Skip to content

briandilley/joraph

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Build And Deploy Joraph codecov

Joraph

Java library for loading and joining object graphs, JOin gRAPH. Joraph's only dependency is the JVM and Kotlin runtime. It can be used with Kotlin and Java (8+) projects alike. It has been used in many high traffic production environments and is battle tested. Joraph is not a data access library, it knows nothing about your RDBMS, cache, or other persistant storage. It is merely a wrapper on top of these things to make it easy to load and join related objects (across many different storage mechanisms) into an object graph. While the sample code in this documentation is Kotlin, care has been taking to ensure that the library is just as easy to use in a Java environment.

Maven / Gradle integration

Joraph is available from Maven Central:

maven

<dependency>
    <groupId>com.github.briandilley.joraph</groupId>
    <artifactId>joraph</artifactId>
    <version>1.1</version>
</dependency>

gradle

dependencies {
    compile('com.github.briandilley.joraph:joraph:1.1')
}

What's it useful for?

Joraph is useful in places where given a few entities that you've already loaded you need to load a graph of dependent (and transitive dependencies) objects to build a response to an API call. For instance, lets say you have an endpoint GET /api/videos/{id} that loads the given video by id and returns the following payload:

{
  "video": {
    "id": 69,
    "title": "My awesome video",
    "url": "https://www.example.com/videos/dg9se8gj.mp4",
    "musicTrack": {
      "id": 64584,
      "name": "California Love",
      "url": "https://www.example.com/tracks/34lkjfg.aac",
      "artist": {
        "id": 23462,
        "name": "Tupac Shakur",
        "description": "The most prolific rapper of all time"
      }
    },
    "createdBy": {
      "id": 6846,
      "display_name": "HipHop Lover",
      "username": "john_smith_loves_hiphop"
    }
  }
}

As you can see, the root object is a video. The video contains a musicTrack (meta data about the music in the video) and a createBy (the user that created it). Further, the musicTrack contains an artist (meta data about the artist who recorded the musicTrack).

A simple data model in an RDMS for this might look like this:

  • Video (id, title, url, creator_id, music_track_id)
  • MusicTrack (id, name, url, music_artist_id)
  • MusicArtist (id, name, description)
  • User (id, displayName, username)

A simple implementation for this endpoint might look something like this:

    val video = videoManager.getVideoById(id)
    val musicTrack = musicTrackManager.getMusicTrackById(video.music_track_id)
    val artist = artistManager.getArtistById(musicTrack.artist_id)
    val createBy = userManager.getUserById(video.createdBy)
    
    val responseObject = assembleVideoResponse(video, musicTrack, artist, createdBy)

While with joraph it would look like this:

    val video = videoManager.getVideoById(id)
    val objectGraph = joraphContext.query(Query()
            .withRootEntity(video))
            
    val responseObject = assembleVideoResponse(objectGraph, video)

Much more simple, but even better - imagine an endpoint GET /api/videos/search?text=funny+memes that returns many video objects and all of the work entailed in loading the dependencies (and transitive dependencies), let alone associating the correct dependencies with their parent objects.

The benefit of Joraph becomes even clearer when:

  • your object graph contains many more objects
  • the graph is deeper (more transitive dependencies)
  • dependencies are optional
  • multiple root object types
  • your endpoints return lists of objects

How to use it

After you've added the required dependencies to your gradle script or maven pom you can begin by setting up your schema. Borrowing from the schema outlined above:

The object model model as exposed by your business or data layer:

data class Video(
  val id: Long,
  val title: String,
  val url: String,
  val creatorId: Long,
  val musicTrackId: Long)

data class MusicTrack(
  val id: Long,
  val name : String,
  val url: String,
  val musicArtistId: Long)

data class MusicArtist(
  val id: Long,
  val name: String,
  val description: String)

data class User(
  val id: Long,
  val displayName: String,
  val username: String)

Create a schema for the object model and then add EntityDescriptors to it that define your schema.

val schema = new Schema()

// add Video class
schema.addEntityDescriptor(Video::class.java)
  // who's primary key is the `id` property
  .withPrimaryKey(Video::id)
  // and has a foreign key to the `User` object from it's `creatorId` property
  .withForeignKey(User::class.java, Video::creatorId)
  // along with a foreign key to the `MusicTrack` object from it's `musicTrackId` property
  .withForeignKey(MusicTrack::class.java, Video::musicTrackId)

// add MusicTrack class
schema.addEntityDescriptor(MusicTrack::class.java)
  .withPrimaryKey(MusicTrack::id)
  .withForeignKey(MusicArtist::class.java, MusicTrack::musicArtistId)

// add MusicArtist class
schema.addEntityDescriptor(MusicArtist::class.java)
  .withPrimaryKey(MusicArtist::id)

// add User class
schema.addEntityDescriptor(User::class.java)
  .withPrimaryKey(User::id)

Every entity must have one PrimaryKey, and may have 0 or more ForeignKeys. Joraph only loads entities by their PrimaryKey, and does so in bulk (ie: multi-get). This makes it extremely easy to implement caching of entities.

