T-SQL SELECT TOP returns duplicates - sql

Question

I'm using SQL Server 2008 R2.
I'm not sure if I've discovered a strange SQL quirk, or (more likely) something in my code is causing this strange behaviour, particularly as Google has turned up nothing. I have a view called vwResponsible_Office_Address.
SELECT * FROM vwResponsible_Office_Address
..returns 403 rows
This code:
SELECT TOP 1000 * FROM vwResponsible_Office_Address
..returns 409 rows, as it includes 6 duplicates.
However this:
SELECT TOP 1000 * FROM vwResponsible_Office_Address
ORDER BY ID
..returns 403 rows again.
I can post the code for the view if it's relevant, but does it make sense for SELECT TOP to ever work in this way? I understand that SELECT TOP is free to return records in any order but don't understand why the number of records returned should vary.
The view does use cross apply which might be affecting the result set some how?
EDIT: View definition as requested
CREATE VIEW [dbo].[vwResponsible_Office_Address]
AS
SELECT fp.Entity_ID [Reg_Office_Entity_ID],
fp.Entity_Name [Reg_Office_Entity_Name],
addr.Address_ID
FROM [dbo].[Entity_Relationship] er
INNER JOIN [dbo].[Entity] fp
ON er.[Related_Entity_ID] = fp.[Entity_ID]
INNER JOIN [dbo].[Entity_Address] ea
ON ea.[Entity_ID] = fp.[Entity_ID]
CROSS APPLY (
SELECT TOP 1 Address_ID
FROM [dbo].[vwEntity_Address] vea
WHERE [vea].[Entity_ID] = fp.Entity_ID
ORDER by ea.[Address_Type_ID] ASC, ea.[Address_ID] DESC
) addr
WHERE [Entity_Relationship_Type_ID] = 25 -- fee payment relationship
UNION
SELECT ets.[Entity_ID],
ets.[Entity_Name],
addr.[Address_ID]
FROM dbo.[vwEntity_Entitlement_Status] ets
INNER JOIN dbo.[Entity_Address] ea
ON ea.[Entity_ID] = ets.[Entity_ID]
CROSS APPLY (
SELECT TOP 1 [Address_ID]
FROM [dbo].[vwEntity_Address] vea
WHERE vea.[Entity_ID] = ets.[Entity_ID]
ORDER by ea.[Address_Type_ID] ASC, ea.[Address_ID] DESC
) addr
WHERE ets.[Entitlement_Type_ID] = 40 -- registered office
AND ets.[Entitlement_Status_ID] = 11 -- active

Answer 1

I would assume that there is some non determinism going on which means that different access methods can return different results.
Looking at the view definition the only place that appears likely would be if vwEntity_Address has some duplicates for Entity_ID.
This would make the top 1 Address_ID returned arbitrary in that case which will effect the result of the union operation when it removes duplicates.
Definitely this does look extremely suspect
SELECT TOP 1 [Address_ID]
FROM [dbo].[vwEntity_Address] vea
WHERE vea.[Entity_ID] = ets.[Entity_ID]
ORDER by ea.[Address_Type_ID] ASC, ea.[Address_ID] DESC
You are ordering by values from the outer query in the cross apply. This will have absolutely no effect whatsoever as these will be constant for a particular CROSS APPLY invocation.
Can you try changing to
SELECT TOP 1 [Address_ID]
FROM [dbo].[vwEntity_Address] vea
WHERE vea.[Entity_ID] = ets.[Entity_ID]
ORDER by vea.[Address_ID] DESC

Answer 2

I was wondering if your view included a function, until I got to the end, where you say you use cross-apply. I would assume that is your problem, if your interested in the details, take a look at the various query plans.
EDIT: Expansion of answer
I.e. your function is non-deterministic and can either return more than one row per input or return the same row for different input. In combination, this means that you'll get exactly what you are seeing: duplicate rows under some circumsntaces. Adding a distinct to your view is the costly way to solve your problem, a better way would be to change your function so that for any input there is only one row output, and for a row output only one input will produce that row.
EDIT: I didn't see that you're now including your view definition.
Your problem is definitely the cross apply, in particular you are sorting inside the cross apply by values from OUTSIDE of the cross apply, making the top 1 effectively random.