GitHub - nateware/redis-objects: Map Redis types directly to Ruby objects

Important 2.0 changes

Several longstanding bugs have been addressed in this release. However, this may require some code changes as part of the upgrade.

Renaming of `lock` Method

The lock method that collided with ActiveRecord::Base has been renamed redis_lock. This means your classes need to be updated to call redis_lock instead:

class YouClassNameHere < ActiveRecord::Base
  include Redis::Objects
  redis_lock :mylock  # formerly just "lock"
end

For more details on the issue and fix refer to #196.

New key naming method

A new method to determine the internal key naming scheme has been added to fix a longstanding bug. (Refer to #213) By default, backwards compatibility is maintained, but this means that in some cases, Nested classes (for example Dog::Behavior and Cat::Behavior) would have key names that collide (they would both start with Behavior as anything before :: is stripped).

To ensure names are unique, you can set Redis::Objects.prefix_style = :modern after you load the module. By default, it is set to Redis::Objects.prefix_style = :legacy.

Migrating old keys

If your Redis::Object subclasses are nested such as Dog::Behavior and Cat::Behavior, then you should upgrade them using the below proceedure. (only needed once)

Create a script like this:

class Dog::Behavior < ActiveRecord::Base
  include Redis::Objects
  # ... your relevant redis_object definitions here (counters/sets) ...
end

Dog::Behavior.migrate_redis_legacy_keys

You need to find a time when you can temporarily pause writes to your redis server so that you can run that script. It uses redis.scan internally so it should be able to handle a high number of keys. For large data sets, it could take a while.

After migrating all of your redis keys, update Redis::Objects.prefix_style = :modern to start using the new keys.

Note: Your existing data in Redis will not be accessible after running migrate_redis_legacy_keys until the prefix_style has been changed.

Overview

This is not an ORM (Object-Relational Mapping). People that are wrapping ORM’s around Redis are missing the point.

The killer feature of Redis is that it allows you to perform atomic operations on individual data structures, like counters, lists, and sets. The atomic part is HUGE. Using an ORM wrapper that retrieves a "record", updates values, then sends those values back, removes the atomicity, and thus the major advantage of Redis. Just use MySQL, k?

This gem provides a Rubyish interface to Redis, by mapping Redis data types to Ruby objects, via a thin layer over the redis gem. It offers several advantages over the lower-level redis-rb API:

Easy to integrate directly with existing ORMs - ActiveRecord, DataMapper, etc. Add counters to your model!
Complex data structures are automatically Marshaled (if you set :marshal => true)
Integers are returned as integers, rather than '17'
Higher-level types are provided, such as Locks, that wrap multiple calls

This gem originally arose out of a need for high-concurrency atomic operations; for a fun rant on the topic, see An Atomic Rant, or scroll down to Atomic Counters and Locks in this README.

There are two ways to use Redis::Objects, either as an include within a model class (to tightly integrate with ORMs or other classes), or standalone by using classes such as Redis::List and Redis::SortedSet.

Installation and Setup

Add it to your Gemfile as:

Redis::Objects needs a handle created by Redis.new or a ConnectionPool.

If you're using Rails, config/initializers/redis.rb is a good place for this. However, there are no dependencies on Rails. Redis::Objects can be used in any Ruby code; Sinatra, Resque, or Standalone - no problem.

The recommended approach is to use a ConnectionPool since this guarantees that most timeouts in the redis client do not pollute your existing connection.

require 'connection_pool'
Redis::Objects.redis = ConnectionPool.new(size: 5, timeout: 5) { Redis.new(:host => '127.0.0.1', :port => 6379) }

However, you need to make sure that both :timeout and :size are set appropriately in a multithreaded environment.

Alternatively, you can set the redis handle directly:

Redis::Objects.redis = Redis.new(...)

Redis::Objects will also default to Redis.current if Redis::Objects.redis is not set.

