Skip to content

Commit

Permalink
Document TPC
Browse files Browse the repository at this point in the history
Fixes #3764
  • Loading branch information
AndriySvyryd committed Oct 11, 2022
1 parent 8261728 commit c4e107a
Show file tree
Hide file tree
Showing 12 changed files with 392 additions and 33 deletions.
1 change: 1 addition & 0 deletions .gitignore
Original file line number Diff line number Diff line change
Expand Up @@ -12,6 +12,7 @@

# SQLite databases
*.sqlite
*.db

# User-specific files (MonoDevelop/Xamarin Studio)
*.userprefs
Expand Down
2 changes: 1 addition & 1 deletion entity-framework/core/modeling/generated-properties.md
Original file line number Diff line number Diff line change
Expand Up @@ -39,7 +39,7 @@ The above creates a *virtual* computed column, whose value is computed every tim

By convention, non-composite primary keys of type short, int, long, or Guid are set up to have values generated for inserted entities if a value isn't provided by the application. Your database provider typically takes care of the necessary configuration; for example, a numeric primary key in SQL Server is automatically set up to be an IDENTITY column.

For more information, [see the documentation about keys](xref:core/modeling/keys).
For more information, [see the documentation about keys](xref:core/modeling/keys) and [guidance for specific inheritance mapping strategies](xref:core/modeling/inheritance#key-generation).

## Explicitly configuring value generation

Expand Down
261 changes: 259 additions & 2 deletions entity-framework/core/modeling/inheritance.md
Original file line number Diff line number Diff line change
Expand Up @@ -2,7 +2,7 @@
title: Inheritance - EF Core
description: How to configure entity type inheritance using Entity Framework Core
author: AndriySvyryd
ms.date: 10/01/2020
ms.date: 10/10/2022
uid: core/modeling/inheritance
---
# Inheritance
Expand Down Expand Up @@ -61,12 +61,15 @@ By default, when two sibling entity types in the hierarchy have a property with
## Table-per-type configuration

> [!NOTE]
> The table-per-type (TPT) feature was introduced in EF Core 5.0. Table-per-concrete-type (TPC) is supported by EF6, but is not yet supported by EF Core.
> The table-per-type (TPT) feature was introduced in EF Core 5.0.
In the TPT mapping pattern, all the types are mapped to individual tables. Properties that belong solely to a base type or derived type are stored in a table that maps to that type. Tables that map to derived types also store a foreign key that joins the derived table with the base table.

[!code-csharp[Main](../../../samples/core/Modeling/Inheritance/FluentAPI/TPTConfiguration.cs?name=TPTConfiguration)]

> [!TIP]
> Instead of calling `ToTable` on each entity type you can call `modelBuilder.Entity<Blog>().UseTptMappingStrategy()` on each root entity type and the table names will be generated by EF.
EF will create the following database schema for the model above.

```sql
Expand All @@ -93,3 +96,257 @@ If you are employing bulk configuration you can retrieve the column name for a s

> [!WARNING]
> In many cases, TPT shows inferior performance when compared to TPH. [See the performance docs for more information](xref:core/performance/modeling-for-performance#inheritance-mapping).
## Table-per-concrete-type configuration

> [!NOTE]
> The table-per-concrete-type (TPC) feature was introduced in EF Core 7.0.
In the TPC mapping pattern, all the types are mapped to individual tables. Each table contains columns for all properties on the corresponding entity type. This addresses some common performance issues with the TPT strategy.

> [!TIP]
> The EF Team demonstrated and talked in depth about TPC mapping in an episode of the .NET Data Community Standup. As with [all Community Standup episodes](https://aka.ms/efstandups), you can [watch the TPC episode now on YouTube](https://youtu.be/HaL6DKW1mrg).
[!code-csharp[Main](../../../samples/core/Modeling/Inheritance/FluentAPI/TPCConfiguration.cs?name=TPCConfiguration)]

> [!TIP]
> Instead of calling `ToTable` on each entity type just calling `modelBuilder.Entity<Blog>().UseTpcMappingStrategy()` on each root entity type will generate the table names by convention.
EF will create the following database schema for the model above.

