How to write an xquery containing sequence elements?

Question

I've got a very large xml data set that is structured like the following:

<root>
    <person>
        <personid>HH3269732</personid>
        <firstname>John</firstname>
        <lastname>Smith</lastname>
        <entertime>01/02/2008 10:15</entertime>
        <leavetime>01/02/2008 11:45</leavetime>
        <entertime>03/01/2008 08:00</entertime>
        <leavetime>03/01/2008 10:00</leavetime>
        ... 
        // number of enter times and leave times vary from person to person
        // there may not be a final leave time (ie, they haven't left yet)
    </person>
    ...
</root>

The structure of the data is not under my control. This data is currently residing in a single xml column in a single row in MS SQL Server 2005. I am trying to construct a query which results in the following output:

HH3269732   John   Smith   01/02/2008 10:15   01/02/2008 11:45
HH3269732   John   Smith   03/01/2008 08:00   01/02/2008 10:00
HH3269735   Mark   Pines   02/01/2008 09:00   NULL
HH3263562   James  Frank   NULL               NULL
HH3264237   Harold White   04/18/2008 03:00   04/18/2008 05:00
...

My query currently looks like the following:

DECLARE @xml xml
SELECT @xml = XmlCol FROM Data

SELECT
    [PersonId] = Persons.PersonCollection.value('(personid)[1]', 'NVARCHAR(50)')
    ,[First Name] = Persons.PersonCollection.value('(firstname)[1]', 'NVARCHAR(50)')
    ,[Last Name] = Persons.PersonCollection.value('(lastname)[1]', 'NVARCHAR(50)')
    ??????
FROM @xml.nodes('root\person') Persons(PersonCollection)

That query may not be 100% right as I'm pulling it from memory, but the problem I'm having is that I don't know how to include the entertime leavetime sequence elements in such a way as to get the desired rowset that I indicated above.

Thanks.

UPDATE: I wanted to add that a given person record may have no entertime/leavetime sequence elements at all, but still needs to be returned in the rowset. I have updated the example of the desired output to reflect this.

Answer 1

Haven't realized you may have multiple persons in the document. My query would be incorrect in that case anyway. I thought maybe if you first shred out each person into its own XML fragment and ten extract the enter/leave times might perform better. I don't have 215k person XML to try, but here is an idea:

declare @x xml;
select @x = N'<root>
    <person>
        <personid>HH3269732</personid>
        <firstname>John</firstname>
        <lastname>Smith</lastname>
        <entertime>01/02/2008 10:15</entertime>
        <leavetime>01/02/2008 11:45</leavetime>
        <entertime>03/01/2008 08:00</entertime>
        <leavetime>03/01/2008 10:00</leavetime>
        <entertime>04/01/2008 08:00</entertime>
    </person>
    <person>
        <personid>HH3269733</personid>
        <firstname>Jane</firstname>
        <lastname>Doe</lastname>
        <entertime>01/03/2008 10:15</entertime>
        <leavetime>01/03/2008 11:45</leavetime>
        <entertime>03/04/2008 08:00</entertime>
        <leavetime>03/04/2008 10:00</leavetime>
        <entertime>04/04/2008 08:00</entertime>
    </person>
</root>';


with cte_person as (
    select
        t.c.value('(personid)[1]', 'NVARCHAR(50)') as personid
        , t.c.value('(firstname)[1]', 'NVARCHAR(50)') as firstname
        , t.c.value('(lastname)[1]', 'NVARCHAR(50)') as lastname
        , t.c.query('entertime') as entertime
        , t.c.query('leavetime') as leavetime
    FROM @x.nodes('root/person') t(c))
, cte_cross_enter as (
    select
        p.personid
        , p.firstname
        , p.lastname
        , x.c.value('.', 'datetime') as entertime
        , row_number() over (partition by personid order by x.c) as row_enter
        from cte_person p
        cross apply p.entertime.nodes('/entertime') x(c))
, cte_cross_leave as (
    select
        p.personid 
        , x.c.value('.', 'datetime') as leavetime
        , row_number() over (partition by personid order by x.c) as row_leave
        from cte_person p
        cross apply p.leavetime.nodes('/leavetime') x(c))
select e.personid
    , e.firstname
    , e.lastname
    , e.entertime
    , l.leavetime
    from cte_cross_enter e
    left outer join cte_cross_leave l 
            on e.personid = l.personid and 
            e.row_enter = l.row_leave

Answer 2

I have accepted Remus's answer as it got me 95% to the solution. For informational purposes, here is the final query structure:

with cte_maindata as (
SELECT
    [PersonId] = t.c.value('(personid)[1]', 'NVARCHAR(50)')
    ,[First Name] = t.c.value('(firstname)[1]', 'NVARCHAR(50)')
    ,[Last Name] = t.c.value('(lastname)[1]', 'NVARCHAR(50)')
FROM @x.nodes('root/person') t(c))
, cte_entertime as (
    SELECT
    [PersonId] = t.c.value('(../personid)[1]', 'NVARCHAR(50)')
    ,[Entertime] = t.c.value('.', 'NVARCHAR(50)')
FROM @x.nodes('root/person/entertime') t(c))
, cte_leavetime as (
    SELECT
    [PersonId] = t.c.value('(../personid)[1]', 'NVARCHAR(50)')
    ,[Leavetime] = t.c.value('.', 'NVARCHAR(50)')
FROM @x.nodes('root/person/leavetime') t(c))
SELECT 
    m.PersonID
    ,[First Name]
    ,[Last Name] 
    ,[Entertime]
    ,[Leavetime]
FROM cte_maindata m
    LEFT OUTER JOIN cte_entertime e on m.PersonId = e.PersonId
    LEFT OUTER JOIN cte_leavetime l on m.PersonId = l.PersonId

Answer 3

with cte_entertime as (
SELECT
    [PersonId] = t.c.value('(../personid)[1]', 'NVARCHAR(50)')
    ,[First Name] = t.c.value('(../firstname)[1]', 'NVARCHAR(50)')
    ,[Last Name] = t.c.value('(../lastname)[1]', 'NVARCHAR(50)')
    ,[Entertime] = t.c.value('.', 'NVARCHAR(50)')
    ,[entry_number] = ROW_NUMBER() OVER (ORDER BY t.c)
FROM @x.nodes('root/person/entertime') t(c))
, cte_leavetime as (
    SELECT
    [Leavetime] = t.c.value('.', 'NVARCHAR(50)')
    ,[entry_number] = ROW_NUMBER() OVER (ORDER BY t.c)
FROM @x.nodes('root/person/leavetime') t(c))
SELECT PersonID
    , [First Name]
    , [Last Name] 
    , [Entertime]
    , [Leavetime]
    FROM cte_entertime e 
    LEFT OUTER JOIN cte_leavetime l on e.entry_number = l.entry_number