Redis.current = Redis.new(:host => '127.0.0.1', :port => 6379)

Finally, you can even set different handles for different classes:

class User
  include Redis::Objects
end
class Post
  include Redis::Objects
end

# you can also use a ConnectionPool here as well
User.redis = Redis.new(:host => '1.2.3.4')
Post.redis = Redis.new(:host => '5.6.7.8')

As of 0.7.0, redis-objects now autoloads the appropriate Redis::Whatever classes on demand. Previous strategies of individually requiring redis/list or redis/set are no longer required.

Option 1: Model Class Include

Including Redis::Objects in a model class makes it trivial to integrate Redis types with an existing ActiveRecord, DataMapper, Mongoid, or similar class. Redis::Objects will work with any class that provides an id method that returns a unique value. Redis::Objects automatically creates keys that are unique to each object, in the format:

For illustration purposes, consider this stub class:

class User
  include Redis::Objects
  counter :my_posts
  def id
    1
  end
end

user = User.new
user.id  # 1
user.my_posts.increment
user.my_posts.increment
user.my_posts.increment
puts user.my_posts.value # 3
user.my_posts.reset
puts user.my_posts.value # 0
user.my_posts.reset 5
puts user.my_posts.value # 5

Here's an example that integrates several data types with an ActiveRecord model:

class Team < ActiveRecord::Base
  include Redis::Objects

  redis_lock :trade_players, :expiration => 15  # sec
  value :at_bat
  counter :hits
  counter :runs
  counter :outs
  counter :inning, :start => 1
  list :on_base
  list :coaches, :marshal => true
  set  :outfielders
  hash_key :pitchers_faced  # "hash" is taken by Ruby

  # Customized keys
  counter :player_totals, :key => 'players/#{username}/total'
  list :all_player_stats, :key => 'players:all_stats', :global => true
  set :total_wins,        :key => 'players:#{id}:all_stats'
  value :my_rank,         :key => 'players:my_rank:#{username}'

  def id; @id; end
  def username; "user#{id}"; end
end

Familiar Ruby array operations Just Work™:

@team = Team.find_by_name('New York Yankees')
@team.on_base << 'player1'
@team.on_base << 'player2'
@team.on_base << 'player3'
@team.on_base    # ['player1', 'player2', 'player3']
@team.on_base.pop
@team.on_base.shift
@team.on_base.length  # 1
@team.on_base.delete('player2')
@team.on_base = ['player1', 'player2']  # ['player1', 'player2']

Sets work too:

@team.outfielders << 'outfielder1'
@team.outfielders << 'outfielder2'
@team.outfielders << 'outfielder1'   # dup ignored
@team.outfielders  # ['outfielder1', 'outfielder2']
@team.outfielders.each do |player|
  puts player
end
player = @team.outfielders.detect{|of| of == 'outfielder2'}
@team.outfielders = ['outfielder1', 'outfielder3']  # ['outfielder1', 'outfielder3']

Hashes work too:

@team.pitchers_faced['player1'] = 'pitcher2'
@team.pitchers_faced['player2'] = 'pitcher1'
@team.pitchers_faced = { 'player1' => 'pitcher2', 'player2' => 'pitcher1' }

And you can do unions and intersections between objects (kinda cool):

@team1.outfielders | @team2.outfielders   # outfielders on both teams
@team1.outfielders & @team2.outfielders   # in baseball, should be empty :-)

Counters can be atomically incremented/decremented (but not assigned):

@team.hits.increment  # or incr
@team.hits.decrement  # or decr
@team.hits.incr(3)    # add 3
@team.runs = 4        # exception

Defining a different method as the id field is easy

class User
  include Redis::Objects
  redis_id_field :uid
  counter :my_posts
end

user.uid                # 195137a1bdea4473
user.my_posts.increment # 1

You can also define globals redis attributes that are accessed through the class itself. No id needed/used for these.

class Team < ActiveRecord::Base
  include Redis::Objects

  sorted_set :rank, :global => true
end

Team.rank['Yankees']   = 12
Team.rank['Red Socks'] = 5
Team.rank['Mariners']  = 7
Team.rank.members(:with_scores => true) # => [["Red Socks", 5], ["Mariners", 7], ["Yankees", 12]]

Finally, for free, you get a redis method that points directly to a Redis connection:

Team.redis.get('somekey')
@team = Team.new
@team.redis.get('somekey')
@team.redis.smembers('someset')

You can use the redis handle to directly call any Redis API command.

