moneta

Moneta: A unified interface for key/value stores

Gem Version Build Status Code Climate Flattr this git repo

Moneta provides a standard interface for interacting with various kinds of key/value stores. Moneta supports the well-known NoSQL and document based stores.

A short overview of the features:

If you are not yet convinced, you might ask why? What are the goals of the project?

Moneta is tested thoroughly using GitHub Actions.

Getting started

Install Moneta via Rubygems

$ gem install moneta

or add it to your Gemfile

gem 'moneta'

Now you are ready to go:

require 'moneta'

# Create a simple file store
store = Moneta.new(:File, dir: 'moneta')

# Store some entries
store['key'] = 'value'

# Read entry
store.key?('key') # returns true
store['key'] # returns 'value'

store.close

In case you are wondering, Moneta uses Semantic Versioning since v1.0.0.

Supported backends

Out of the box, it supports the following backends. Use the backend name symbol in the Moneta constructor (e.g. Moneta.new(:Memory)).

Some of the backends are not exactly based on key/value stores, e.g. the relational ones. These are useful if you already use the corresponding backend in your application. You get a key/value store for free then without installing any additional services and you still have the possibility to upgrade to a real key/value store.

Backend feature matrix

NOTE: The backend matrix is much more readable on rubydoc.info than on github. Go there!

