Datastore library

This library provides consistent API for data management (including backup, restore, and metrics logging) on Android platform.

Notably, it is designed to be flexible and could be utilized for a wide range of data store besides the settings preferences.

Overview

In the high-level design, a persistent datastore aims to support two key characteristics:

• observable: triggers backup and metrics logging whenever data is changed.
• transferable: offers users with a seamless experience by backing up and restoring data on to new devices.

More specifically, Android framework supports data backup to preserve user experiences on a new device. And the observer pattern allows to monitor data change.

Backup and restore

Currently, the Android backup framework provides BackupAgentHelper and BackupHelper to facilitate data backup. However, there are several caveats to consider when implementing BackupHelper:

• performBackup: The data is updated incrementally but it is not well documented. The ParcelFileDescriptor state parameters are normally ignored and data is updated even there is no change.
• restoreEntity: The implementation must take care not to seek or close the underlying data source, nor read more than size() bytes from the stream when restore (see BackupDataInputStream). It is possible that a BackupHelper interferes with the restore process of other BackupHelpers.
• writeNewStateDescription: Existing implementations rarely notice that this callback is invoked after all entities are restored. Instead, they check if necessary data are all restored in the restoreEntity (e.g. BatteryBackupHelper), which is not robust sometimes.

The datastore library will mitigate these problems by providing alternative APIs. For instance, library users make use of InputStream / OutputStream to back up and restore data directly.

Observer pattern

In the current implementation, the Android backup framework requires a manual call to BackupManager.dataChanged. However, it‘s often observed that this API call is forgotten when using SharedPreferences. Additionally, there’s a common need to log metrics when data changed. To address these limitations, datastore API employed the observer pattern.

API design and advantages

Datastore must extend the BackupRestoreStorage class (subclass of BackupHelper). The data in a datastore is group by entity, which is represented by BackupRestoreEntity. Basically, a datastore implementation only needs to focus on the BackupRestoreEntity.

If the datastore is key-value based (e.g. SharedPreferences), implements the KeyedObservable interface to offer fine-grained observer. Otherwise, implements Observable. There are builtin thread-safe implementations of the two interfaces (KeyedDataObservable / DataObservable). If it is Kotlin, use delegation to simplify the code.

Keep in mind that the implementation should call KeyedObservable.notifyChange / Observable.notifyChange whenever internal data is changed, so that the registered observer will be notified properly.

For SharedPreferences use case, leverage the SharedPreferencesStorage directly. To back up other file based storage, extend the BackupRestoreFileStorage class.

Here are some highlights of the library:

• The restore InputStream will ensure bounded data are read, and close the stream is no-op. That being said, all entities are isolated.
• Data checksum is computed automatically, unchanged data will not be sent to Android backup system.
• Data compression is supported:
  • ZIP best compression is enabled by default, no extra effort needs to be taken.
  • It is safe to switch between compression and no compression in future, the backup data will add 1 byte header to recognize the codec.
  • To support other compression algorithms, simply wrap over the InputStream and OutputStream. Actually, the checksum is computed in this way by CheckedInputStream and CheckedOutputStream, see BackupRestoreStorage implementation for more details.
• Enhanced forward compatibility for file is enabled: If a backup includes data that didn't exist in earlier versions of the app, the data can still be successfully restored in those older versions. This is achieved by extending the BackupRestoreFileStorage class, and BackupRestoreFileArchiver will treat each file as an entity and do the backup / restore.
• Manual BackupManager.dataChanged call is unnecessary now, the framework will invoke the API automatically.

Usages

This section provides examples of datastore.

