There are many questions about seeding many-to-many relationships in Entity Framework. However, most of them are extremely old, and many-to-many behavior has changed significantly in EFCore5. The official docs recommend overriding OnModelCreating to implement ModelBuilder.Entity<>.HasData().
However, with the new many-to-many behavior (without explicit mappings), I can find no clear path to seed the intermediate tables. To use the example of this tutorial, the BookCategories class is now implicit. Therefore, there is no path to explicitly declare the intermediate table values while seeding.
I've also tried simply assigning the arrays, e.g.,:
public class Book
{
public int BookId { get; set; }
public string Title { get; set; }
public ICollection<Category> Categories { get; set; }
}
public class Category
{
public int CategoryId { get; set; }
public string CategoryName { get; set; }
public ICollection<Book> Books { get; set; }
}
And then at seed time:
Book book = new Book() { BookId = 1, Title = "Brave New World" }
Category category = new Category() { CategoryId = 1, CategoryName = "Dystopian" }
category.Books = new List<Book>() { book };
book.Categories = new List<Category>() { category };
modelBuilder.Entity<Book>().HasData(book);
modelBuilder.Entity<Category>().HasData(category);
... but there are no entries created for BookCategories in the resulting migration. This was somewhat expected, as this article suggests that one must explicitly seed the intermediate table. What I want is something like this:
modelBuilder.Entity<BookCategory>().HasData(
new BookCategory() { BookId = 1, CategoryId = 1 }
);
However, again, since there is no concrete class to describe BookCategories in EFCore5, the only way I can think of to seed the table is to manually edit the migration with additional MigrationBuilder.InsertData commands, which rather defeats the purpose of seeding data via application code.
However, again, since there is no concrete class to describe BookCategories in EFCore5
Actually, as explained in the What's new link, EF Core 5 allows you to have explicit join entity
public class BookCategory
{
public int BookId { get; set; }
public EBook Book { get; set; }
public int CategoryId { get; set; }
public Category Category { get; set; }
}
and configure the many-to-many relationship to use it
modelBuilder.Entity<Book>()
.HasMany(left => left.Categories)
.WithMany(right => right.Books)
.UsingEntity<BookCategory>(
right => right.HasOne(e => e.Category).WithMany(),
left => left.HasOne(e => e.Book).WithMany().HasForeignKey(e => e.BookId),
join => join.ToTable("BookCategories")
);
This way you can use all normal entity operations (query, change tracking, data model seeding etc.) with it
modelBuilder.Entity<BookCategory>().HasData(
new BookCategory() { BookId = 1, CategoryId = 1 }
);
still having the new many-to-many skip navigations mapping.
This is probably the simplest as well as the type-safe approach.
In case you thing it's too much, using the conventional join entity is also possible, but you need to know the shared dictionary entity type name, as well as the two shadow property names. Which as you will see by convention might not be what you expect.
So, by convention the join entity (and table) name is
{LeftEntityName}{RightEntityName}
and the shadow property (and column) names are
{LeftEntityNavigationPropertyName}{RightEntityKeyName}
{RightEntityNavigationPropertyName}{LeftEntityKeyName}
The first question would be - which is the left/right entity? The answer is (not documented yet) - by convention the left entity is the one which name is less in alphabetical order. So with your example Book is left, Category is right, so the join entity and table name would be BookCategory.
It can be changed adding explicit
modelBuilder.Entity<Category>()
.HasMany(left => left.Books)
.WithMany(right => right.Categories);
and now it would be CategoryBook.
In both cases the shadow property (and column) names would be
CategoriesCategoryId
BooksBookId
So neither the table name, nor the property/column names are what you'd normally do.
And apart from the database table/column names, the entity and property names are important because you'd need them for entity operations, including the data seeding in question.
With that being said, even if you don't create explicit join entity, it's better to configure fluently the one created automatically by EF Core convention:
modelBuilder.Entity<Book>()
.HasMany(left => left.Categories)
.WithMany(right => right.Books)
.UsingEntity("BookCategory", typeof(Dictionary<string, object>),
right => right.HasOne(typeof(Category)).WithMany().HasForeignKey("CategoryId"),
left => left.HasOne(typeof(Book)).WithMany().HasForeignKey("BookId"),
join => join.ToTable("BookCategories")
);
Now you can use the entity name to access the EntityTypeBuilder
modelBuilder.Entity("BookCategories")
and you can seed it similar to normal entities with shadow FK properties with anonymous type
modelBuilder.Entity("BookCategory").HasData(
new { BookId = 1, CategoryId = 1 }
);
or for this specific property bag type entity, also with Dictionary<string, object> instances
modelBuilder.Entity("BookCategory").HasData(
new Dictionary<string, object> { ["BookId"] = 1, ["CategoryId"] = 1 }
);
Update:
People seem to misinterpret the aforementioned "extra" steps and find them redundant and "too much", not needed.
