Using a view with no primary key with Entity - sql

Question

I just started on a project converting an application from raw ADO.NET and embedded SQL to Entity. I ran in to a problem with one of the views used by the application. The view has no primary key and no column (or combination of columns) that uniquely identifies a row. Here is the select the view is created with:
SELECT
filingmonth,
CEIL(filingmonth / 3),
licnum,
filingyear,
DECODE(GROUPING(insurername), '1', '- All Insured -', insurername),
insurername,
policylinecode,
linedescription,
SUM(NVL(grosspremium, 0)),
SUM(DECODE(taxexempt, 1, grosspremium, 0)),
TRUNC(
CASE
WHEN
(
b.rsn IS NOT NULL
OR A.zeroreport = 1
)
AND b.datereceived IS NULL
THEN A.datereceived
ELSE b.datereceived
END),
SUM(aip.iscompanyadmitted(b.naiccocode, b.naicalienid)),
A.insuredid
FROM
aip.slbtransinsured A
LEFT OUTER JOIN aip.slbtransinsurer b
ON
A.insuredid = b.insuredid
LEFT OUTER JOIN aip.slblinecodes C
ON
b.policylinecode = C.linecode
WHERE
A.submitted = 1
AND A.entryincomplete = 0
GROUP BY
licnum,
filingmonth,
filingyear,
TRUNC(
CASE
WHEN
(
b.rsn IS NOT NULL
OR A.zeroreport = 1
)
AND b.datereceived IS NULL
THEN A.datereceived
ELSE b.datereceived
END),
ROLLUP(insurername, aip.iscompanyadmitted(b.naiccocode, b.naicalienid),
policylinecode, linedescription), A.insuredid;
And here is some sample data showing that there are some rows that are completely duplicated (rows 3 and 4):
FILINGMONTH CEIL(FILINGMONTH/3) LICNUM FILINGYEAR DECODE(GROUPING(INSURERNAME),'1','-ALLINSURED-',INSURERNAME) INSURERNAME POLICYLINECODE LINEDESCRIPTION SUM(NVL(GROSSPREMIUM,0)) SUM(DECODE(TAXEXEMPT,1,GROSSPREMIUM,0)) TRUNC(CASEWHEN(B.RSNISNOTNULLORA.ZEROREPORT=1)ANDB.DATERECEIVEDISNULLTHENA.DATERECEIVEDELSEB.DATERECEIVEDEND) SUM(AIP.ISCOMPANYADMITTED(B.NAICCOCODE,B.NAICALIENID)) INSUREDID
6 2 8150 2007 SAVERS PROPERTY AND CASUALTY INSURANCE CO SAVERS PROPERTY AND CASUALTY INSURANCE CO 17 OTHER LIABILITY 721.25 0 18-JUL-07 0 81
6 2 8150 2007 SAVERS PROPERTY AND CASUALTY INSURANCE CO SAVERS PROPERTY AND CASUALTY INSURANCE CO 17 721.25 0 18-JUL-07 0 81
6 2 8150 2007 SAVERS PROPERTY AND CASUALTY INSURANCE CO SAVERS PROPERTY AND CASUALTY INSURANCE CO 721.25 0 18-JUL-07 0 81
6 2 8150 2007 SAVERS PROPERTY AND CASUALTY INSURANCE CO SAVERS PROPERTY AND CASUALTY INSURANCE CO 721.25 0 18-JUL-07 0 81
insuredid is the pk for the aip.slbtransinsured table, rsn is the pk for aip.slbtransinsurer and aip.slblinecodes.
Is it at all possible to add a view to the Entity model without a unique identifier? Or is there an easy way to add a unique row identifier to the view? The view is only read from, never written to.

Answer 1

Is it at all possible to add a view to the Entity model without a
unique identifier?
If without a primary key, no. That will result to this kind of error:
One or more validation errors were detected during model generation:
System.Data.Edm.EdmEntityType: : EntityType 'SalesOnEachCountry' has
no key defined. Define the key for this EntityType.
System.Data.Edm.EdmEntitySet: EntityType: The EntitySet
SalesOnEachCountryList is based on type SalesOnEachCountry that has no
keys defined.
If without a unique identifier, yes, albeit it has a non-desirable output. Records with same identifier would reference the same object, this is called Identity Map Pattern
An example, even if your view produces these two rows:
Country Year TotalSales
Philippines 2010 20.000000
Philippines 2011 40.000000
If you will just map the primary key on Country field only, e.g.
public class SalesOnEachCountry
{
[Key]
public int CountryId { get; set; }
public string CountryName { get; set; }
public int OrYear { get; set; }
public long SalesCount { get; set; }
public decimal TotalSales { get; set; }
}
, even your view produces the above two rows on your Oracle query editor, Entity Framework produces this incorrect output:
Country Year TotalSales
Philippines 2010 20.000000
Philippines 2010 20.000000
Entity Framework will take it that the second row is same object as first row.
To guarantee uniqueness, you must identify what columns that makes each row unique. In the above example, Year must be included so the primary key is unique. i.e.
public class SalesOnEachCountry
{
[Key, Column(Order=0)] public int CountryId { get; set; }
public string CountryName { get; set; }
[Key, Column(Order=1)] public int OrYear { get; set; }
public long SalesCount { get; set; }
public decimal TotalSales { get; set; }
}
Making your primary key similar to the attributes above, Entity Framework can correctly map your each view's row to their own objects. Hence, Entity Framework can now display exactly the same rows your view have.
Country Year TotalSales
Philippines 2010 20.000000
Philippines 2011 40.000000
Full details here: http://www.ienablemuch.com/2011/06/mapping-class-to-database-view-with.html
Then regarding your views which don't have any columns to make a row unique, the easiest way to guarantee Entity Framework can map each of your view's row to their own objects is to create a separate column for your view's primary key, a good candidate is to just create a row number column on each row. e.g.
create view RowNumberedView as
select
row_number() over(order by <columns of your view sorting>) as RN
, *
from your_existing_view
Then assign the [Key] attribute on RN property of your class RowNumberedView

