Design3：使用HierarchyID构建数据的分层结构

1，传统的分层结构是父子结构，表结构中有一个ParentID字段自引用表的主键，表示“归属”关系，例如

create table dbo.emph
(
ID int not null primary key,
ParentID int foreign key references dbo.emph(id),
Descr varchar(100) not null
)

示例数据是一个简单的HR职称结构，Boss，M表示的Mananger，L表示的是Leader，E表示的是Employee。

2，将父子结构转换为使用HierarchyID的结构

2.1 首先创建一个新表 dbo.emph2，代码以下

create table dbo.emph2
(
idpath hierarchyid not null primary key,
id int not null,
--parentpath as idpath.GetAncestor(1) persisted foreign key references dbo.emph2(idpath),
descr varchar(100)
)

create unique nonclustered  index idx_emph2_unique on dbo.emph2(id)

代码分析：

• 不需要ParentID字段，因为HierarchyID类型能够获取Ancestor(1)函数来获取ParentID，所以不需要通过外键关系来限制。
• IDPath能够标识每一行在分层结构中的位置，但是不能标识每一行的ID，所以ID字段是必须的。
• 创建一个唯一索引，强制ID唯一。

2.2 将父子结构的数据填充到使用HierarchyID构建的表中

;with cte(idpath,id,descr) as
(
select cast(HierarchyID::GetRoot().ToString() as Varchar(max)) as idpath,ID,Descr
from dbo.emph
where ParentID is null

union all

select cast(c.idpath+cast(e.id as varchar)+'/' as varchar(max)) as idpath,e.id,e.descr
from dbo.emph e
inner join cte c on e.parentid=c.id
)
insert dbo.emph2 (idpath,id,descr)
select idpath,id,descr
from cte

HierarchyID类型的字符串格式如：‘/1/2/3/’，字符串以‘/’开头，并以‘/’结尾；

HierarchyID类型不会自动生成节点的位置，需要在代码中将父子关系拼接成字符串，

2.3 查询新表dbo.emph2，将HierarchyID类型转换为字符串，能更直观地看出其结构和位置。例如，idpathstring='/2/6/15/'的上一级父节点的path是‘/2/6/’，idpath=‘/2/6/’的ID是6。

select idpath, idpath.ToString() as idpathstring,id,descr
from dbo.emph2

2.4 查看某一个节点的父节点，@NodeID是一个节点ID，@Upcnt是指@Node向上的第几个父节点，如果为Null，那么查询出所有的父节点。

alter procedure dbo.usp_GetAncestor
    @NodeID int,
    @Upcnt  int=null
as
begin

    declare @node HierarchyID

    select @node=idpath
    from dbo.emph2
    where id=@NodeID

    --get all ancestors
    if @Upcnt is null
    begin
        set @Upcnt=@node.GetLevel()

        --;with cte(idpath,id,descr,Level) as
        --(
        --    select idpath,id,descr, 0 as Level
        --    from dbo.emph2
        --    where id=@NodeID
        --    union all

        --    select e.idpath,e.id,e.descr,c.Level+1 as Level
        --    from dbo.emph2 e
        --    inner join cte c on e.idpath=c.idpath.GetAncestor(1)
        --    where c.Level<@Upcnt
        --)
        --select idpath,idpath.ToString() as idpathstring,id,descr
        --from cte

        declare @rt table(idpath hierarchyid,idpathstring varchar(max),id int ,descr varchar(100))
        declare @i int=0

        while @i<=@Upcnt
        begin
            insert into @rt
            select idpath, idpath.ToString() as idpathstring,id,descr
            from dbo.emph2
            where idpath=@node.GetAncestor(@i);

            set @i=@i+1
        end
        select idpath, idpath.ToString() as idpathstring,id,descr
        from @rt
    end
    else
    begin
        select idpath, idpath.ToString() as idpathstring,id,descr
        from dbo.emph2
        where idpath=@node.GetAncestor(@Upcnt);
    end
end

2.5 查询子节点

alter procedure dbo.usp_GetDescendant
    @NodeID int,
    @Downcnt  int=null
as
begin
    declare @Node hierarchyid

    select @node=idpath
    from dbo.emph2
    where id=@NodeID

    select idpath, idpath.ToString() as idpathstring,id,descr
    from dbo.emph2
    where idpath.IsDescendantOf(@Node)=1
        and (@Downcnt is null or (idpath.GetLevel()=@Node.GetLevel()+@Downcnt))
end