```sql
CREATE TABLE [Blogs] (
[BlogId] int NOT NULL DEFAULT (NEXT VALUE FOR [BlogSequence]),
[Url] nvarchar(max) NULL,
CONSTRAINT [PK_Blogs] PRIMARY KEY ([BlogId])
);

CREATE TABLE [RssBlogs] (
[BlogId] int NOT NULL DEFAULT (NEXT VALUE FOR [BlogSequence]),
[Url] nvarchar(max) NULL,
[RssUrl] nvarchar(max) NULL,
CONSTRAINT [PK_RssBlogs] PRIMARY KEY ([BlogId])
);
```

### TPC database schema

The TPC strategy is similar to the TPT strategy except that a different table is created for every _concrete_ type in the hierarchy, but tables are **not** created for _abstract_ types - hence the name “table-per-concrete-type”. As with TPT, the table itself indicates the type of the object saved. However, unlike TPT mapping, each table contains columns for every property in the concrete type and its base types. TPC database schemas are denormalized.

For example, consider mapping this hierarchy:

<!--
public abstract class Animal
{
protected Animal(string name)
{
Name = name;
}
public int Id { get; set; }
public string Name { get; set; }
public abstract string Species { get; }
public Food? Food { get; set; }
}
public abstract class Pet : Animal
{
protected Pet(string name)
: base(name)
{
}
public string? Vet { get; set; }
public ICollection<Human> Humans { get; } = new List<Human>();
}
public class FarmAnimal : Animal
{
public FarmAnimal(string name, string species)
: base(name)
{
Species = species;
}
public override string Species { get; }
[Precision(18, 2)]
public decimal Value { get; set; }
public override string ToString()
=> $"Farm animal '{Name}' ({Species}/{Id}) worth {Value:C} eats {Food?.ToString() ?? "<Unknown>"}";
}
public class Cat : Pet
{
public Cat(string name, string educationLevel)
: base(name)
{
EducationLevel = educationLevel;
}
public string EducationLevel { get; set; }
public override string Species => "Felis catus";
public override string ToString()
=> $"Cat '{Name}' ({Species}/{Id}) with education '{EducationLevel}' eats {Food?.ToString() ?? "<Unknown>"}";
}
public class Dog : Pet
{
public Dog(string name, string favoriteToy)
: base(name)
{
FavoriteToy = favoriteToy;
}
public string FavoriteToy { get; set; }
public override string Species => "Canis familiaris";
public override string ToString()
=> $"Dog '{Name}' ({Species}/{Id}) with favorite toy '{FavoriteToy}' eats {Food?.ToString() ?? "<Unknown>"}";
}
public class Human : Animal
{
public Human(string name)
: base(name)
{
}
public override string Species => "Homo sapiens";
public Animal? FavoriteAnimal { get; set; }
public ICollection<Pet> Pets { get; } = new List<Pet>();
public override string ToString()
=> $"Human '{Name}' ({Species}/{Id}) with favorite animal '{FavoriteAnimal?.Name ?? "<Unknown>"}'" +
$" eats {Food?.ToString() ?? "<Unknown>"}";
}
-->
[!code-csharp[AnimalsHierarchy](../../../samples/core/Miscellaneous/NewInEFCore7/TpcInheritanceSample.cs?name=AnimalsHierarchy)]

When using SQL Server, the tables created for this hierarchy are:

```sql
CREATE TABLE [Cats] (
[Id] int NOT NULL DEFAULT (NEXT VALUE FOR [AnimalSequence]),
[Name] nvarchar(max) NOT NULL,
[FoodId] uniqueidentifier NULL,
[Vet] nvarchar(max) NULL,
[EducationLevel] nvarchar(max) NOT NULL,
CONSTRAINT [PK_Cats] PRIMARY KEY ([Id]));

