SysTec.EF.ChangeTracking.DetachedGraphTracker 7.0.1

EF-Detached-Graph-Tracker

A library to support easier change tracking of a complex detached graph with EF-Core.

Created by Tobias Theiss @ https://www.systec-computer.de

What does this library do?

It handles the change tracking of a complex detached graph using pure vanilla EF-Core. Let's begin by listing the three main problems EF-Core has with updating detached graphs.

Why does this library depend on a postreSQL NuGet-Package?

1. EF supports database generated values. For example a number may be generated on INSERT and may not be overwritten by an UPDATE statement (number may not be present in the statement).
2. EF also supports looking up if a value is database generated for every property on an entity entry (property.Metadata.ValueGenerated).<br/>

The problem is that when using postgreSQL´s UseIdentityAlwaysColumn() the value of Metadata.ValueGenerated is set to ValueGenerated.OnAdd (what would still be correct in this case).<br/> When setting UseIdentityByDefaultColumn() the value of Metadata.ValueGenerated is also set to ValueGenerated.OnAdd which is incorrect because the value can be overwritten in a update statement. <br/> <br/> So relying on the Metadata.ValueGenerated property is not possible.<br/> PostgreSQL provides the method Metadata.GetValueGenerationStrategy() to find out what strategy is used. The documentation states that when no PostreSQL specific strategy is found, it defaults to the IModel of EF, so even when using a different provider the library should work (except when they also don´t set Metadata.ValueGenerated correctly). In this case generated values (except key values, ef manges them correctly) must be handled by custom code (set to Unchanged on update).

EF-Core Detached Graph Problems

• Identity Resolution: When attaching a graph to the change tracker, the caller needs to assure that the graph contains only one element of the same type and the same key.<br/> See following example:

RootNode
├── Item: {Id: 1, Text: "This text should be persisted"}
└── List<Item>: {Id: 1, Text "This text should not be persisted"}, {...}

// Throws Item with Id 1 can not be tracked because another instance with the same key is already tracked...
dbContext.Attach(RootNode);

• Another problem are associations where you don't want to persist changes to the entity inside the navigation property. <br/> In relation to the previous example consider that the client UI allows updates of Item on top of a page and just allows the user to associate Items in another part of the page. <br/> <br/> This problem is connected to the identity resolution problem. When identity resolution is performed automatically, it must somehow be determined which entity values in the graph to persist (track as Modified or Added) and which to ignore (track as Ùnchanged).
• The third problem is severing 1:many or many:many relationships by just removing items from a detached collection ont the client side.<br/> Another example makes this clear:

  // First Request -> New Item will get Id 1
  RootNode
  └── List<Item>: {Id: 0, "This will be Added"}
  
  // Second Request -> New Item will get Id 2
  RootNode
  └── List<Item>: {Id: 1, "This will be Modified"}, {Id: 0, "This will be Added"}
  
  // Third Request -> Only the text of Item with Id 2 will change. Nothing else. No relationship gets severed.
  RootNode
  └── List<Item>: {Id: 2, "This will be Modified"}
  

How does this library solve the problems?

• Identity Resolution & Associations: You can set the [UpdateAssociationOnly] attribute over any navigation property. <br/> It makes sure entities in nodes annotated with this attribute are kept in the unchanged state (only relationships / FKs are updated). If an entity with a temporary key value (e.g. Id = 0, Id = -123) is being tracked over a navigation property annotated with this attribute, an exception will be thrown. However you can configure the behavior to keep the entity in a detached state instead of throwing an exception using the parameter in the attribute constructor. See usage for further details.
• 1:many / many:many missing list items: The library loads the original values of a collection navigation from the database and then removes all items that are not present in the current collection but in the original DB collection. Depending on the configuration of the relationship and the delete behavior, cascading actions may occur.

How to use this library?

1. Install the nuget package SysTec.EF.ChangeTracking.DetachedGraphTracker
2. Create a new instance of DetachedGraphTracker and make sure one instance is used per request (for example register it as a scoped service in a Web-API application).
3. Set attributes in your models.
4. Call graphTrackerInstance.TrackGraphAsync(rootNode); with the root of your graph.
5. After the call every node in the graph that has been traversed is tracked in a Modified, Unchanged, Added or Deleted state. <br/> The states depend on the attributes set in the model classes (refer to attributes for details). <br/> When no attributes are set, the node is tracked depending on its key value. If the key is set (e. g. Id = 42) the state is set to Modified otherwise the entity is tracked in an Added state.

