Improve performance of SQL query with a FOR XML PATH in SQL Server 2008 - sql

Question

I am trying to concatenate LoadIds for a user using For XML path which is a portion of my whole query, I verified that the maximum time gets elapsed in calculating this concatenated LoadId column. Below is the syntax, can anyone suggest a way to rewrite this efficiently?
SELECT
Col1, col2,
LoadIds = STUFF((SELECT ' , ' + CAST([LoadId] AS varchar(5))
FROM Table1 AS t1
WHERE t1.[UserId] = [t2].[UserId]
FOR XML PATH ('')), 1, 2, '')
FROM
Table1 AS t2
GROUP BY
[UserId]

Answer 1

1). I would try to collect the required data in a temp table or table variable, create an index on it and then play with the concatination.
2). FOR XML PATH works good for small sets of records, for large sets I would try a recursion.
declare #T table (
UserId int not null,
RowNumber int not null,
LoadId varchar(5) not null
primary key clustered (UserId, RowNumber)
);
insert into #T
select
UserId,
row_number() over(partition by UserId order by LoadId),
CAST(LoadId AS varchar(5))
from
Table1 ;
with cte (UserId, RowNumber, LoadIds) as
(
select
UserId,
RowNumber,
LoadIds = convert(varchar(8000), LoadId)
from #T
where RowNumber = 1
union all
select
t.UserId,
t.RowNumber,
convert(varchar(8000), cte.LoadIds + ', ' + t.LoadId)
from
cte inner join #T t on t.UserId = cte.UserId and t.RowNumber = cte.RowNumber + 1
)
select UserId, LoadIds = max(LoadIds) from cte group by UserId;

Answer 2

One thing you can do is select the distinct user ids before doing the string aggregation:
SELECT t2.UserId,
STUFF((SELECT ' , ' + CAST([LoadId] AS varchar(8000))
FROM Table1 t1
WHERE t1.[UserId] = [t2].[UserId]
FOR xml PATH ('')
), 1, 2, ''
) as LoadIds
FROM (SELECT DISTINCT userId
FROM Table1 t2
) t2;
For performance, an index on table1(UserId, LoadId) would also help.