AdapterRequired gemsMRI support1JRuby support1Multi-thread safe2Multi-process safe3Atomic increment4Atomic create5Native expires6PersistentKey TraversalBulk read7Bulk write8Description
Persistent stores
Mongomongo✓✓✓✓✓✓✓✓✓✓✓MongoDB database
Redisredis✓✓✓✓✓✓✓✓✓✓✓Redis database
ActiveRecordactiverecord✓✓✓✓✓✓✗✓✓✓✓ActiveRecord ORM
File-✓✓✓✓✓✓✗✓✓✗✗File store
LMDBlmdb✓✗✓✓✓✓✗✓✓✓✓Symas Lightning Memory-Mapped Database (LMDB)
Sequelsequel✓✓✓✓✓✓✗✓✓✓✓Sequel ORM
TokyoTyranttokyotyrant or ruby-tokyotyrant✓✓✗✓✓✓✗✓✗✓✗TokyoTyrant database
PStore-✓✗✗✓9✓✓✗✓✓✓✓PStore store
YAML-✓✓✗✓9✓✓✗✓✓✗✗YAML store
Sqlitesqlite3✓✗?✓9✓✓✗✓✓✓✓Sqlite3 database
Daybreakdaybreak✓✓✗✓✓✓✗✓✓✗✓Incredibly fast pure-ruby key/value store Daybreak
DBM-✓✗✗✗✓✓✗✓✓✓✓Berkeley DB using DBM interface or NDBM (Depends on Ruby environment)
GDBMffi-gdbm on JRuby✓✓✗✗✓✓✗✓✓✓✓GDBM database
LevelDBleveldb✓✗✗✗✓✓✗✓✓✓✓LevelDB database
SDBM-✓✗✗✗✓✓✗✓✓✓✓SDBM database
TDBtdb✓✗✗✗✓✓✗✓✓✗✗TDB database
KyotoCabinetkyotocabinet-ruby or kyotocabinet-ruby-reanimated✓✗✗✗✓✓✗✓✓✓✓KyotoCabinet database
TokyoCabinettokyocabinet✓✗✗✗✓✓✗✓✓✗✗TokyoCabinet database
DataMapperdm-core, dm-migrations✓✗✓✓✗✓✗✓✗✗✗DataMapper ORM
Couchfaraday, multi_json✓✓✗✓✗✓✗✓✓✓✓CouchDB database
HBasehbaserb✗✗?✓✓✗✗✓✗✗✗HBase database
Cassandracassandra✓✓?✓✗✗✓✓✓✓✓Cassandra distributed database
LocalMemCachelocalmemcache✓✗✓✓✗✗✗✓✗✗✗LocalMemCache database
Fogfog✓✓?✓✗✗✗✓✗✗✗Fog cloud store
Riakriak-client✗✗✗✓✗✗✗✓✗✗✗Riak database
Non-persistent stores
MemcachedDallidalli✓✓✓✓✓✓✓✗10✗✓✓Memcached database with Dalli library
Memcacheddalli or memcached✓?11?11✓✓✓✓✗10✗?11?11Memcached database
MemcachedNativememcached✓✗✗✓✓✓✓✗10✗✗✗Memcached database with native library
Cookie-✓✓✗✓12✓✓✓✗✓✗✗Cookie in memory store
LRUHash-✓✓✗✓12✓✓✗✗✓✗✗LRU memory store
Memory-✓✓✗✓12✓✓✗✗✓✓✓Memory store
Null-✓✓✓✓✗✗✗✗✗✗✗No database
Network clients
Client-✓✓✗✓?13?13?13?13?13✗✗Moneta client adapter
RestClient-✓✓✗✓✗✗✗?13✗✗✗Moneta REST client adapter
  1. Indicates that the adapter is expected to work on this platform. Most adapters will at least work on MRI, but some are curently considered unstable, in which case they are not supported on any platform.
  2. Make adapters thread-safe by using Moneta::Lock or by passing the option threadsafe: true to Moneta#new. There is also Moneta::Pool which can be used to share a store between multiple threads if the store is multi-process safe. I recommend to add the option :threadsafe to ensure thread-safety since for example under JRuby and Rubinius even the basic datastructures are not thread safe due to the lack of a global interpreter lock (GIL). This differs from MRI where some adapters might appear thread safe already but only due to the GIL.
  3. Share a Moneta store between multiple processes using Moneta::Shared (See below).
  4. If a store provides atomic increment it can be used with Moneta::Semaphore. You can add weak #increment support using the Moneta::WeakIncrement proxy.
  5. If a store provides atomic creation it can be used with Moneta::Mutex. You can add weak #create support using the Moneta::WeakCreate proxy.
  6. Add expiration support by using Moneta::Expires or by passing the option expires: true to Moneta#new.
  7. This indicates that there is some performance gain when fetching multiple values at once using #values_at/#fetch_values or #slice. For instance, the MGET instruction in Redis, or the ability to retrieve several rows in one query in SQL.
  8. This indicates that there is some performance gain when storing multiple key/value pairs at once using #merge!/#update.
  9. Sqlite/YAML/PStore are multiprocess safe, but the performance suffers badly since the whole database file must be locked for writing. Use a key/value server if you want multiprocess concurrency!
  10. There are some servers which use the memcached protocol but which are persistent (e.g. MemcacheDB, Kai, IronCache, Roma, Flare and Kumofs)
  11. This feature is only available if the dalli backend is selected
  12. Store is multi-process safe because it is an in-memory store, values are not shared between multiple processes
  13. Depends on server

Proxies

In addition it supports proxies (Similar to Rack middlewares) which add additional features to storage backends:

Check the YARD documentation for more information and examples.

Serializers and compressors (Moneta::Transformer)

Supported serializers:

Supported value compressors:

Supported encoders:

Special transformers:

Moneta API

The Moneta API is purposely extremely similar to the Hash API with a few minor additions. Every method takes also a optional option hash. In order so support an identical API across stores, Moneta does not support partial matches.

#initialize(options)                      options differs per-store, and is used to set up the store.

#[](key)                                  retrieve a key. If the key is not available, return nil.

#load(key, options = {})                  retrieve a key. If the key is not available, return nil.

#fetch(key, options = {}, &block)         retrieve a key. If the key is not available, execute the
                                          block and return its return value.

#fetch(key, value, options = {})          retrieve a key. If the key is not available, return the value,

#[]=(key, value)                          set a value for a key. If the key is already used, clobber it.
                                          keys set using []= will never expire.

#store(key, value, options = {})          same as []=, but you can supply options.

#delete(key, options = {})                delete the key from the store and return the current value.

#key?(key, options = {})                  true if the key exists, false if it does not.

