stash

Overview

The stash library is a key-value store abstraction with a pluggable backend architecture that provides support for vault and cache objects. The vault is a simple key value storage for primitives and objects but the cache goes one step further and adds caching semantics. The design of the cache was heavily influenced by the JCache spec from the Java world, albeit it draws inspiration from other libraries as well. It supports the most traditional capabilities found on well know caching libraries like expiration or eviction and all it's core concepts were designed from ground up with extensibility in mind.

3rd party library support was a major concern since the inception, as such, a library stash_test is provided with a complete set of tests that allow the developers of novel storage backends to test their implementations against the same baseline tests that were used by the main library.

History

The stash library started it's life as a pure cache library, however with the multiple versions, it become clear that the pluggable storage backend architecture could be reused to provide a key value abstraction that can be leveraged in other scenarios like the storage of preferences or assets or as a general purpose key value database. With version 4.0.0, comes the first major revision and most of the breaking changes since it's inception.

Features

• :rocket: Binary serialization - Provides out-of-box highly performant binary serialization using an implementation of msgpack inspired on the msgpack_dart package and adapted to the specific needs of this library .
• :twisted_rightwards_arrows: Type Agnostic Reuse the same store for multiple types. Each store is partitioned and it's possible to configure a type and the respective fromEncodable function per partition.
• :sparkles: Events - Subscribable events for vault and cache instances, providing notifications on creation, update or removal of an entry and also expiry and eviction notifications for caches.
• :bar_chart: Statistics - Supports the capture of vault and cache statistics
• :alarm_clock: Cache Expiry policies - out-of-box support for Eternal,Created, Accessed, Modified and Touched policies.
• :outbox_tray: Cache Eviction policies - out-of-box support for FIFO(first-in, first-out), FILO (first-in, last-out), LRU (least-recently used), MRU (most-recently used), LFU (least-frequently used, the default), MFU (most frequently used) and Hyperbolic
• :game_die: Cache entry sampling - Sampling of eviction candidates randomly or via reservoir sampling
• :hamburger: Tiered cache - Allows the configuration of a primary highly performing cache (in-memory for example) and a secondary second-level cache
• :loop: Extensible - Pluggable implementations of custom encoding/decoding, storage, stats, expiry, eviction and sampling strategies.
• :wrench: Testable - Storage, vault and cache harness for 3d party support of novel storage backend strategies

Storage Implementations

There's a vast array of storage implementations available which you can use.

Package	Version	SDK	Platform	Description
stash_memory		DART FLUTTER	ANDROID IOS LINUX MACOS WEB WINDOWS	A memory storage implementation
stash_file		DART FLUTTER	ANDROID IOS LINUX MACOS WEB WINDOWS	A storage implementation using the file package
stash_sqlite		DART FLUTTER	ANDROID IOS LINUX MACOS WINDOWS	A storage implementation using the drift package
stash_hive		DART FLUTTER	ANDROID IOS LINUX MACOS WEB WINDOWS	A storage implementation using the hive package
stash_sembast		DART FLUTTER	ANDROID IOS LINUX MACOS WINDOWS	A storage implementation using the sembast package
stash_sembast_web		DART FLUTTER	WEB	A storage implementation using the sembast_web package
stash_objectbox		DART FLUTTER	ANDROID IOS LINUX MACOS WINDOWS	A storage implementation using the objectbox package
stash_cbl		DART FLUTTER	ANDROID IOS LINUX MACOS WINDOWS	A storage implementation using the cbl package
stash_isar		DART FLUTTER	ANDROID IOS LINUX MACOS WINDOWS	A storage implementation using the isar package
stash_shared_preferences		FLUTTER	ANDROID IOS LINUX MACOS WEB WINDOWS	A storage implementation using the shared_preferences package
stash_secure_storage		FLUTTER	ANDROID IOS LINUX MACOS WEB WINDOWS	A storage implementation using the flutter_secure_storage package