Attributes

• This libary provides 3 attributes:
  1. [UpdateAssociationOnly]

  RootNode
  └── Item: {Id: 2, "Initial"}
  
  // This will update RootNode, the relationship between RootNode and Item and the text of Item with Id 2.
  await graphTrackerInstance.TrackGraphAsync(RootNode);
  
  RootNode
  └── [UpdateAssociationOnly] Item: {Id: 2, "Updated"}
  
  // This will update RootNode and the relationship between RootNode and Item. The text is not being updated.
  await graphTrackerInstance.TrackGraphAsync(RootNode);
  

  RootNode
  ├── Item: {Id: 1, Text: "This text should be persisted"}
  └── List<Item>: {Id: 1, Text "This text should not be persisted"}, {...}
  
  // Throws MultipleCompositionException
  await graphTrackerInstance.TrackGraphAsync(RootNode);
  
  RootNode
  ├── Item: {Id: 1, Text: "This text should be persisted"}
  └── [UpdateAssociationOnly] List<Item>: {Id: 1, Text "This text should not be persisted"}, {...}
  
  // Does not throw, updates Text of Item with Id 1 to "This text should be persisted" and updates both relationships between Item and RootNode.
  await graphTrackerInstance.TrackGraphAsync(RootNode);
  

  2. [ForceDeleteOnMissingEntries]: In some cases relationships are configured to be non required (e. g. for TPH configuration where multiple types get mapped to a singe table) but should not exist in the database when the relationship to their principal is severed. <br/> Therefore this attribute sets the state of all entities, that are missing in a collection navigation but are still present in the original database collection, to Deleted. This overrides the default behavior of the change detection for collection entries.

  3. [ForceKeepExistingRelationship]: There may be cases where a navigation property is needed for logic in the backend but is never seen from the client because it may be ignored in a DTO for example. Now when the client sends back the data to the server, the value of the "hidden" navigation property is null and the relationship would be severed by the TrackGraphAsync() method. In case this behavior is unwanted, use the attribute to keep the existing relationship if the value is null (or the item is missing in a collection navigation). Important: New relationships can always be connected even when this attribute is set. Also when providing a valid entity to a reference navigation annotated with this attribute, the relationship also gets changed.

How does the library work (in a more detailed way)

1. After the call to graphTrackerInstance.TrackGraphAsync(rootNode) graph traversal is performed using the DbContext.ChangeTracker.TrackGraph(object root, Action<EntityEntryGraphNode> callback) method.
2. In the callback it is first determined if the entity is already present in the change tracker.
   1. If it is, and the current node is not annotated with an [UpdateAssociationOnly] attribute, an exception will be thrown, because in this case multiple nodes of the same type with the same key value are present in the graph and are not annotated with an [UpdateAssociationOnly] attribute.<br/>
   2. If it´s not, and the node is not annotated with an [UpdateAssociationOnly] attribute, the state will be set depending on the key value of the entity (Id > 0 → Modified, otherwise Added) <br/>
   3. If it´s not, and the node is annotated with an [UpdateAssociationOnly] attribute, the entity will be kept in a detached state and stored in a list to be handled later .
3. After the graph traversal is finished, Identity Resolution is now performed for the yet detached association entities (2.iii).
   1. It an entity of the same type and key as the association is already present in the change tracker, the association gets "replaced" with the entity from the change tracker because only entities that are not annotated with the [UpdateAssociationOnly] attribute are present in the change tracker (2.ii)
   2. If no entity of the same type and key as the association is present in the change tracker and the key value of the association is set (e. g. Id > 0), it will be tracked in an Unchanged state.
   3. If no entity of the same type and key as the association is present in the change tracker and the key value of the association is not set (e. g. Id < 0) the entity will remain in the Detached state or an exception will be thrown, depending on the parameter passed to the [UpdateAssociationOnly] attribute.
4. Finally list items that are not present in collection navigations in the graph, but still are in the original collection loaded from the database, are removed using the collection.Remove() method on the original collection. <br/> This way the change tracker is aware of the change in the collection an relationships are severed. Cascading actions may occur depending on the configuration of the relationships and the configured delete behaviors.