#increment(key, amount = 1, options = {}) increment numeric value. This is an atomic operation
                                          which is not supported by all stores. Returns current value.

#decrement(key, amount = 1, options = {}) increment numeric value. This is an atomic operation
                                          which is not supported by all stores. Returns current value.
                                          This is just syntactic sugar for incrementing with a negative value.

#create(key, value, options = {})         create entry. This is an atomic operation which is not supported by all stores.
                                          Returns true if the value was created.

#values_at(*keys, **options)              retrieve multiple keys. Returns an array of equal length to the keys.
                                          Each entry in the array is either the value corresponding to the key
                                          in the same position, or nil if the key is not available.

#fetch_values(*keys, **options, &block)   retrieve multiple keys. Return is identical to values_at, except that
                                          when a block is given it will be called once for each key that is not
                                          available, and the return value of the block will be used in place of
                                          nil in the array.

#slice(*keys, **options)                  retrieve multiple keys. Returns an enumerable of key-value pairs,
                                          one for each of the supplied keys that is present in the store.

#merge!(pairs, options = {})              set values for multiple keys. "pairs" must be an enumerable of
                                          key-value pairs to be stored. Any existing keys will be clobbered.

#merge!(pairs, options = {}, &block)      set values for multiple keys. For each existing key, execute the block
                                          passing the key, existing value and new value, and store the return
                                          value.

#update(pairs, options = {}, &block)      same as merge!

#each_key                                 return an enumerable which will yield all keys in the store, one at a
                                          time. This method is present if and only if the store supports the
                                          :each_key feature.

#each_key(&block)                         yield all keys in the store to the block, one at a time. Again, this
                                          method is present if and only if the store supports the :each_key
                                          feature.

#clear(options = {})                      clear all keys in this store.

#close                                    close database connection.

#features                                 return array of features, e.g. [:create, :expires, :increment]

#supports?(feature)                       returns true if store supports a given feature

Creating a Store

There is a simple interface to create a store using Moneta.new. You will get automatic key and value serialization which is provided by Moneta::Transformer. This allows you to store arbitrary Ruby objects. You can tune some options when you call Moneta.new. However for very fine tuning use Moneta.build.

store = Moneta.new(:Memcached, server: 'localhost:11211')
store['key'] = 'value'
store['hash_key'] = {a: 1, b: 2}
store['object_key'] = MarshallableRubyObject.new

If you want to have control over the proxies, you have to use Moneta.build:

store = Moneta.build do
  # Adds expires proxy
  use :Expires

  # Transform key using Marshal and Base64 and value using Marshal
  use :Transformer, key: [:marshal, :base64], value: :marshal

  # IMPORTANT: adapter must be defined last for the builder to function properly.

  # Memory backend
  adapter :Memory
end

You can also directly access the underlying adapters if you don’t want to use the Moneta stack.

db = Moneta::Adapters::File.new(dir: 'directory')
db['key'] = {a: 1, b: 2} # This will fail since you can only store Strings

# However for Mongo and Couch this works
# The hash will be mapped directly to a Mongo/Couch document.
db = Moneta::Adapters::Couch.new
db['key'] = {a: 1, b: 2}

db = Moneta::Adapters::Mongo.new
db['key'] = {a: 1, b: 2}

Expiration

The Cassandra, Memcached, Redis and Mongo backends support expiration natively.

cache = Moneta::Adapters::Memcached.new

# Or using the builder...
cache = Moneta.build do
  adapter :Memcached
end

# Expires in 60 seconds
cache.store(key, value, expires: 60)

# Never expire
cache.store(key, value, expires: 0)
cache.store(key, value, expires: false)

# Update expires time if value is found
cache.load(key, expires: 30)
cache.key?(key, expires: 30)

# Or remove the expiration if found
cache.load(key, expires: false)
cache.key?(key, expires: 0)

You can add the expires feature to other backends using the Moneta::Expires proxy. But be aware that expired values are not deleted automatically if they are not looked up.