Option 2: Standalone Usage

There is a Ruby class that maps to each Redis type, with methods for each Redis API command. Note that calling new does not imply it's actually a "new" value - it just creates a mapping between that Ruby object and the corresponding Redis data structure, which may already exist on the redis-server.

Counters

The counter_name is the key stored in Redis.

@counter = Redis::Counter.new('counter_name')
@counter.increment  # or incr
@counter.decrement  # or decr
@counter.increment(3)
puts @counter.value

This gem provides a clean way to do atomic blocks as well:

@counter.increment do |val|
  raise "Full" if val > MAX_VAL  # rewind counter
end

See the section on Atomic Counters and Locks for cool uses of atomic counter blocks.

Locks

A convenience class that wraps the pattern of using setnx to perform locking.

@lock = Redis::Lock.new('serialize_stuff', :expiration => 15, :timeout => 0.1)
@lock.lock do
  # do work
end

This can be especially useful if you're running batch jobs spread across multiple hosts.

Values

Simple values are easy as well:

@value = Redis::Value.new('value_name')
@value.value = 'a'
@value.delete

Complex data is no problem with :marshal => true:

@account = Account.create!(params[:account])
@newest  = Redis::Value.new('newest_account', :marshal => true)
@newest.value = @account.attributes
puts @newest.value['username']

Compress data to save memory usage on Redis with :compress => true:

@account = Account.create!(params[:account])
@marshaled_value = Redis::Value.new('marshaled', :marshal => true, :compress => true)
@marshaled_value.value = @account.attributes
@unmarshaled_value = Redis::Value.new('unmarshaled', :compress => true)
@unmarshaled_value = 'Really Long String'
puts @marshaled_value.value['username']
puts @unmarshaled_value.value

Lists

Lists work just like Ruby arrays:

@list = Redis::List.new('list_name')
@list << 'a'
@list << 'b'
@list.include? 'c'   # false
@list.values  # ['a','b']
@list << 'c'
@list.delete('c')
@list[0]
@list[0,1]
@list[0..1]
@list.shift
@list.pop
@list.clear
# etc

You can bound the size of the list to only hold N elements like so:

# Only holds 10 elements, throws out old ones when you reach :maxlength.
@list = Redis::List.new('list_name', :maxlength => 10)

Complex data types are serialized with :marshal => true:

@list = Redis::List.new('list_name', :marshal => true)
@list << {:name => "Nate", :city => "San Diego"}
@list << {:name => "Peter", :city => "Oceanside"}
@list.each do |el|
  puts "#{el[:name]} lives in #{el[:city]}"
end

Note: If you run into issues, with Marshal errors, refer to the fix in Issue #176.

Hashes

Hashes work like a Ruby Hash, with a few Redis-specific additions. (The class name is "HashKey" not just "Hash", due to conflicts with the Ruby core Hash class in other gems.)

@hash = Redis::HashKey.new('hash_name')
@hash['a'] = 1
@hash['b'] = 2
@hash.each do |k,v|
  puts "#{k} = #{v}"
end
@hash['c'] = 3
puts @hash.all  # {"a"=>"1","b"=>"2","c"=>"3"}
@hash.clear

Redis also adds incrementing and bulk operations:

@hash.incr('c', 6)  # 9
@hash.bulk_set('d' => 5, 'e' => 6)
@hash.bulk_get('d','e')  # "5", "6"

Remember that numbers become strings in Redis. Unlike with other Redis data types, redis-objects can't guess at your data type in this situation, since you may actually mean to store "1.5".

Sets

Sets work like the Ruby Set class. They are unordered, but guarantee uniqueness of members.

