Many applications deal with data that is nested or relational in nature. For example, a blog editor could have many Posts, each Post could have many Comments, and both Posts and Comments would be written by a User. Data for this kind of application might look like:
const blogPosts = [ { id : "post1", author : {username : "user1", name : "User 1"}, body : "......", comments : [ { id : "comment1", author : {username : "user2", name : "User 2"}, comment : ".....", }, { id : "comment2", author : {username : "user3", name : "User 3"}, comment : ".....", } ] }, { id : "post2", author : {username : "user2", name : "User 2"}, body : "......", comments : [ { id : "comment3", author : {username : "user3", name : "User 3"}, comment : ".....", }, { id : "comment4", author : {username : "user1", name : "User 1"}, comment : ".....", }, { id : "comment5", author : {username : "user3", name : "User 3"}, comment : ".....", } ] } // and repeat many times ]
Notice that the structure of the data is a bit complex, and some of the data is repeated. This is a concern for several reasons:
Because of this, the recommended approach to managing relational or nested data in a Redux store is to treat a portion of your store as if it were a database, and keep that data in a normalized form.
The basic concepts of normalizing data are:
An example of a normalized state structure for the blog example above might look like:
{ posts : { byId : { "post1" : { id : "post1", author : "user1", body : "......", comments : ["comment1", "comment2"] }, "post2" : { id : "post2", author : "user2", body : "......", comments : ["comment3", "comment4", "comment5"] } } allIds : ["post1", "post2"] }, comments : { byId : { "comment1" : { id : "comment1", author : "user2", comment : ".....", }, "comment2" : { id : "comment2", author : "user3", comment : ".....", }, "comment3" : { id : "comment3", author : "user3", comment : ".....", }, "comment4" : { id : "comment4", author : "user1", comment : ".....", }, "comment5" : { id : "comment5", author : "user3", comment : ".....", }, }, allIds : ["comment1", "comment2", "comment3", "commment4", "comment5"] }, users : { byId : { "user1" : { username : "user1", name : "User 1", } "user2" : { username : "user2", name : "User 2", } "user3" : { username : "user3", name : "User 3", } }, allIds : ["user1", "user2", "user3"] } }
This state structure is much flatter overall. Compared to the original nested format, this is an improvement in several ways:
Note that a normalized state structure generally implies that more components are connected and each component is responsible for looking up its own data, as opposed to a few connected components looking up large amounts of data and passing all that data downwards. As it turns out, having connected parent components simply pass item IDs to connected children is a good pattern for optimizing UI performance in a React Redux application, so keeping state normalized plays a key role in improving performance.
A typical application will likely have a mixture of relational data and non-relational data. While there is no single rule for exactly how those different types of data should be organized, one common pattern is to put the relational "tables" under a common parent key, such as "entities". A state structure using this approach might look like:
{ simpleDomainData1: {....}, simpleDomainData2: {....} entities : { entityType1 : {....}, entityType2 : {....} } ui : { uiSection1 : {....}, uiSection2 : {....} } }
This could be expanded in a number of ways. For example, an application that does a lot of editing of entities might want to keep two sets of "tables" in the state, one for the "current" item values and one for the "work-in-progress" item values. When an item is edited, its values could be copied into the "work-in-progress" section, and any actions that update it would be applied to the "work-in-progress" copy, allowing the editing form to be controlled by that set of data while another part of the UI still refers to the original version. "Resetting" the edit form would simply require removing the item from the "work-in-progress" section and re-copying the original data from "current" to "work-in-progress", while "applying" the edits would involve copying the values from the "work-in-progress" section to the "current" section.
Because we're treating a portion of our Redux store as a "database", many of the principles of database design also apply here as well. For example, if we have a many-to-many relationship, we can model that using an intermediate table that stores the IDs of the corresponding items (often known as a "join table" or an "associative table"). For consistency, we would probably also want to use the same byId
and allIds
approach that we used for the actual item tables, like this:
{ entities: { authors : { byId : {}, allIds : [] }, books : { byId : {}, allIds : [] }, authorBook : { byId : { 1 : { id : 1, authorId : 5, bookId : 22 }, 2 : { id : 2, authorId : 5, bookId : 15, } 3 : { id : 3, authorId : 42, bookId : 12 } }, allIds : [1, 2, 3] } } }
Operations like "Look up all books by this author" can then accomplished with a single loop over the join table. Given the typical amounts of data in a client application and the speed of Javascript engines, this kind of operation is likely to have sufficiently fast performance for most use cases.
Because APIs frequently send back data in a nested form, that data needs to be transformed into a normalized shape before it can be included in the state tree. The Normalizr library is usually used for this task. You can define schema types and relations, feed the schema and the response data to Normalizr, and it will output a normalized transformation of the response. That output can then be included in an action and used to update the store. See the Normalizr documentation for more details on its usage.
© 2015–2017 Dan Abramov
Licensed under the MIT License.
http://redux.js.org/docs/recipes/reducers/NormalizingStateShape.html