Library Integrations

There's also some integrations with well know dart libraries

Package	Pub	Description
stash_dio		Integrates with the Dio HTTP client

Test Support

Finally a testing library is provided to aid in the development of third party storage backend extensions

Package	Pub	Description
stash_test		Testing support for stash extensions

Getting Started

Select one of the storage libraries and add the api and storage packages to your pubspec.yaml replacing x.x.x and y.y.y with the latest version of both. The example below uses the stash_memory package which provides an in-memory storage:

dependencies:
    stash_api: ^x.x.x
    stash_memory: ^y.y.y

Run the following command to install dependencies:

dart pub get

Finally, to start developing import the api and the selected storage implementation. In the example bellow the selected storage implementation is stash_memory thus we import the stash_apiand the stash_memory libraries:

import 'package:stash/stash_api.dart';
import 'package:stash/stash_memory.dart';
// In a more general sense 'package:stash/stash_xxx.dart' where xxx is the name of the
// storage provider, memory, hive and so on

Usage

Start by creating an instance of the storage backend and to decide if the stash needs caching or plain key-value semantics. For the former we need a cache while for the latter we need a vault. For example, to use an in-memory vault you we need to call the newMemoryVaultStore function which is exported by the stash_memory package. This function allows the user to bootstrap a in-memory store for vaults. Note that there is a similar one for caches dubbed newMemoryCacheStore

  // Create a in-memory store
  final store = await newMemoryVaultStore();
  // In a more general sense 'newXXXVaultStore' or 'newXXXCacheStore' where xxx is the name of the storage provider,
  // memory, file, sqlite, hive and so on

Then you can create as many vault's as you wish from this store (for caches you need a different storage). Let's create one to store string's and add an element to it

  // Creates a vault from the previously created store
  final stringVault = await store.vault<String>();

  // Adds a new value to the vault
  await stringVault.put('key1', 'value1');

Let's create a untyped vault and add two values to it with different types

  // Creates a second vault from the previously created store
  final anyVault = await store.vault();

    // Adds a string value to the vault
  await anyVault.put('key1', 'value1');

    // Adds an int value to the vault
  await anyVault.put('key1', 1);

On both cases if the name of the vault is not provided a uuid is automatically assigned as the name. By using different names you can reuse the same storage and partition your data accordingly

Now let's create a cache. In this case a new type of store is needed as the cache needs additional fields to support expiry and eviction. In the example bellow a store is created followed by a cache with a max capacity of 10. Note: that the eviction policy is only applied if maxEntries is specified

  // Creates a in-memory store
  final store = await newMemoryCacheStore();

  // Creates a cache with a max capacity of 10 from the previously created store
  final cache = await store.cache<String>(maxEntries: 10);

Then add a element to the cache:

  // Adds a 'value1' under 'key1' to the cache
  await cache.put('key1', 'value1');

Finally, retrieve that element:

  // Retrieves the value from the cache
  final value = await cache.get('key1');

The in-memory example is the simplest one, on that case there is no persistence so encoding/decoding of elements is not needed. Conversely when the storage mechanism uses persistence and we need to add custom objects they need to be json serializable and the appropriate configuration provided to allow the serialization/deserialization of those objects. This means that on those cases additional configuration is needed to allow the serilization/deserialization to happen.

Find bellow and example that uses stash_file as the storage implementation. In this case an object is stored, so in order to deserialize it the user needs to provide a way to decode it, like so: fromEncodable: (json) => Task.fromJson(json). The lambda should make a call to a user provided function that deserializes the object. Conversly, the serialization happens by convention i.e. by calling the toJson method on the object. Note that this example is sufficiently simple to warrant the usage of manual coded functions to serialize/deserialize the objects but it could be paired with the json_serializable package or similar for the automatic generation of the Json serialization / deserialization code.

import 'dart:io';

import 'package:stash/stash_api.dart';
import 'package:stash_file/stash_file.dart';

class Task {
  final int id;
  final String title;
  final bool completed;

  Task({required this.id, required this.title, this.completed = false});