Now create the JoraphContext, you'll use this object to execute queries - you only need one instance of it for your entire application to use:

val joraphContext = JoraphContext(
  // use the schema configured
  schema = schema,
  // Joraph loads entities in parallel when possible,
  // this defines how many threads are available to
  // joraph for doing so.
  parallelExecutorCount = 50)

Now you need to tell joraph how to load the entities that you expect it to load, do this with the EntityLoaderContext:

val loaderContext: EntityLoaderContext = joraphContext.loaderContext

loaderContext.withLoader(Video::class.java) { ids: Collection<Long> ->
  videoManager.findVideosByIds(ids)
}

loaderContext.withLoader(MusicTrack::class.java) { ids: Collection<Long> ->
  musicTrackRestClient.getMusicTracks(ids)
}

loaderContext.withLoader(MusicArtist::class.java, MusicArtistDao::loadArtistsByIds)

loaderContext.withLoader(User::class) { ids: Collection<Long> ->
  cachedUserManager.getUsersByIds(ids)
}

A loader is merely a function (or lambda) that is given a Collection of ids and is expected to return the objects for those ids. The ids can be of any type and the loader can load from anywhere.

Now your context is created and ready to be used:

search for some videos:

val videos: List<Video> = videoManager.searchVideos("Tupac")

and use Joraph to load the rest of the object graph for each video:

val objectGraph = joraphManager.query(Query()
  .withRootEntities(videos))

Now all of the videos and all of the dependent objects are inside of the objectGraph:

val video0 = objectGraph.get(Video::class.java, videos.get(0).id)
val video0CreatedByUser = objectGraph.get(User::class.java, videos.get(0).creatorId)
etc.

Your ObjectGraph is populated and ready to be used to create your response to the API call. Follows is a common pattern used to do just that:

class VideoResponse(
  val id: Long,
  val title: String,
  val url: String,
  val creator: UserResponse,
  val musicTrack: MusicTrackResponse) {
    
  companion object {
      
    fun from(objectGraph: ObjectGraph, video: Video?): VideoResponse? {
      if (video == null) { 
        return null
      }
      
      val creator = objectGraph.get(User::class.java, video.creatorId)
      val musicTrack = objectGraph.get(MusicTrack::class.java, video.musicTrackId)
      
      return VideoResponse(
        id          = video.id,
        title       = video.title,
        url         = video.url,
        creator     = UserResponse.from(objectGraph, creator),
        musicTrack  = MusicTrackResponse.from(objectGraph, musicTrack))
    }
  }
}

class MusicTrackResponse(
  val id: Long,
  val name: String,
  val url: String,
  val musicArtist: MusicArtistResponse) {

  companion object {

    fun from(objectGraph: ObjectGraph, musicTrack: MusicTrack?): MusicTrackResponse? {
      if (musicTrack == null) {
        return null
      }

      val musicArtist = objectGraph.get(MusicArtist::class.java, musicTrack.musicArtistId)

      return MusicTrackResponse(
        id          = musicTrack.id,
        name        = musicTrack.title,
        url         = musicTrack.url,
        musicArtist = MusicArtistResponse.from(objectGraph, musicArtist))
    }
  }
}

class MusicArtistResponse(
  val id: Long,
  val name: String,
  val description: String) {

  companion object {

    fun from(objectGraph: ObjectGraph, artist: MusicArtist?): MusicArtistResponse? {
      if (artist == null) {
        return null
      }

      return MusicArtistResponse(
        id          = artist.id,
        name        = artist.title,
        description = artist.description)
    }
  }
}

class UserResponse(
  val id: Long,
  val displayName: String,
  val username: String) {

  companion object {

    fun from(objectGraph: ObjectGraph, user: User?): UserResponse? {
      if (user == null) {
        return null
      }

      return UserResponse(
        id          = user.id,
        displayName = user.displayName,
        username    = user.username)
    }
  }
}

The pattern is pretty simple. Response objects each contain a static method called from that take an ObjectGraph as the first parameter and the actual model object as it's second parameter. It then populates itself using the model object and referring to other response objects's from methods to hydrate child objects.

And an example usage of a search endpoint returning a list of Video objects.

    @GetMapping("/api/videos/search")
    fun searchVideos(@RequestParam("search") search: String): List<VideoResponse> {
      
      val videos = videoManager.searchVideos(search)
      
      val objectGraph = joraphContext.query(Query()
        .withRootEntities(videos))

      return videos
        .map { video -> VideoResponse.from(objectGraph, video)}
    }

Future work

  • View hydration layer. Something implementing the above described pattern that keeps you from having to create the response classes.

  • GraphQL front-end. Joraph could be a great back-end to a GraphQL front-end.

  • Documentation of all of the functionality (ie: conditional foreign keys, loader arguments, etc.)

Who has used Joraph in Production?

  • Flipagram: (now Vigo Video) #1 app in 81 countries, hundreds of millions of users, serving request throughput in excess of ~500k requests per minute and joining objects across a 24 node postgresql shard, 50+ node Cassandra cluster, Redis, Memcached, and ElasticSearch.

  • Blockfolio: #1 crypto portfolio tracking app, tens of millions of users,s ervice request throughput in excess of 50k requests per minute and joining objects across PostgreSQL, Cassandra, Redis, Memcached, InfluxDB, and more.

  • QuickRide: Dealership shuttle management software, thousands of users, joining objects from an RDBMS.

  • Add an issue if you'd like to be added to this list, include a description of your usage.

About

Java library for joining objects into a graph, JOin gRAPH

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 3

  •  
  •  
  •