数据域

db.createDomain("domainbefore",["datagroup1","datagroup2","datagroup3"], {AutoSplit:true});

建立集合空间

db.createCS("pri5_image",{Domain:"domainbefore"});

建立主表集合

db.pri5_image.createCL("atnnotate",{ShardingKey:{"dated":1},IsMainCL:true});

建立子表集合

db.pri5_image.createCL("atnnotate2016",{ShardingKey:{"name":1},ShardingType:"hash"});

db.pri5_image.createCL("atnnotate2017",{ShardingKey:{"name":1},ShardingType:"hash"});

attach子表集合

db.pri5_image.atnnotate.attachCL("pri5_image.atnnotate2016",{LowBound:{dated:"2016-01-01"},UpBound:{dated:"2017-01-01"}});

db.pri5_image.atnnotate.attachCL("pri5_image.atnnotate2017",{LowBound:{dated:"2017-01-01"},UpBound:{dated:"2018-01-01"}});

查看主表编目信息

> db.snapshot(SDB_SNAP_CATALOG,{"Name":"pri5_image.atnnotate"});

{…省略

  "CataInfo": [

    {

      "ID": 1,

      "SubCLName": "pri5_image.atnnotate2016",

      "LowBound": {"dated": "2016-01-01"},

      "UpBound": {"dated": "2017-01-01"}

    },

    {

      "ID": 2,

      "SubCLName": "pri5_image.atnnotate2017",

      "LowBound": {

        "dated": "2017-01-01"

      },

      "UpBound": {

        "dated": "2018-01-01"

      }

    }

  ],

  "IsMainCL": true,

  "Name": "pri5_image.atnnotate",

  "ShardingKey": {

    "dated": 1

  }

 …省略

}

此时主表包含两个子表，以dated为分区键

查看子表集合分布的数据组

子表pri5_image.atnnotate2017和pri5_image.atnnotate2016数据组信息一样，在此省略，此外省略了CataInfo的部分。

> db.snapshot(SDB_SNAP_CATALOG,{"Name":"pri5_image.atnnotate2016"});

{…省略

  "CataInfo": [

    {

      "GroupName": "datagroup1",

      "LowBound": {"": 0},

      "UpBound": {"": 1365}

    },

    {

      "GroupName": "datagroup2",

      "LowBound": {"": 1365},

      "UpBound": {"": 2730}

    },

    {

      "GroupName": "datagroup3",

      "LowBound": {"": 2730},

      "UpBound": {"": 4096}

    }

  ],

  "MainCLName": "pri5_image.atnnotate",

  "Name": "pri5_image.atnnotate2016"

  …省略

}

各导入10000条dated 在2016年和2017年的数据

查看数据量

> db.pri5_image.atnnotate.count();

20000

> db.pri5_image.atnnotate2016.count();

10000

> db.pri5_image.atnnotate2017.count();

10000

 

扩容准备

新增数据组datagroup4、datagroup5和datagroup6

添加至原数据域中

db.getDomain("domainbefore").alter({Groups:['datagroup1', 'datagroup2', 'datagroup3', 'datagroup4', 'datagroup5', 'datagroup6']})

新建子表

db.pri5_image.createCL("atnnotate2016_2",{ShardingKey:{"name":1},ShardingType:"hash"});

db.pri5_image.createCL("atnnotate2017_2",{ShardingKey:{"name":1},ShardingType:"hash"});

查看新建子表的编目信息

子表pri5_image.atnnotate2017_2和pri5_image.atnnotate2016_2数据组信息一样，在此省略

> db.snapshot(SDB_SNAP_CATALOG,{"Name":"pri5_image.atnnotate2016_2"});

{

 …省略

  "CataInfo": [

    {

      "GroupName": "datagroup1",

      "LowBound": {"": 0},

      "UpBound": {"": 682}

    },

    {

      "GroupName": "datagroup2",

      "LowBound": {"": 682},

      "UpBound": {"": 1364}

    },

    {

      "GroupName": "datagroup3",

      "LowBound": {"": 1364},

      "UpBound": {"": 2046}

    },

    {

      "GroupName": "datagroup4",

      "LowBound": {"": 2046},

      "UpBound": {"": 2728}

    },

    {

      "GroupName": "datagroup5",

      "LowBound": {"": 2728},

      "UpBound": {"": 3410}

    },

    {

      "GroupName": "datagroup6",

      "LowBound": {"": 3410},

      "UpBound": {"": 4096}

    }

  ],

  "AutoSplit": true

}

新的子表建立在了包含新增的数据组的总共6个数据组上。

 