Answer 2

Expanding on the answer from Michael Buen:
I found that adding the row number to the view with an ISNULL() will allow the entity framework to pull the view in and create the necessary EntitySet data automatically.
create view RowNumberedView as
select
ISNULL(ROW_NUMBER() OVER (ORDER BY <column>), 0) AS RN
, *
from your_existing_view

Answer 3

At work recently, i ran into this same issue. Based on my research, I could not find any answers on how to attach a view to EF6 CodeFirst without a PK. Most seem to involve migrations and were quite confusing. I believe DB first has better support for working SQL VIEWS.
I did try introducing a window function (RowNumber) with the idea being, to use the row identifier as the PK to keep EF6 Happy. But this made my query more expensive overall so i had to drop this idea.
In the end, I had to carefully analyse my data set to see if i could introduce a composite key - one that covers all of the scenarios my business application needed to ensure would work. Remember to use IsNull(ColumnName,0) too to ensure you can satisfy the .IsRequired() in CodeFirst fluent methods.
i.e
HasKey(x => new { x.KfiId, x.ApplicationNumber, x.CustomerId });
I hope this helps someone - the answer for me was analyse the dataset the view denormalises and look for a composite key.
Another cool idea you can try suggested by Marc Cals.

Answer 4

In case you are using Entity Framework with MVC in ASP.NET
As previously said, create your view with column that has auto-increment or ROW_NUMBER.
Let's say you have that column and it's name is rowNumber.
an than go to context file (yourDatabaseNameContext) file in your Models directory of your MVC application, find definition for your view and instead of
modelBuilder.Entity<yourView>(entity =>
{
entity.HasNoKey();
change it to:
modelBuilder.Entity<yourView>(entity =>
{
entity.HasKey(e => e.rowNumber);

Answer 5

Is it at all possible to add a view to the Entity model without a unique identifier?
It's possible to have a views where there isn't a single column or set of columns that create the primary key; thus, you end up with spurious relationships. Data warehouse tables sometimes follow that form. In short, normalization is sometimes not followed for either performance reasons or reporting reasons.
Now to your second point:
Or is there an easy way to add a unique row identifier to the view?
What i suggest you do is to select all the columns from slbtransinsured and see if you can find the one column that uniquely identifies each record. It seems to me that the data is there should be a code type in slblinecodes that you need to select, somewhat like a lookup.
For kicks, try running this and tell me what you get:
SELECT filingmonth,
CEIL (filingmonth / 3),
licnum,
filingyear,
DECODE (GROUPING (insurername), '1', '- All Insured -', insurername),
insurername,
policylinecode,
linedescription,
SUM (NVL (grosspremium, 0)),
SUM (DECODE (taxexempt, 1, grosspremium, 0)),
TRUNC (
CASE
WHEN (b.rsn IS NOT NULL OR a.zeroreport = 1)
AND b.datereceived IS NULL
THEN
a.datereceived
ELSE
b.datereceived
END),
SUM (aip.iscompanyadmitted (b.naiccocode, b.naicalienid)),
a.insuredid
FROM aip.slbtransinsured a
LEFT OUTER JOIN aip.slbtransinsurer b
ON a.insuredid = b.insuredid
LEFT OUTER JOIN aip.slblinecodes c
ON b.policylinecode = c.linecode
WHERE a.submitted = 1 AND a.entryincomplete = 0
GROUP BY filingmonth,
licnum,
filingyear,
DECODE (GROUPING (insurername), '1', '- All Insured -', insurername),
insurername,
policylinecode,
linedescription,
TRUNC (
CASE
WHEN (b.rsn IS NOT NULL OR a.zeroreport = 1)
AND b.datereceived IS NULL
THEN
a.datereceived
ELSE
b.datereceived
END),
a.insuredid;

Answer 6

Consider using AsNoTracking() when consuming the view. This disables the usage of any key fields for EF tracking. Then any non-null field can be defined manually as the EF key (even if it repeats).
It is advised not to create an additional row counter field because most row counters end up requiring that the engine scan the entire domain of the view to generate the proper counter values even if querying using a predicate (where clause).
Please see link.