# Using the :expires option
cache = Moneta.new(:File, dir: '...', expires: true)

# or manually by using the proxy...
cache = Moneta::Expires.new(Moneta::Adapters::File.new(dir: '...'))

# or using the builder...
cache = Moneta.build do
  use :Expires
  adapter :File, dir: '...'
end

Key traversal

Where supported by the store’s backend, it is possible to traverse the keys in the store using the #each_key method. Support for this can be tested by calling store.supports?(:each_key), or checking for the presence of :each_key in store.features.

store.each_key # returns an Enumerable
store.each_key do |key|
  store.load(key)  # read operations are supported within the block
  store[key] = "x" # behaviour of write operations is undefined
end

Atomic operations

Atomic incrementation and raw access

The stores support the #increment which allows atomic increments of unsigned integer values. If you increment a non existing value, it will be created. If you increment a non integer value an exception will be raised.

store.increment('counter')     # returns 1, counter created
store.increment('counter')     # returns 2
store.increment('counter', -1) # returns 1
store.increment('counter', 13) # returns 14
store.increment('counter', 0)  # returns 14
store.decrement('counter')     # returns 13
store['name'] = 'Moneta'
store.increment('name')        # raises an Exception

If you want to access the counter value you have to use raw access to the datastore. This is only important if you have a Moneta::Transformer somewhere in your proxy stack which transforms the values e.g. with Marshal.

store.increment('counter')          # returns 1, counter created
store.load('counter', raw: true) # returns 1

store.store('counter', '10', raw: true)
store.increment('counter') # returns 11

Fortunately there is a nicer way to do this using some syntactic sugar!

store.increment('counter') # returns 1, counter created
store.raw['counter']       # returns 1
store.raw.load('counter')  # returns 1

store.raw['counter'] = '10'
store.increment('counter') # returns 11

You can also keep the raw store in a variable and use it like this:

counters = store.raw

counters.increment('counter') # returns 1, counter created
counters['counter']           # returns 1
counters.load('counter')      # returns 1

counters['counter'] = '10'
counters.increment('counter') # returns 11

Atomic create

The stores support the #create which allows atomic creation of entries. #create returns true if the value was created.

store.create('key', 'value') # returns true
store.create('key', 'other value') # returns false

Atomic bulk operations

All stores support storage and retrieval of multiple keys using #values_at/#fetch_values/#slice and #merge!/#update. Wherever possible, these operations are performed atomically. When this is not possible, the #load and #store methods are called once for each key.

store.merge!('key1' => 'value1', 'key2' => 'value2') # stores two keys
store.values_at('key1', 'key2', 'key3') # returns ['value1', 'value2', nil]
store.fetch('key1', 'key3') { |k| k + ' missing' } # returns ['key1', 'key3 missing']
store.slice('key1', 'key2', 'key3') # returns enumerable of ['key1', 'value1'], ['key2', 'value2']

store.merge!('key2' => 'new value2', 'key3' => 'value3') do |key, value, new_value|
  [value, new_value].join('+')
end # stores "value3" and "value2+new value2"

Shared/distributed synchronization primitives

Moneta provides shared/distributed synchronization primitives which are shared database-wide between all clients.

Moneta::Mutex allows a single thread to enter a critical section.

mutex = Moneta::Mutex.new(store, 'mutex_key')

mutex.synchronize do
   mutex.locked? # returns true

   # Synchronized access to counter
   store['counter'] += 1
end

begin
  mutex.lock
  mutex.locked? # returns true
  # ...
ensure
  mutex.unlock
end

Moneta::Semaphore allows max_concurrent threads to enter a critical section.

semaphore = Moneta::Semaphore.new(store, 'semaphore_counter', max_concurrent)

semaphore.synchronize do
   semaphore.locked? # returns true
   # ...
end

begin
  semaphore.enter
  semaphore.locked? # returns true
  # ...
ensure
  semaphore.leave
end

Weak atomic operations

If an underlying adapter doesn’t provide atomic #create or #increment and #decrement you can use the proxies Moneta::WeakIncrement and Moneta::WeakCreate to add support without atomicity.