开始扩容

创建管道文件pri5_image_atnnotate2016.json

mknod  pri5_image_atnnotate2016.json p

导出原表pri5_image.atnnotate2016数据，导入到pri5_image.atnnotate2016_2

nohup sdbexprt -s localhost -p 11810 --type json  --file ./pri5_image_atnnotate2016.json -c pri5_image -l atnnotate2016 --fields '_id,id,name,doubled,longd,boold,dated'  & >/dev/null

nohup sdbimprt -s localhost -p 11810 --type json  --file ./pri5_image_atnnotate2016.json -c pri5_image -l atnnotate2016_2 --fields '_id oid,id int,name string,doubled double,longd long,boold bool,dated string' & >/dev/null

查看pri5_image.atnnotate2016_2数据量

> db.pri5_image.atnnotate2016_2.count();

10000

按照id字段排序，各导出1000条pri5_image.atnnotate2016和pri5_image.atnnotate2016_2数据，进行diff，确认导出导入的数据是否一致

sdb "db.pri5_image.atnnotate2016.find().sort({"_id":1}).limit(1000)">>pri5_image_atnnotate2016.rec

sdb "db.pri5_image.atnnotate2016_2.find().sort({"_id":1}).limit(1000)">>pri5_image_atnnotate2016_2.rec

diff pri5_image_atnnotate2016.rec pri5_image_atnnotate2016_2.rec

detach原子表pri5_image.atnnotate2016，attach 新子表pri5_image.atnnotate2016

编写脚本delta-t.js，计算detach原表和attach新子表的耗时

var db = new Sdb();

var rdate = new Date();

var begin = rdate.getSeconds()*1000+rdate.getMilliseconds();

db.pri5_image.atnnotate.detachCL("pri5_image.atnnotate2016");

db.pri5_image.atnnotate.attachCL("pri5_image.atnnotate2016_2",{LowBound:{dated:"2016-01-01"},UpBound:{dated:"2017-01-01"}});

var ldate = new Date();

var end = ldate.getSeconds()*1000+ldate.getMilliseconds();

println(end-begin);

执行脚本

sdb –f delta-t.js

过程耗时41毫秒，detachCL旧表和attachCL新表过程很快，应用几乎无感

删除原子表pri5_image.atnnotate2016,释放空间给后面的子表数据迁移

db.pri5_image.dropCL("pri5_image.atnnotate2016")

进行下一个子表pri5_image.atnnotate2017的扩容，步骤和上述步骤一致，不再详述。

 

方案二

环境准备

和第一种方案环境准备章节一致，在以下部分的扩容步骤和第一种方案不同。

准备扩容

新增数据组datagroup4、datagroup5和datagroup6

添加至原数据域中

db.getDomain("domainbefore").alter({Groups:['datagroup1', 'datagroup2', 'datagroup3', 'datagroup4', 'datagroup5', 'datagroup6']})

新建子表

db.pri5_image.createCL("atnnotate2018",{ShardingKey:{"name":1},ShardingType:"hash",AutoSplit:false,Group:"datagroup4"});

查看子表编目信息,省略了部分信息

> db.snapshot(SDB_SNAP_CATALOG,{"Name":"pri5_image.atnnotate2018"});

{  …省略

  "Name": "pri5_image.atnnotate2018",

  "CataInfo": [

    {

      "GroupName": "datagroup4",

      "LowBound": {"": 0},

      "UpBound": {"": 4096}

    }

  ],

  "AutoSplit": false

}

 

开始扩容

使用split拆分数据到数据组datagroup5和datagroup上

db.pri5_image.atnnotate2018.split("datagroup4","datagroup6",{"id":2730},{"id":4096})

db.pri5_image.atnnotate2018.split("datagroup4","datagroup5",{"id":1365},{"id":2730})

查看拆分后的子表编目信息

> db.snapshot(SDB_SNAP_CATALOG,{"Name":"pri5_image.atnnotate2018"});