  /// Creates a [Task] from json map
  factory Task.fromJson(Map<String, dynamic> json) => Task(
      id: json['id'] as int,
      title: json['title'] as String,
      completed: json['completed'] as bool);

  /// Creates a json map from a [Task]
  Map<String, dynamic> toJson() =>
      <String, dynamic>{'id': id, 'title': title, 'completed': completed};

  @override
  String toString() {
    return 'Task $id, "$title" is ${completed ? "completed" : "not completed"}';
  }
}

void main() async {
  // Temporary directory
  final path = Directory.systemTemp.path;

  // Creates a store
  final store = await newFileLocalCacheStore(path: path);

  // Creates a cache with a capacity of 10 from the previously created store
  final cache = await store.cache<Task>(
      name: 'cache1',
      fromEncodable: (json) => Task.fromJson(json),
      maxEntries: 10,
      eventListenerMode: EventListenerMode.synchronous)
    ..on<CacheEntryCreatedEvent<Task>>().listen(
        (event) => print('Key "${event.entry.key}" added to the cache'));

  // Adds a task with key 'task1' to the cache
  await cache.put(
      'task1', Task(id: 1, title: 'Run cache store example', completed: true));
  // Retrieves the value from the cache
  print(await cache.get('task1'));
}

Notice that fromEncodable function is configured per cache which means you can configure a cache with a different type using the same store. This was not previously possible as before version 4.3.0 the fromEncodable function was configured per store. That has changed with version 4.3.0 which introduced this feature. Find bellow a complete example that stores two different objects in the same store and creates two vaults using two different fromEncodable functions.

import 'dart:io';

import 'package:stash/stash_api.dart';
import 'package:stash_file/stash_file.dart';

class Task {
  final int id;
  final String title;
  final bool completed;

  Task({required this.id, required this.title, this.completed = false});

  /// Creates a [Task] from json map
  factory Task.fromJson(Map<String, dynamic> json) => Task(
      id: json['id'] as int,
      title: json['title'] as String,
      completed: json['completed'] as bool);

  /// Creates a json map from a [Task]
  Map<String, dynamic> toJson() =>
      <String, dynamic>{'id': id, 'title': title, 'completed': completed};

  @override
  String toString() {
    return 'Task $id, "$title" is ${completed ? "completed" : "not completed"}';
  }
}

class Contact {
  final int id;
  final String name;

  Contact({required this.id, required this.name});

  /// Creates a [Contact] from json map
  factory Contact.fromJson(Map<String, dynamic> json) =>
      Contact(id: json['id'] as int, name: json['name'] as String);

  /// Creates a json map from a [Contact]
  Map<String, dynamic> toJson() => <String, dynamic>{'id': id, 'name': name};

  @override
  String toString() {
    return 'Contact $id, "$name"';
  }
}

void main() async {
  // Temporary directory
  final path = Directory.systemTemp.path;

  // Creates a store
  final store = await newFileLocalVaultStore(path: path);

  // Creates a vault that stores Tasks from the previously created store
  final taskVault = await store.vault<Task>(
      name: 'taskVault',
      fromEncodable: (json) => Task.fromJson(json),
      eventListenerMode: EventListenerMode.synchronous)
    ..on<VaultEntryCreatedEvent<Task>>().listen(
        (event) => print('Key "${event.entry.key}" added to the task vault'));

  // Creates a vault that stores Contacts from the previously created store
  final contactVault = await store.vault<Contact>(
      name: 'contactVault',
      fromEncodable: (json) => Contact.fromJson(json),
      eventListenerMode: EventListenerMode.synchronous)
    ..on<VaultEntryCreatedEvent<Contact>>().listen((event) =>
        print('Key "${event.entry.key}" added to the contact vault'));

  // Adds a task with key 'task1' to the vault
  await taskVault.put('task1',
      Task(id: 1, title: 'Run task vault store example', completed: true));

  // Adds a contact with key 'contact1' to the vault
  await contactVault.put(
      'contact1', Contact(id: 1, name: 'Run contact vault store example'));

  // Retrieves the value from the task vault
  print(await taskVault.get('task1'));

  // Retrieves the value from the contact vault
  print(await contactVault.get('contact1'));
}

It's possible to automate the fetching of value whenever a vault or cache entry is not present or when the entry expired. You just need to provide a vaultLoader or a cacheLoader and when some entry is absent or expired stash can retrieve a new value for the specified key. The loader is configured when creating a vault or a cache e.g. store.cache(cacheLoader: (key) => ...). Note that this function must return a Future.