I never said they are mandatory. If you know the conventional join entity and property names, go ahead directly to the last step and use anonymous type or Dictionary<string, object>.
I already explained the drawbacks of taking that route - loosing the C# type safety and using "magic" strings out of your control. You have to be smart enough to know the exact EF Core naming conventions and to realize that if you rename class Book to EBook the new join entity/table name will change from "BookCategory" to "CategoryEBook" as well as the order of the PK properties/columns, associated indexes etc.
Regarding the concrete problem with data seeding. If you really want to generalize it (OP attempt in their own answer), at least make it correctly by using the EF Core metadata system rather than reflection and assumptions. For instance, the following will extract these names from the EF Core metadata:
public static void HasJoinData<TFirst, TSecond>(
this ModelBuilder modelBuilder,
params (TFirst First, TSecond Second)[] data)
where TFirst : class where TSecond : class
=> modelBuilder.HasJoinData(data.AsEnumerable());
public static void HasJoinData<TFirst, TSecond>(
this ModelBuilder modelBuilder,
IEnumerable<(TFirst First, TSecond Second)> data)
where TFirst : class where TSecond : class
{
var firstEntityType = modelBuilder.Model.FindEntityType(typeof(TFirst));
var secondEntityType = modelBuilder.Model.FindEntityType(typeof(TSecond));
var firstToSecond = firstEntityType.GetSkipNavigations()
.Single(n => n.TargetEntityType == secondEntityType);
var joinEntityType = firstToSecond.JoinEntityType;
var firstProperty = firstToSecond.ForeignKey.Properties.Single();
var secondProperty = firstToSecond.Inverse.ForeignKey.Properties.Single();
var firstValueGetter = firstToSecond.ForeignKey.PrincipalKey.Properties.Single().GetGetter();
var secondValueGetter = firstToSecond.Inverse.ForeignKey.PrincipalKey.Properties.Single().GetGetter();
var seedData = data.Select(e => (object)new Dictionary<string, object>
{
[firstProperty.Name] = firstValueGetter.GetClrValue(e.First),
[secondProperty.Name] = secondValueGetter.GetClrValue(e.Second),
});
modelBuilder.Entity(joinEntityType.Name).HasData(seedData);
}
Also here you don't need to know which type is "left" and which is "right", neither requires special base class or interface. Just pass sequence of entity pairs and it will properly seed the conventional join entity, e.g. with OP example, both
modelBuilder.HasJoinData((book, category));
and
modelBuilder.HasJoinData((category, book));
would do.
Update (EF Core 5.0.2)
It's working well using the name of the associative table:
builder.Entity("ContractDeclarationType").HasData(
new { ContractsId = 1L, DeclarationTypesId = 1L },
new { ContractsId = 1L, DeclarationTypesId = 2L },
new { ContractsId = 1L, DeclarationTypesId = 3L });
I ended up whipping up a generic solution to this problem based upon the answer from Ivan (thanks!). I'm now able to seed all my M2M tables with this syntax:
// Add book1 and book2 to category1:
modelBuilder.HasM2MData(new [] { book1, book2 }, new [] { category1 });
This may not be fully robust, but it should work with conventional M2M mappings.
It makes some assumptions:
T1 & T2 Inherit from some ModelBase that provides an Id property.
T1 & T2 Have exactly one ICollection<OtherType> property.
You know the correct order (which model is T1 and which is T2) — this can be discovered by running the migration for the tables first and inspecting the migration.
You're running EFCore5 RC2 or later (see this issue).