{…省略

  "AutoSplit": false,

  "CataInfo": [

    {

      "GroupName": "datagroup4",

      "LowBound": {"": 0},

      "UpBound": {"": 1365}

    },

    {

      "GroupName": "datagroup5",

      "LowBound": {"": 1365},

      "UpBound": {"": 2730}

    },

    {

      "GroupName": "datagroup6",

      "LowBound": {"": 2730},

      "UpBound": {"": 4096}

    }

  ],

  "Name": "pri5_image.atnnotate2018",

  …省略

}

新增的子表均匀切分到了新增的数据组上。

attach到主表

db.pri5_image.atnnotate.attachCL("pri5_image.atnnotate2018",{LowBound:{dated:"2018-01-01"},UpBound:{dated:"2019-01-01"}});

如果还有更多子表需要添加，重复以上步骤，进行扩容。

 

方案三

环境准备

和第一种方案环境准备章节一致，在以下部分的扩容步骤和第一种方案不同。

准备扩容

新建域

db.createDomain("domainafter",["datagroup4","datagroup5","datagroup6"], {AutoSplit:true});

建立集合空间

db.createCS("pri5_image_2",{Domain:"domainafter"});

创建子表

db.pri5_ image2.createCL("atnnotate2018",{ShardingKey:{"name":1},ShardingType:"hash"});

查看新建子表的编目信息

省略了部分输出信息

> db.snapshot(SDB_SNAP_CATALOG,{"Name":"pri5_image2.atnnotate2018"});

{

  "CataInfo": [

    {

      "GroupName": "datagroup4",

      "LowBound": {"": 0},

      "UpBound": {"": 1365}

    },

    {

      "GroupName": "datagroup5",

      "LowBound": {"": 1365},

      "UpBound": {"": 2730}

    },

    {

      "GroupName": "datagroup6",

      "LowBound": {"": 2730},

      "UpBound": {"": 4096}

    }

  ],

  "AutoSplit": true

}

 

开始扩容

attach到主表

db.pri5_image.atnnotate.attachCL("pri5_image_2.atnnotate2018",{LowBound:{dated:"2018-01-01"},UpBound:{dated:"2019-01-01"}});

查看主表编目信息

> db.snapshot(SDB_SNAP_CATALOG,{"Name":"pri5_image.atnnotate"});

{

  "CataInfo": [

    {

      "SubCLName": "pri5_image.atnnotate2017",

      "LowBound": {"dated": "2017-01-01"},

      "UpBound": {"dated": "2018-01-01"}

    },

    {

      "SubCLName": "pri5_image.atnnotate2016",

      "LowBound": {"dated": "2016-01-01"},

      "UpBound": {"dated": "2017-01-01"}

    },

    {

      "SubCLName": "pri5_image_2.atnnotate2018",

      "LowBound": {"dated": "2018-01-01"},

      "UpBound": {"dated": "2019-01-01"}

    }

  ],

  "EnsureShardingIndex": true,

  "IsMainCL": true,

  "Name": "pri5_image.atnnotate",

  "ShardingKey": {"dated": 1},

  "ShardingType": "range",

}

新增的子表均匀切分到了新增的数据组上。

如果还有更多子表需要添加，重复以上步骤，进行扩容。

 

方案四

环境准备

创建集合空间及集合

>db.createCS("foo").createCL("bar",{ShardingKey:{"id":1},ShardingType:"hash"});

查看编目信息

> db.snapshot(SDB_SNAP_CATALOG,{"Name":"foo.bar"});

{

  "CataInfo": [

    {

      "GroupName": "datagroup3",

      "LowBound": {"": 0},

      "UpBound": {"": 4096}

    }

  ]

}

 

开始扩容

开始切分

> db.foo.bar.split("datagroup3","datagroup1",{"id":0},{"id":2048});

查看编目信息

> db.snapshot(SDB_SNAP_CATALOG,{"Name":"foo.bar"});

{

  "CataInfo": [

    {

      "GroupName": "datagroup1",

      "LowBound": {"": 0},

      "UpBound": {"": 2048}

    },

    {

      "GroupName": "datagroup3",

      "LowBound": {"": 2048},

      "UpBound": {"": 4096}

    }

  ]

}

分区键在0-2048范围内的被切分到了datagroup1上。