Vault

Operations

The Vault frontend provides a number of operations which are presented in the table bellow:

Events

The user of a Vault can subscribe to entry events if they are enabled as, by default, no events are propagated. The three possible configurations are setted via the eventListenerMode parameter. When creating a Vault the default value is Disabled as in, no events are published, but can be configured with Sync providing synchronous events or with Async in which case it provides assynchronous events. A user can subscribe all events or only to a specific set of events. On the example bellow a memory vault is created subscribing all the events synchronously.

  // Creates a vault from a previously created store subscribing to the created, updated, removed events
  final vault = await store.vault<String>(
      eventListenerMode: EventListenerMode.synchronous)
    ..on<VaultEntryCreatedEvent<String>>().listen(
        (event) => print('Key "${event.entry.key}" added'))
    ..on<VaultEntryUpdatedEvent<String>>().listen(
        (event) => print('Key "${event.newEntry.key}" updated'))
    ..on<VaultEntryRemovedEvent<String>>().listen(
        (event) => print('Key "${event.entry.key}" removed'));

Statistics

The user of a Vault can collect a number of statistics about it's usage by enabling stats through the statsEnabled flag. He can also provide a custom implementation of the VaultStats interface allowing custom statistics backends. The VaultStats interface defines a number of statistics that the user can collect through the following operations:

Cache

Operations

The Cache frontend provides a number of operations which are presented in the table bellow.

Types

To create a Cache we can use the function exported by a specific storage library, newMemoryCacheStore in case of the stash_memory library (generically newXXXCacheStore where xxx is the name of the storage provider).

Note that this is not the only type of cache provided, stash also provies a tiered cache which can be created with a call to newTieredCache function which is exported by the base stash library. It allows the creation of a cache that uses primary and secondary cache surrogates. The idea is to have a fast in-memory cache as the primary and a persistent cache as the secondary altough other combinations are definitely supported. In this cases it's normal to have a bigger capacity for the secondary and a lower capacity for the primary cache. In the example bellow a new tiered cache is created using two in-memory caches the first with a maximum capacity of 10 and the second with unlimited capacity.

  // Creates a in-memory store
  final store = await newMemoryCacheStore();

  /// Creates a tiered cache with both the primary and the secondary caches using 
  /// a memory based storage. The first cache with a maximum capacity of 10 and 
  /// the second with unlimited capacity
  final cache = newTieredCache(
      await store.cache<String>(maxEntries: 10),
      await store.cache<String>());

A more common use case is to have the primary cache using a memory storage and the secondary a cache backed by a persistent storage like the one provided by stash_file or stash_sqlite packages. The example bellow illustrates one of those use cases with the stash_file package as the provider of the storage backend of the secondary cache.

  // Creates a in-memory store
  final memoryStore = await newMemoryCacheStore();

  // Creates a file store
  final fileStore = await newFileLocalCacheStore(path: Directory.systemTemp.path);  

  final cache = newTieredCache(
      await memoryStore.cache<String>(name: 'memoryCache', maxEntries: 10),
      await fileStore.cache(name: 'diskCache', maxEntries: 1000));

Expiry Policies

It's possible to define how the expiration of cache entries works based on creation, access and modification operations. A number of pre-defined expiry polices are provided out-of-box that define multiple combinations of those interactions. Note that, most of the expiry policies can be configured with a specific duration which is used to increase the expiry time when some type of operation is executed on the cache. This mechanism was heavily inspired on the JCache expiry semantics. By default the configuration does not enforce any kind of expiration, thus it uses the Eternal expiry policy. It is of course possible to configure an alternative expiry policy setting the expiryPolicy parameter e.g. store.cache(expiryPolicy: const AccessedExpiryPolicy(Duration(days: 1))). Another alternative is to configure a custom expiry policy through the implementation of the ExpiryPolicy interface.