public static void HasM2MData<T1, T2>
(this ModelBuilder mb, T1[] t1s, T2[] t2s)
where T1 : ModelBase where T2 : ModelBase
{
string table = $"{typeof(T1).Name}{typeof(T2).Name}";
PropertyInfo t1Prop = GetM2MProperty<T1, T2>();
PropertyInfo t2Prop = GetM2MProperty<T2, T1>();
string t1Key = $"{t1Prop.Name}Id";
string t2Key = $"{t2Prop.Name}Id";
foreach (T1 t1 in t1s) {
foreach (T2 t2 in t2s) {
mb.Entity(table).HasData(new Dictionary<string, object>() { [t2Key] = t1.Id, [t1Key] = t2.Id });
}
}
}
// Get a property on T1 which is assignable to type ICollection<T2>, representing the m2m relationship
private static PropertyInfo GetM2MProperty<T1, T2>() {
Type assignableType = typeof(ICollection<T2>);
List<PropertyInfo> props = typeof(T1).GetProperties()
.Where(pi => pi.PropertyType.IsAssignableTo(assignableType))
.ToList();
if (props.Count() != 1) {
throw new SystemException(
$"Expected {typeof(T1)} to have exactly one column of type {assignableType}; got: {props.Count()}");
}
return props.First();
}
In the migration, we see something like:
migrationBuilder.InsertData(
table: "BookCategory",
columns: new[] { "BooksId", "CategoriesId" },
values: new object[,]
{
{ "book1", "category1" },
{ "book2", "category1" }
});
My question is similar to serviceHub.vaultQueryService.queryBy returns all related states by default? except we have schemas mapped for states and use TypedQueries to fetch states from database.
The flow which queries a state from database:
#StartableByRPC
#Transactional
open class FetchServiceById(
private val serviceId: UniqueIdentifier
) : FlowLogic<List<ServiceSchemaV1.PersistentService>>() {
#Suspendable
override fun call(): List<ServiceSchemaV1.PersistentService> {
return serviceHub.withEntityManager {
createQuery(
"SELECT s FROM $TABLE_NAME s WHERE s.linearId = ?1",
ServiceSchemaV1.PersistentService::class.java
).setParameter(
1,
serviceId.id
).resultList
}
}
private companion object {
val TABLE_NAME = ServiceSchemaV1.PersistentService::class.jvmName
}
}
This returns a state correctly but after updating the particular state and running query again it returns both the old, consumed state and the new unconsumed state.
I checked the h2 database and can see that after a state update, in VAULT_STATE there is a date in CONSUMED_TIMESTAMP and its STATE_STATUS is 1.
How can we only query unconsumed states with typed queries? I know we could just use linearquery but when we need to query and join data from different states I don't think that would be computationally feasible.
The schema the state uses
object ServiceSchema
object ServiceSchemaV1 : MappedSchema(
schemaFamily = ServiceSchema.javaClass,
version = 1,
mappedTypes = listOf(
PersistentService::class.java
)) {
#Entity
#Table(name = "service_states")
class PersistentService(
#Column(name = "accountoperator")
var accountOperator: String,
#Column(name = "serviceprovider")
var serviceProvider: String,
#Column(name = "servicename")
var serviceName: String,
#Column(name = "servicedescription")
var serviceDescription: String,
#Column(name = "datacreated")
var dataCreated: LocalDate,
#Column(name = "linear_id")
var linearId: UUID,
#ElementCollection
#Column(name = "service_data_ids")
var serviceDataIds: MutableList<UUID>,
#ElementCollection
#Column(name = "service_partners")
var servicePartners: MutableList<String>
) : PersistentState() {
constructor() : this(
"",
"",
"",
"",
LocalDate.now(),
UUID.randomUUID(),
mutableListOf(),
mutableListOf()
)
}
}
You could perform a join query on your State's Schema and the VAULT_STATE table on the basis of the TRANSACTION_ID and OUTPUT_INDEX and filter based on STATE_STATUS.
Select X, Y Z, from MY_STATE M, VAULT_STATES V WHERE M.TRANSACTION_ID =
V.TRANSACTION_ID AND M.OUTPUT_INDEX = V.OUTPUT_INDEX AND STATE_STATUS = 0
You should check that all the necessary parties are included for the service call. Otherwise you can only see the states the particular node was included with the last state update's contract.
I am able to get a somewhat dynamic Entity Framework search result with the following simplified example, which pulls a single result from the DB or Cache:
string strTableName = "TableName2"
string strColumnName = "MyColumnName"
int intPrimaryKey = 1
Type returnType;
returnType = typeof(TableName1);
string queryResults = null;
switch (strTableName)
{
case "TableName2":
returnType = typeof(TableName2);
break;
}
var refColumnName = returnType.GetProperty(strColumnName );
var query = mydbEntity.Set(returnType).Find(intPrimaryKey );
var queryResults = refColumnName.GetValue(query).ToString();
This can also be adapted for Updating a record:
DataQuery.LeadsEntity.Entry(query).Property(strColumnName ).CurrentValue = "Whatever";
DataQuery.LeadsEntity.SaveChanges();
Is there an equivalent for way for .set(returnType).Add()? I'm not sure if there is a way to do this type of thinking using variable table and column names:
DataQuery.LeadsEntity.Set(returnType).Add(new returnType { PrimayKeyName = 1, refColumnName = "Something" });
If you don't know a priori what's the name of the primary key property, it could be a little painful to get it from the type.