Here is a datastore with a string data:

class ExampleStorage : ObservableBackupRestoreStorage() {
  @Volatile // field is manipulated by multiple threads, synchronization might be needed
  var data: String? = null
    private set

  @AnyThread
  fun setData(data: String?) {
    this.data = data
    // call notifyChange to trigger backup and metrics logging whenever data is changed
    if (data != null) {
      notifyChange(ChangeReason.UPDATE)
    } else {
      notifyChange(ChangeReason.DELETE)
    }
  }

  override val name: String
    get() = "ExampleStorage"

  override fun createBackupRestoreEntities(): List<BackupRestoreEntity> =
    listOf(StringEntity("data"))

  override fun enableRestore(): Boolean {
    return true // check condition like flag, environment, etc.
  }

  override fun enableBackup(backupContext: BackupContext): Boolean {
    return true // check condition like flag, environment, etc.
  }

  @BinderThread
  private inner class StringEntity(override val key: String) : BackupRestoreEntity {
    override fun backup(backupContext: BackupContext, outputStream: OutputStream) =
      if (data != null) {
        outputStream.write(data!!.toByteArray(UTF_8))
        EntityBackupResult.UPDATE
      } else {
        EntityBackupResult.DELETE // delete existing backup data
      }

    override fun restore(restoreContext: RestoreContext, inputStream: InputStream) {
      // DO NOT call setData API here, which will trigger notifyChange unexpectedly.
      // Under the hood, the datastore library will call notifyChange(ChangeReason.RESTORE)
      // later to notify observers.
      data = String(inputStream.readBytes(), UTF_8)
      // Handle restored data in onRestoreFinished() callback
    }
  }

  override fun onRestoreFinished() {
    // TODO: Update state with the restored data. Use this callback instead of "restore()" in
    //       case the restore action involves several entities.
    // NOTE: The library will call notifyChange(ChangeReason.RESTORE) for you
  }
}

And this is a datastore with key value data:

class ExampleKeyValueStorage :
  BackupRestoreStorage(), KeyedObservable<String> by KeyedDataObservable() {
  // thread safe data structure
  private val map = ConcurrentHashMap<String, String>()

  override val name: String
    get() = "ExampleKeyValueStorage"

  fun updateData(key: String, value: String?) {
    if (value != null) {
      map[key] = value
      notifyChange(ChangeReason.UPDATE)
    } else {
      map.remove(key)
      notifyChange(ChangeReason.DELETE)
    }
  }

  override fun createBackupRestoreEntities(): List<BackupRestoreEntity> =
    listOf(createMapBackupRestoreEntity())

  private fun createMapBackupRestoreEntity() =
    object : BackupRestoreEntity {
      override val key: String
        get() = "map"

      override fun backup(
        backupContext: BackupContext,
        outputStream: OutputStream,
      ): EntityBackupResult {
        // Use TreeMap to achieve predictable and stable order, so that data will not be
        // updated to Android backup backend if there is only order change.
        val copy = TreeMap(map)
        if (copy.isEmpty()) return EntityBackupResult.DELETE
        val dataOutputStream = DataOutputStream(outputStream)
        dataOutputStream.writeInt(copy.size)
        for ((key, value) in copy) {
          dataOutputStream.writeUTF(key)
          dataOutputStream.writeUTF(value)
        }
        return EntityBackupResult.UPDATE
      }

      override fun restore(restoreContext: RestoreContext, inputStream: InputStream) {
        val dataInputString = DataInputStream(inputStream)
        repeat(dataInputString.readInt()) {
          val key = dataInputString.readUTF()
          val value = dataInputString.readUTF()
          map[key] = value
        }
      }
    }
}

All the datastore should be added in the application class:

class ExampleApplication : Application() {
  override fun onCreate() {
    super.onCreate()
    BackupRestoreStorageManager.getInstance(this)
      .add(ExampleStorage(), ExampleKeyValueStorage())
  }
}

Additionally, inject datastore to the custom BackupAgentHelper class:

class ExampleBackupAgent : BackupAgentHelper() {
  override fun onCreate() {
    super.onCreate()
    BackupRestoreStorageManager.getInstance(this).addBackupAgentHelpers(this)
  }

  override fun onRestoreFinished() {
    BackupRestoreStorageManager.getInstance(this).onRestoreFinished()
  }
}

Development

Please preserve the code coverage ratio during development. The current line coverage is 100% (444/444) and branch coverage is 93.6% (176/188).