But then you have to ensure that the store is not shared by multiple processes and thread-safety is provided by Moneta::Lock.

Syntactic sugar and option merger

For raw data access as described before the class Moneta::OptionMerger is used. It works like this:

# All methods after 'with' get the options passed
store.with(raw: true).load('key')

# You can also specify the methods
store.with(raw: true, only: :load).load('key')
store.with(raw: true, except: [:key?, :increment]).load('key')

# Syntactic sugar for raw access
store.raw.load('key')

# Access substore where all keys get a prefix
substore = store.prefix('sub')
substore['key'] = 'value'
store['key']    # returns nil
store['subkey'] # returns 'value'

# Set expiration time for all keys
short_lived_store = store.expires(60)
short_lived_store['key'] = 'value'

Add proxies to existing store

You can add proxies to an existing store. This is useful if you want to compress only a few values for example.

compressed_store = store.with(prefix: 'compressed') do
  use :Transformer, value: :zlib
end

store['key'] = 'this value will not be compressed'
compressed_store['key'] = 'value will be compressed'

Framework Integration

Inspired by redis-store there exist integration classes for Rails and Rack/Rack-Cache. You can also use all the Rack middlewares together with Rails and the Sinatra framework. There exist the following integration classes:

Rack

Session store

You can use Moneta as a Rack session store. Use it in your config.ru like this:

require 'rack/session/moneta'

# Use only the adapter name
use Rack::Session::Moneta, store: :Redis

# Use Moneta.new
use Rack::Session::Moneta, store: Moneta.new(:Memory, expires: true)

# Set rack options
use Rack::Session::Moneta, key: 'rack.session',
domain: 'foo.com',
path: '/',
expire_after: 2592000,
store: Moneta.new(:Memory, expires: true)

# Use the Moneta builder
use Rack::Session::Moneta do
  use :Expires
  adapter :Memory
end

Moneta middleware

There is a simple middleware which places a Moneta store in the Rack environment at env['rack.moneta_store']. It supports per-request caching if you add the option cache: true. Use it in your config.ru like this:

require 'rack/moneta_store'

# Add Rack::MonetaStore somewhere in your rack stack
use Rack::MonetaStore, :Memory, cache: true

run lambda { |env|
  env['rack.moneta_store'] # is a Moneta store with per-request caching
}

# Pass it a block like the one passed to Moneta.build
use Rack::MonetaStore do
  use :Transformer, value: :zlib
  adapter :Cookie
end

run lambda { |env|
  env['rack.moneta_store'] # is a Moneta store without caching
}

REST server

If you want to expose your Moneta key/value store via HTTP, you can use the Rack/Moneta REST service. Use it in your config.ru like this:

require 'rack/moneta_rest'

map '/moneta' do
  run Rack::MonetaRest.new(:Memory)
end

# Or pass it a block like the one passed to Moneta.build
run Rack::MonetaRest.new do
  use :Transformer, value: :zlib
  adapter :Memory
end

Rack-Cache

You can use Moneta as a Rack-Cache store. Use it in your config.ru like this:

require 'rack/cache/moneta'

use Rack::Cache,
      metastore:   'moneta://Memory?expires=true',
      entitystore: 'moneta://Memory?expires=true'

# Or used named Moneta stores
Rack::Cache::Moneta['named_metastore'] = Moneta.build do
  use :Expires
  adapter :Memory
end
use Rack::Cache,
      metastore: 'moneta://named_metastore',
      entity_store: 'moneta://named_entitystore'

Cookies

Use Moneta to store cookies in Rack. It uses the Moneta::Adapters::Cookie. You might wonder what the purpose of this store or Rack middleware is: It makes it possible to use all the transformers on the cookies (e.g. :prefix, :marshal and :hmac for value verification).

require 'rack/moneta_cookies'

use Rack::MonetaCookies, domain: 'example.com', path: '/path'
run lambda { |env|
  req = Rack::Request.new(env)
  req.cookies #=> is now a Moneta store!
  env['rack.request.cookie_hash'] #=> is now a Moneta store!
  req.cookies['key'] #=> retrieves 'key'
  req.cookies['key'] = 'value' #=> sets 'key'
  req.cookies.delete('key') #=> removes 'key'
  [200, {}, []]
}

