NHibernate SetFirstResult causes duplicate results - sql

Question

We are having trouble with the way NHibernate (version 4.0.0.4000 AND 4.0.4.4000 tested) returns duplicate results. In the sample below, I get 566 results (the correct number of results), but only 549 are unique, meaning there are 17 duplicates.
#region Get Record IDs
public IList<string> GetRecordIds(string user, string agency, DateTime utcFrom, DateTime utcTo, SearchDateRangeType dateRangeType, IEnumerable<string> status, IEnumerable<string> billingStatus, IEnumerable<string> qaStatus, IEnumerable<string> transmissionStatus, IEnumerable<string> scheduledTransmissions, int pageSize = -1, int pageNumber = -1)
{
using (ISession session = NHibernateHelper.OpenSession())
{
ICriteria crit = session.CreateCriteria<Metadata>();
var dateDisjunction = Restrictions.Disjunction();
dateDisjunction.Add(Restrictions.Between("IncidentDate", utcFrom, utcTo));
crit.Add(dateDisjunction);
if (string.IsNullOrEmpty(agency) == false)
{
crit.CreateAlias("Ownership._entities.AsIList", "entities");
crit.Add(Restrictions.Eq("entities._entityName._value", agency));
crit.Add(Restrictions.Eq("entities._isDeleted._value", false) || Restrictions.IsNull("entities._isDeleted._value"));
}
crit.AddOrder(Order.Asc(Projections.Property("RecordId")));
crit.SetProjection(Projections.Property("RecordId"));
if (pageSize > 0 && pageNumber > 0)
{
crit.SetFirstResult(pageSize * (pageNumber - 1)).SetMaxResults(pageSize);
}
var ret = crit.List<string>();
return ret;
}
}
#endregion
SQL Sample 1 is the generated first iteration code from NHibernate. Subsequent pages (second page onward) use ROW_NUMBER() OVER. SQL Sample 2 is a manually-created first page, which uses ROW_NUMBER() OVER as if it was a subsequent page. NHibernate has apparently "optimized" away the ROW_NUMBER() OVER for the first page and that seems(?) to be the cause of our issues.
SQL Sample 1: Generated by NHibernate. Causes duplicates.
SELECT
TOP (100) this_.RecordId as y0_
FROM
PcrMetadata this_
inner join
PcrEntities entities1_
on this_.Id=entities1_.ListKey
WHERE
(
this_.IncidentDate between '0001-01-01 00:00:00.0000000' and '9999-01-01 00:00:00.0000000'
)
and entities1_.Name = 'ClientIDNumber'
and (
entities1_.Entities_IsDeleted = 0
or entities1_.Entities_IsDeleted is null
)
SQL Sample 2: Manually created based on NHibernate second page on. Does not cause duplicates.
SELECT
TOP (100) this_.RecordId as y0_
FROM
(SELECT
this_.Record as y0_,
ROW_NUMBER() OVER(
ORDER BY
CURRENT_TIMESTAMP) as __hibernate_sort_row
FROM
PcrMetadata this_
inner join
PcrEntities entities1_
on this_.Id=entities1_.ListKey
WHERE
(
this_.IncidentDate between '0001-01-01 00:00:00.0000000' and '9999-01-01 00:00:00.0000000'
)
and entities1_.Name = 'ClientIDNumber'
and (
entities1_.Entities_IsDeleted = 0
or entities1_.Entities_IsDeleted is null
)) as query
WHERE
query.__hibernate_sort_row > 0 -- CHANGE THIS NUMBER
Am I doing something wrong? Or there anything I can do to force NHibernate to use ROW_NUMBER?
Thanks in advance for any help!

Answer 1

We cannot JOIN collections and apply paging. Because we are getting cartesian product, which is paged (experience described above).
The solution I would suggest, is to (my way NEVER) join collection. To get the similar results, we should:
use subquery to apply WHERE
use fetch batching to later recieve all collection items without 1 + N issue
There is detailed answer about this issue.
see also:
How to Eager Load Associations without duplication in NHibernate?
What is the solution for the N+1 issue in hibernate?
There is more about making result distinct, but this could not help here:
Criteria.DISTINCT_ROOT_ENTITY vs Projections.distinct