CREATE TABLE [Dogs] (
[Id] int NOT NULL DEFAULT (NEXT VALUE FOR [AnimalSequence]),
[Name] nvarchar(max) NOT NULL,
[FoodId] uniqueidentifier NULL,
[Vet] nvarchar(max) NULL,
[FavoriteToy] nvarchar(max) NOT NULL,
CONSTRAINT [PK_Dogs] PRIMARY KEY ([Id]));

CREATE TABLE [FarmAnimals] (
[Id] int NOT NULL DEFAULT (NEXT VALUE FOR [AnimalSequence]),
[Name] nvarchar(max) NOT NULL,
[FoodId] uniqueidentifier NULL,
[Value] decimal(18,2) NOT NULL,
[Species] nvarchar(max) NOT NULL,
CONSTRAINT [PK_FarmAnimals] PRIMARY KEY ([Id]));

CREATE TABLE [Humans] (
[Id] int NOT NULL DEFAULT (NEXT VALUE FOR [AnimalSequence]),
[Name] nvarchar(max) NOT NULL,
[FoodId] uniqueidentifier NULL,
[FavoriteAnimalId] int NULL,
CONSTRAINT [PK_Humans] PRIMARY KEY ([Id]));
```

Notice that:

- There are no tables for the `Animal` or `Pet` types, since these are `abstract` in the object model. Remember that C# does not allow instances of abstract types, and there is therefore no situation where an abstract type instance will be saved to the database.
- The mapping of properties in base types is repeated for each concrete type. For example, every table has a `Name` column, and both Cats and Dogs have a `Vet` column.

- Saving some data into this database results in the following:

**Cats table**

| Id | Name | FoodId | Vet | EducationLevel |
|:----|:-------|:-------------------------------------|:---------------------|:---------------|
| 1 | Alice | 99ca3e98-b26d-4a0c-d4ae-08da7aca624f | Pengelly | MBA |
| 2 | Mac | 99ca3e98-b26d-4a0c-d4ae-08da7aca624f | Pengelly | Preschool |
| 8 | Baxter | 5dc5019e-6f72-454b-d4b0-08da7aca624f | Bothell Pet Hospital | BSc |

**Dogs table**

| Id | Name | FoodId | Vet | FavoriteToy |
|:----|:------|:-------------------------------------|:---------|:-------------|
| 3 | Toast | 011aaf6f-d588-4fad-d4ac-08da7aca624f | Pengelly | Mr. Squirrel |

**FarmAnimals table**

| Id | Name | FoodId | Value | Species |
|:----|:------|:-------------------------------------|:-------|:-----------------------|
| 4 | Clyde | 1d495075-f527-4498-d4af-08da7aca624f | 100.00 | Equus africanus asinus |

**Humans table**

| Id | Name | FoodId | FavoriteAnimalId |
|:----|:-------|:-------------------------------------|:----------------------|
| 5 | Wendy | 5418fd81-7660-432f-d4b1-08da7aca624f | 2 |
| 6 | Arthur | 59b495d4-0414-46bf-d4ad-08da7aca624f | 1 |
| 9 | Katie | null | 8 |

Notice that, unlike with TPT mapping, all the information for a single object is contained in a single table. And, unlike with TPH mapping, there is no combination of column and row in any table where that is never used by the model. We'll see below how these characteristics can be important for queries and storage.

## Key generation

The inheritance mapping strategy chosen has consequences for how primary key values are generated and managed. Keys in TPH are easy, since each entity instance is represented by a single row in a single table. Any kind of key value generation can be used, and no additional constraints are needed.

For the TPT strategy, there is always a row in the table mapped to the base type of the hierarchy. Any kind of key generation can be used on this row, and the keys for other tables are linked to this table using foreign key constraints.

Things get a bit more complicated for TPC. First, it’s important to understand that EF Core requires that all entities in a hierarchy must have a unique key value, even if the entities have different types. So, using our example model, a Dog cannot have the same Id key value as a Cat. Second, unlike TPT, there is no common table that can act as the single place where key values live and can be generated. This means a simple `Identity` column cannot be used.

For databases that support sequences, key values can be generated by using a single sequence referenced in the default constraint for each table. This is the strategy used in the TPC tables shown above, where each table has the following:

```sql
[Id] int NOT NULL DEFAULT (NEXT VALUE FOR [AnimalSequence])
```

`AnimalSequence` is a database sequence created by EF Core. This strategy is used by default for TPC hierarchies when using the EF Core database provider for SQL Server. Database providers for other databases that support sequences should have a similar default. Other key generation strategies that use sequences, such as Hi-Lo patterns, may also be used with TPC.

While standard Identity columns will not work with TPC, it is possible to use Identity columns if each table is configured with an appropriate seed and increment such that the values generated for each table will never conflict. For example:

<!--
modelBuilder.Entity<Cat>().ToTable("Cats", tb => tb.Property(e => e.Id).UseIdentityColumn(1, 4));
modelBuilder.Entity<Dog>().ToTable("Dogs", tb => tb.Property(e => e.Id).UseIdentityColumn(2, 4));
modelBuilder.Entity<FarmAnimal>().ToTable("FarmAnimals", tb => tb.Property(e => e.Id).UseIdentityColumn(3, 4));
modelBuilder.Entity<Human>().ToTable("Humans", tb => tb.Property(e => e.Id).UseIdentityColumn(4, 4));
-->
[!code-csharp[UsingIdentity](../../../../samples/core/Miscellaneous/NewInEFCore7/TpcInheritanceSample.cs?name=UsingIdentity)]

SQLite does not support sequences or Identity seed/increment, and hence integer key value generation is not supported when using SQLite with the TPC strategy. However, client-side generation or globally unique keys; for example, GUID keys are supported on any database, including SQLite.

## Foreign key constraints

The TPC mapping strategy creates a denormalized SQL schema - this is one reason why some database purists are against it. For example, consider the foreign key column `FavoriteAnimalId`. The value in this column must match the primary key value of some animal. This can be enforced in the database with a simple FK constraint when using TPH or TPT. For example:

```sql
CONSTRAINT [FK_Animals_Animals_FavoriteAnimalId] FOREIGN KEY ([FavoriteAnimalId]) REFERENCES [Animals] ([Id])
```

But when using TPC, the primary key for any given animal is stored in the table corresponding to the concrete type of that animal. For example, a cat's primary key is stored in the `Cats.Id` column, while a dog's primary key is stored in the `Dogs.Id` column, and so on. This means an FK constraint cannot be created for this relationship.

In practice, this is not a problem as long as the application does not attempt to insert invalid data. For example, if all the data is inserted by EF Core and uses navigations to relate entities, then it is guaranteed that the FK column will contain valid PK values at all times.

> [!NOTE]
> Also see [stored computed column](xref:core/modeling/table-splitting#computed-columns)
## Summary and guidance

In summary, TPC is a good mapping strategy to use when your code will mostly query for entities of a single leaf type. This is because the storage requirements are smaller, and there is no discriminator column that may need an index. Inserts and updates are also efficient. It also makes it easy to evolve the model as changes to a given type don't need to affect tables corresponding to the derived types.

That being said, TPH is usually fine for most applications, and is a good default for a wide range of scenarios, so don't add the complexity of TPC if you don't need it. Specifically, if your code will mostly query for entities of many types, such as writing queries against the base type, then lean towards TPH over TPC.

Use TPT only if constrained to do so by external factors.
2 changes: 1 addition & 1 deletion entity-framework/core/modeling/keys.md
Original file line number Diff line number Diff line change
Expand Up @@ -36,7 +36,7 @@ You can also configure multiple properties to be the key of an entity - this is

## Value generation

For non-composite numeric and GUID primary keys, EF Core sets up value generation for you by convention. For example, a numeric primary key in SQL Server is automatically set up to be an IDENTITY column. For more information, see [the documentation on value generation](xref:core/modeling/generated-properties).
For non-composite numeric and GUID primary keys, EF Core sets up value generation for you by convention. For example, a numeric primary key in SQL Server is automatically set up to be an IDENTITY column. For more information, see [the documentation on value generation](xref:core/modeling/generated-properties) and [guidance for specific inheritance mapping strategies](xref:core/modeling/inheritance#key-generation).

## Primary key name

Expand Down
2 changes: 1 addition & 1 deletion entity-framework/core/modeling/owned-entities.md
Original file line number Diff line number Diff line change
Expand Up @@ -33,7 +33,7 @@ If the `ShippingAddress` property is private in the `Order` type, you can use th

The model above is mapped to the following database schema:

![Sceenshot of the database model for entity containing owned reference](_static/owned-entities-ownsone.png)
![Screenshot of the database model for entity containing owned reference](_static/owned-entities-ownsone.png)

See the [full sample project](https://github.com/dotnet/EntityFramework.Docs/tree/main/samples/core/Modeling/OwnedEntities) for more context.

Expand Down
3 changes: 3 additions & 0 deletions entity-framework/core/modeling/relationships.md
Original file line number Diff line number Diff line change
Expand Up @@ -146,6 +146,9 @@ After the navigation property has been created, you may need to further configur

[!code-csharp[Main](../../../samples/core/Modeling/Relationships/FluentAPI/NavigationConfiguration.cs?name=NavigationConfiguration&highlight=7-9)]

> [!TIP]
> The non-collection navigations can also be marked as required, see [Required one-to-one dependents](xref:core/modeling/relationships#one-to-one) for more information.
> [!NOTE]
> This call cannot be used to create a navigation property. It is only used to configure a navigation property which has been previously created by defining a relationship or from a convention.
Expand Down
14 changes: 13 additions & 1 deletion entity-framework/core/modeling/table-splitting.md
Original file line number Diff line number Diff line change
Expand Up @@ -2,7 +2,7 @@
title: Table Splitting - EF Core
description: How to configure table splitting using Entity Framework Core
author: AndriySvyryd
ms.date: 11/15/2021
ms.date: 10/10/2022
uid: core/modeling/table-splitting
---
# Table Splitting
Expand Down Expand Up @@ -38,10 +38,22 @@ Saving and querying entities using table splitting is done in the same way as ot

If all of the columns used by a dependent entity are `NULL` in the database, then no instance for it will be created when queried. This allows modeling an optional dependent entity, where the relationship property on the principal would be null. Note that this would also happen if all of the dependent's properties are optional and set to `null`, which might not be expected.

However, the additional check can impact the performance on queries. And if the dependent entity type has dependents of its own then determining whether an instance should be created becomes non-trivial. To avoid these issues the dependent entity type can be marked as required, see [Required one-to-one dependents](xref:core/modeling/relationships#one-to-one) for more information.

## Concurrency tokens

If any of the entity types sharing a table has a concurrency token then it must be included in all other entity types as well. This is necessary in order to avoid a stale concurrency token value when only one of the entities mapped to the same table is updated.

To avoid exposing the concurrency token to the consuming code, it's possible the create one as a [shadow property](xref:core/modeling/shadow-properties):

[!code-csharp[TableSplittingConfiguration](../../../samples/core/Modeling/TableSplitting/TableSplittingContext.cs?name=ConcurrencyToken&highlight=2)]

## Inheritance

It's recommended to read [the dedicated page on inheritance](xref:core/modeling/inheritance) before continuing with this section.

The dependent types using table splitting can have an inheritance hierarchy, but there are some limitations:

- The dependent entity type __cannot__ use TPC mapping as the derived types wouldn't be able to map to the same table.
- The dependent entity type __can__ use TPT mapping, but only the root entity type can use table splitting.
- If the principal entity type uses TPC then only the entity types that don't have any descendants can use table splitting. Otherwise the dependent columns would need to be duplicated on the tables corresponding to the derived types, complicating all interactions.
Loading

0 comments on commit c4e107a

Please sign in to comment.