This is the way I found more reliable to retrieve the primary key from the entity type:
private string[] GetKeyNames(DbContext context, Type entityType)
{
ObjectContext objectContext = ((IObjectContextAdapter)context).ObjectContext;
//create method CreateObjectSet with the generic parameter of the base-type
MethodInfo method = typeof(ObjectContext).GetMethod("CreateObjectSet", Type.EmptyTypes)
.MakeGenericMethod(entityType);
dynamic objectSet = method.Invoke(objectContext, null);
IEnumerable<dynamic> keyMembers = objectSet.EntitySet.ElementType.KeyMembers;
string[] keyNames = keyMembers.Select(k => (string)k.Name).ToArray();
_keyNamesCache[entityType] = keyNames;
return keyNames;
}
But assuming your primary keys are always a single property, you could use reflection to create the entity object and set its properties like this:
private void CreateEntity(Type entityType, object pkValue, Dictionary<string, object> Columns)
{
// Create the new entity
var entity = Activator.CreateInstance(entityType);
// Get the primary key property name
var pkName = GetKeyNames(context, entityType).First();
// Set Pk value
entityType.GetProperty(pkName).SetValue(entity, pkValue);
// Set other column(s)
foreach (var col in Columns)
{
entityType.GetProperty(col.Key).SetValue(entity, col.Value);
}
// Add the entity to the DbSet
using (var context = new YourContext())
{
context.Set(entityType).Add(entity);
}
}
I am trying to setup a product key system in my application, but I want to ensure the attribute has the right size (16 characters).
I tried the following
public class ProductKey
{
public const int ProductKeyLength = 16;
[StringLength(ProductKeyLength, MinimumLength = ProductKeyLength)]
private string _value;
[Required]
[Index(IsUnique = true)]
public string Value {
get
{
var temp = Regex.Replace(this._value, ".{4}", "$0-");
return temp.Trim('-');
}
set { this._value = value.Replace("-", "");}
}
}
I want to enable the user to insert the key with our without hyphen. I get the following error with above code:
Column 'Value' in table 'dbo.ProductKeys' is of a type that is invalid for use as a key column in an index.
As I understood, I need to set a limit to Value so it can be used as a unique key. But, _value has a limit and _value is the actual representation of Value in the database.
Is there a way to set the limit correctly in this case?
Thanks in advance.
You are getting the error because without a StringLength attribute on the Value field, the database column gets created as VARCHAR(MAX) which cannot be used as a key. You need a [StringLength] on the field being used as a key. However, as your getter is returning the key formatted with dashes, you need the key length to be 19:
public class ProductKey
{
public const int ProductKeyLength = 19;
private string _value { get; set; }
[Key]
[Required]
[StringLength(ProductKeyLength, MinimumLength = ProductKeyLength)]
[Index(IsUnique = true)]
public string Value
{
get
{
var temp = Regex.Replace(this._value, ".{4}", "$0-");
return temp.Trim('-');
}
set { this._value = value.Replace("-", ""); }
}
}
You might be better off doing your format conversion in ViewModels and client-side code, as one problem you'll have here is searching - for example...
db.Keys.Add(new ProductKey { Value = "1234-5678-9012-3456" });
db.Keys.Add(new ProductKey { Value = "1234567890123455" });
db.SaveChanges();
Console.WriteLine(db.Keys.Count(k => k.Value.Contains("89"))); // 0
Console.WriteLine(db.Keys.Count(k => k.Value.Contains("8-9"))); // 2
I have a domain called Test that has been running look-up SQL queries to select from another database.
I want to get away from this implementation and goto multiple datasource support of Grails 2.0. However, the table name in the other database is a called Panel.
Is it possible to map the domain to the alternative database and also map which table it selects from?
// Datasource.groovy
development {
dataSource {
dbCreate = 'create-drop' // one of 'create', 'create-drop','update'
url = "jdbc:sqlserver://machine\\SQLEXPRESS;databaseName=primarydb"
username = "user"
password = "password"
}
dataSource_otherdb {
url = "jdbc:sqlserver://remoteserver:1433;databaseName=otherdb"
}
}
// Test.groovy
class Test {
String name
int key
String abbreviation
boolean active = true
static mapping = {
sort name: "asc"
datasource("otherdb")
}
}
There is a "table"configuration in mapping closure for that.
// Test.groovy
class Test {
String name
int key
String abbreviation
boolean active = true
static mapping = {
table 'Panel' //customize table name
sort name: "asc"
datasource("otherdb")
}
}