Eviction Policies

As discussed Cache supports eviction and provides a number of pre-defined eviction policies that are described in the table bellow. Note that it's mandatory to configure the cache with a number for maxEntries e.g. store.cache(maxEntries: 10). Without this configuration the eviction algorithm is not triggered since there is no limit defined for the number of items on the cache. The default algorithm is LRU (least-recently used) but other algorithms can be configured through the use of the evictionPolicy parameter e.g. store.cache(evictionPolicy: const HyperbolicEvictionPolicy()). Another alternative is to configure a custom eviction policy through the implementation of the EvictionPolicy interface.

When the maximum capacity of a cache is exceeded eviction of one or more entries is inevitable. At that point the eviction algorithm works with a set of entries that are defined by the sampling strategy used. In the default configuration the whole set of entries is used which means that the cache statistics will be retrieved from each and every one of the entries. This works fine for modest sized caches but can became a performance burden for bigger caches. On that cases a more efficient sampling strategy should be selected to avoid sampling the whole set of entities from storage. On those cases it's possible to configure the sampling strategy with the sampler parameter e.g. newMemoryCache(sampler: ReservoirSampler(0.5)) uses a ReservoirSampler sampler to select only half of the entries as candidates for eviction. The configuration of a custom sampler is also possible through the implementation of the KeySampler interface.

Events

The user of a Cache can subscribe to entry events if they are enabled as, by default, no events are propagated. The three possible configurations are setted via the eventListenerMode parameter. When creating a Cache the default value is Disabled as in, no events are published, but can be configured with Sync providing synchronous events or with Async in which case it provides assynchronous events. A user can subscribe all events or only to a specific set of events. On the example bellow a memory cache is created with a maximum of 10 entries subscribing all the events synchronously.

  // Creates a cache with a capacity of 10 from a previously created store subscribing to the created, updated, removed, expired and evicted events
  final cache = await store.cache<String>(
      maxEntries: 10,
      eventListenerMode: EventListenerMode.synchronous)
    ..on<CacheEntryCreatedEvent<String>>().listen(
        (event) => print('Key "${event.entry.key}" added'))
    ..on<CacheEntryUpdatedEvent<String>>().listen(
        (event) => print('Key "${event.newEntry.key}" updated'))
    ..on<CacheEntryRemovedEvent<String>>().listen(
        (event) => print('Key "${event.entry.key}" removed'))
    ..on<CacheEntryExpiredEvent<String>>().listen(
        (event) => print('Key "${event.entry.key}" expired'))
    ..on<CacheEntryEvictedEvent<String>>().listen(
        (event) => print('Key "${event.entry.key}" evicted')); 

Statistics

The user of a Cache can collect a number of statistics about it's usage by enabling stats through the statsEnabled. He can also provide a custom implementation of the CacheStats interface allowing custom statistics backends.

The CacheStats interface defines a number of statistics that the user can collect through the following operations:

Contributing

Contributions are always welcome!

If you would like to contribute with other parts of the API, feel free to make a Github pull request as I'm always looking for contributions for:

• Tests
• Documentation
• New APIs

See CONTRIBUTING.md for ways to get started.

Thanks goes to these wonderful people (emoji key), that have already contributed to the project:

Gabriel Terwesten 📖 💻 ⚠️ 💡

Mike Hoolehan ⚠️

Thomas Järvstrand 🐛

Romain Franceschini 💻

Simon Stevens 💻 🐛

This project follows the all-contributors specification. Contributions of any kind welcome!

Features and Bugs

Please file feature requests and bugs at the issue tracker.

License

This project is licensed under the MIT License - see the LICENSE file for details