Select all rows for the first N distinct child table rows - sql-server

Question

It seems like this should be a common problem with a simple solution, but I haven't found it.
I would like to compute child_order, which is the order of appearance of distinct child table rows as is shown below in the following data:
child_order PK1 PK2 ACCESS ACCESS_ID
1 99 Al NULL NULL
2 55 Charles Accounts 1
2 55 Charles Desktop 2
2 55 Charles Printer 3
2 55 Charles Servers 4
2 55 Charles VMs 5
3 66 Charles Desktop 2
3 66 Charles VMs 5
4 22 Chris Desktop 2
4 22 Chris Printer 3
4 22 Chris Servers 4
5 89 Evan Desktop 2
Retrieved by a query like:
SELECT sub1.*
FROM (
SELECT ??? as child_order, sub2.*
FROM (
SELECT ct.PK1, ct.PK2, pt1.ACCESS, pt1.ACCESS_ID
FROM child_table ct
LEFT JOIN some_linktable lt ON lt.child_id = ct.id
LEFT JOIN parent_table1 pt1 ON lt.parent_id = pt1.id
WHERE ct.PK2 IN ('Charles', 'Evan', 'Al', 'Chris')
ORDER BY ct.PK2, pt1.ACCESS -- Order must be preserved
) sub2
) sub1
WHERE child_order < 10 AND (other_conditions)
I can use subqueries, aggregates, analytics, etc. but not really CTEs/"WITH" statements or temporary tables because of the complexity of generating SQL for them dynamically.
Specifically, I am generating pagination SQL (for several DBMSs) for search results from a query joining several tables.
I am trying to figure out how to simply show the top N rows, not counting repeats due to a join (e.g. Chris counts as only one row. Access shows "Desktop, Printer, Servers").
I've tried DENSE_RANK() OVER (ORDER BY PK1, PK2), but of course I get ranking in PK1 PK2 order, which is useless for the WHERE clause. Al, for example, would get a value higher than 1.
I've tried DENSE_RANK() OVER (ORDER BY PK2, ACCESS), but it enumerates only the search terms, not the child table rows.
I've tried DENSE_RANK() OVER (PARTITION BY PK2, ACCESS ORDER BY (SELECT NULL)) (to get DENSE_RANK to use the row order it is given, which is how I want to rank values) but only "1" is returned.
I'll omit my other "try random stuff"-phase attempts.
I would like to avoid having a SELECT DISTINCT PK1, PK2 WHERE (search) ORDER BY (sortorder) subquery because there may be zero or very many primary key fields so dynamic SQL generation would be tricky and, additionally, I suspect the performance would suck with all the WHERE sub3.field1 = sub2.field1 AND sub3.field2 = sub2.field2... checks.

Answer 1

Despite your misgivings about SELECT DISTINCT, this might be the best choice for a sub-query:
SELECT row_number() OVER (ORDER BY PK2) AS child_order, PK1, PK2
FROM (
SELECT DISTINCT PK1, PK2
FROM child_table
WHERE PK2 IN ('Charles', 'Evan', 'Al', 'Chris')
ORDER BY PK2
LIMIT 9) sub2
The child_order field depends only on table child_table and you want only 9 rows of them, so compute the child_order in a sub-query on that table only. After you have that you can join to the other tables. If you have an index on child_table(PK1, PK2) this should be a very fast index-only search. It takes some of the filtering and the limiting inside, so the enveloping query is much simpler:
SELECT sub1.child_order, PK1, PK2, pt1.ACCESS, pt1.ACCESS_ID
FROM child_table ct
JOIN (
SELECT row_number() OVER (ORDER BY PK2) AS child_order, PK1, PK2
FROM (
SELECT DISTINCT PK1, PK2
FROM child_table
WHERE PK2 IN ('Charles', 'Evan', 'Al', 'Chris')
ORDER BY PK2
LIMIT 9) sub2
) sub1 USING (PK1, PK2)
LEFT JOIN some_linktable lt ON lt.child_id = ct.id
LEFT JOIN parent_table1 pt1 ON lt.parent_id = pt1.id
WHERE <other conditions>
ORDER BY sub1.child_order, pt1.ACCESS; -- Faster to order by int