Rails

Session store

Add the session store in your application configuration config/environments/*.rb.

require 'action_dispatch/middleware/session/moneta_store'

# Only by adapter name
config.cache_store :moneta_store, store: :Memory

# Use Moneta.new
config.cache_store :moneta_store, store: Moneta.new(:Memory)

# Use the Moneta builder
config.cache_store :moneta_store, store: Moneta.build do
  use :Expires
  adapter :Memory
end

Cache store

Add the cache store in your application configuration config/environments/*.rb. Unfortunately the Moneta cache store doesn’t support matchers. If you need these features use a different server-specific implementation.

require 'active_support/cache/moneta_store'

# Only by adapter name
config.cache_store :moneta_store, store: :Memory

# Use Moneta.new
config.cache_store :moneta_store, store: Moneta.new(:Memory)

# Use the Moneta builder
config.cache_store :moneta_store, store: Moneta.build do
  use :Expires
  adapter :Memory
end

Padrino

Padrino adopted Moneta to replace their cache stores in padrino-cache. You use it like this

# Global Padrino caching
# Don't forget the expires: [true, Integer] if you want expiration support!
Padrino.cache = Moneta.new(:Memory, expires: true)

# Application caching
# Don't forget the expires: [true, Integer] if you want expiration support!
set :cache, Moneta.new(:Memory, expires: true)

Advanced

Build your own key value server

You can use Moneta to build your own key/value server which is shared between multiple processes. If you run the following code in two different processes, they will share the same data which will also be persistet in the database shared.db.

require 'moneta'

store = Moneta.build do
  use :Transformer, key: :marshal, value: :marshal
  use :Shared do
    use :Cache do
      cache do
        adapter :LRUHash
      end
      backend do
        adapter :GDBM, file: 'shared.db'
      end
    end
  end
end

If you want to go further, you might want to take a look at Moneta::Server and Moneta::Adapters::Client which are used by Moneta::Shared and provide the networking communication. But be aware that they are experimental and subjected to change. They provide an acceptable performance (for being ruby only), but don’t have a stable protocol yet.

You might wonder why I didn’t use DRb to implement server and client - in fact my first versions used it, but with much worse performance and it was real fun to implement the networking directly :) There is still much room for improvement and experiments, try EventMachine, try Kgio, …

ToyStore ORM

If you want something more advanced to handle your objects and relations, use John Nunemaker’s ToyStore which works together with Moneta. Assuming that Person is a ToyStore::Object you can add persistence using Moneta as follows:

# Use the Moneta Redis backend
Person.adapter :memory, Moneta.new(:Redis)

Testing and Benchmarks

Testing is done using GitHub Actions. Currently we support MRI Ruby >= 2.4.0 (but not yet 3.x) and the JRuby >= 9.2.9.0. MRI 2.3.0 should mostly still work, but is no longer tested in CI.

Benchmarks for each store are done on Travis-CI for each build. At the time of writing, benchmarks still need to be migrated from Travis to GitHub Actions. Take a look there to compare the speed of the different key value stores for different key/value sizes and size distributions. Feel free to add your own configurations! The impact of Moneta should be minimal since it is only a thin layer on top of the different stores.

How to contribute?

Always feel free to open an issue on https://github.com/moneta-rb/moneta/issues if something doesn’t work as you expect it to work. Feedback is also very welcome!

My only request about patches is that you please try to test them before submitting.

Contribute an adapter

If you want support for another adapter you can at first at it to the list of missing adapters at https://github.com/moneta-rb/moneta/issues/16

If you choose to implement an adapter please also add tests. Please check also if anything in .github/workflows needs changes, for example if you need to start additional services.

Check if the default settings in Moneta#new are appropriate for your adapter. If not specify a better one.

Don’t forget to edit the README.md and the CHANGES.

Alternatives

Authors

This site is open source. Improve this page.