2.6 增加一个节点，有时节点的idpath是有顺序的，为了保证顺序，必须使用GetDescendant函数。

--在子节点序列的末尾加入新节点

create procedure dbo.usp_addnode
    @parentid    int,
    @id            int,
    @descr varchar(100)
as
begin
    declare @parentnode hierarchyid
    declare @maxchildnode hierarchyid

    select @parentnode=idpath
    from dbo.emph2
    where id=@parentid

    select @maxchildnode=max(idpath)
    from dbo.emph2
    where idpath.GetAncestor(1)=@parentnode

    insert into dbo.emph2(idpath,id,descr)
    select @parentnode.GetDescendant(@maxchildnode,null),@id,@descr
end

--按照一定的顺序插入子节点

alter procedure dbo.usp_addnode_order
    @parentid    int,
    @childleft    int,
    @childright int,
    @id            int,
    @descr varchar(100)
as
begin

declare @childrightnode hierarchyid
declare @childleftnode hierarchyid
declare @parentnode hierarchyid

select @childleftnode=idpath
from dbo.emph2
where id=@childleft

select @childrightnode=idpath
from dbo.emph2
where id=@childright

select @parentnode=idpath
from dbo.emph2
where id=@parentid

insert into dbo.emph2(idpath,id,descr)
select @parentnode.GetDescendant(@childleftnode,@childrightnode),@id,@descr

end

对stored procedure 进行测试，并对查询结果进行排序，如下图

exec dbo.usp_addnode 5,25,'L41'
exec dbo.usp_addnode 5,26,'L42'
exec dbo.usp_addnode_order 5,25,26,27,'L43'

select idpath, idpath.ToString() as idpathstring,id,descr
from dbo.emph2
order by idpath

2.7 删除一个节点

如果删除的是叶子节点，非常简单，删除叶子节点不会影响其他节点，但是，如果删除的是非叶子节点，必须处理好其子节点，否则，其子节点将会从层次结构游离出来，成为非法存在，所以在删除一个节点的同时，必须为其可能存在的子节点指定一个新的父节点。

alter procedure dbo.usp_deletenode
    @deleteid int,
    @childnewparentid int
as
begin
    declare @deletenode hierarchyid

    select @deletenode=idpath
    from dbo.emph2
    where id=@deleteid

    declare @id int
    declare @descr varchar(100)

    declare cur_child cursor
    for select id,descr from dbo.emph2
    where idpath.IsDescendantOf(@deletenode)=1 and id!=@deleteid

    open cur_child

    fetch next from cur_child into @id,@descr

    while @@FETCH_STATUS=0
    begin
        delete dbo.emph2 where id=@id
        exec dbo.usp_addnode @childnewparentid,@id,@descr

        fetch next from cur_child into @id,@descr
    end 

    close cur_child
    deallocate cur_child

    delete dbo.emph2 where id=@deleteid
end

注意：IsDescendantOf函数包含当前节点，要想获取当前节点的所有子孙节点，必须将当前节点过滤掉。

测试，将id=5的节点删除，并将其子节点挂在id=4的节点下。

exec dbo.usp_deletenode 5,4

查询结果

select idpath, idpath.ToString() as idpathstring,id,descr
from dbo.emph2
order by idpath

使用存储过程查询id=4的子孙节点

exec dbo.usp_getdescendant 4

2.8 更新一个节点

更新一个节点，变更其父节点，同样面临如何处理其子节点的问题。

create procedure dbo.usp_updatenode
    @id int,
    @parentid int,
    @childnewparentid int
as
begin
    --获取节点的idpath
    declare @deletenode hierarchyid

    select @deletenode=idpath
    from dbo.emph2
    where id=@id

    --删除旧节点，并变更节点的父节点
    declare @descr varchar(100)

    select @descr=descr
    from dbo.emph2
    where id=@id

    delete dbo.emph2 where id=@id

    exec dbo.usp_addnode @parentid,@id,@descr

    --逐个变更子节点的父节点
    declare @childid int
    declare @childdescr varchar(100)

    declare cur_child cursor
    for select id,descr from dbo.emph2
    where idpath.IsDescendantOf(@deletenode)=1 and id!=@id

    open cur_child
    fetch next from cur_child into @childid,@childdescr

    while @@FETCH_STATUS=0
    begin
        delete dbo.emph2 where id=@childid
        exec dbo.usp_addnode @childnewparentid,@childid,@childdescr

        fetch next from cur_child into @childid,@childdescr
    end 

    close cur_child
    deallocate cur_child
end

测试数据如下

测试的目的是将id=5的所有子节点挂在id=4的节点下，并强id=5的父节点变更为id=8的节点

exec dbo.usp_updatenode 5,8,4

查询结果

select idpath, idpath.ToString() as idpathstring,id,descr
from dbo.emph2
order by idpath