@set = Redis::Set.new('set_name')
@set << 'a'
@set << 'b'
@set << 'a'  # dup ignored
@set.member? 'c'      # false
@set.members          # ['a','b']
@set.members.reverse  # ['b','a']
@set.each do |member|
  puts member
end
@set.clear
# etc

You can perform Redis intersections/unions/diffs easily:

@set1 = Redis::Set.new('set1')
@set2 = Redis::Set.new('set2')
@set3 = Redis::Set.new('set3')
members = @set1 & @set2   # intersection
members = @set1 | @set2   # union
members = @set1 + @set2   # union
members = @set1 ^ @set2   # difference
members = @set1 - @set2   # difference
members = @set1.intersection(@set2, @set3)  # multiple
members = @set1.union(@set2, @set3)         # multiple
members = @set1.difference(@set2, @set3)    # multiple

Or store them in Redis:

@set1.interstore('intername', @set2, @set3)
members = @set1.redis.get('intername')
@set1.unionstore('unionname', @set2, @set3)
members = @set1.redis.get('unionname')
@set1.diffstore('diffname', @set2, @set3)
members = @set1.redis.get('diffname')

And use complex data types too, with :marshal => true:

@set1 = Redis::Set.new('set1', :marshal => true)
@set2 = Redis::Set.new('set2', :marshal => true)
@set1 << {:name => "Nate",  :city => "San Diego"}
@set1 << {:name => "Peter", :city => "Oceanside"}
@set2 << {:name => "Nate",  :city => "San Diego"}
@set2 << {:name => "Jeff",  :city => "Del Mar"}

@set1 & @set2  # Nate
@set1 - @set2  # Peter
@set1 | @set2  # all 3 people

Sorted Sets

Due to their unique properties, Sorted Sets work like a hybrid between a Hash and an Array. You assign like a Hash, but retrieve like an Array:

@sorted_set = Redis::SortedSet.new('number_of_posts')
@sorted_set['Nate']  = 15
@sorted_set['Peter'] = 75
@sorted_set['Jeff']  = 24

# Array access to get sorted order
@sorted_set[0..2]           # => ["Nate", "Jeff", "Peter"]
@sorted_set[0,2]            # => ["Nate", "Jeff"]

@sorted_set['Peter']        # => 75
@sorted_set['Jeff']         # => 24
@sorted_set.score('Jeff')   # same thing (24)

@sorted_set.rank('Peter')   # => 2
@sorted_set.rank('Jeff')    # => 1

@sorted_set.first           # => "Nate"
@sorted_set.last            # => "Peter"
@sorted_set.revrange(0,2)   # => ["Peter", "Jeff", "Nate"]

@sorted_set['Newbie'] = 1
@sorted_set.members         # => ["Newbie", "Nate", "Jeff", "Peter"]
@sorted_set.members.reverse # => ["Peter", "Jeff", "Nate", "Newbie"]

@sorted_set.rangebyscore(10, 100, :limit => 2)   # => ["Nate", "Jeff"]
@sorted_set.members(:with_scores => true)        # => [["Newbie", 1], ["Nate", 16], ["Jeff", 28], ["Peter", 76]]

# atomic increment
@sorted_set.increment('Nate')
@sorted_set.incr('Peter')   # shorthand
@sorted_set.incr('Jeff', 4)

The other Redis Sorted Set commands are supported as well; see Sorted Sets API.

Atomic Counters and Locks

You are probably not handling atomicity correctly in your app. For a fun rant on the topic, see An Atomic Rant.

Atomic counters are a good way to handle concurrency:

@team = Team.find(1)
if @team.drafted_players.increment <= @team.max_players
  # do stuff
  @team.team_players.create!(:player_id => 221)
  @team.active_players.increment
else
  # reset counter state
  @team.drafted_players.decrement
end

An atomic block gives you a cleaner way to do the above. Exceptions or returning nil will rewind the counter back to its previous state:

@team.drafted_players.increment do |val|
  raise Team::TeamFullError if val > @team.max_players  # rewind
  @team.team_players.create!(:player_id => 221)
  @team.active_players.increment
end

Here's a similar approach, using an if block (failure rewinds counter):

@team.drafted_players.increment do |val|
  if val <= @team.max_players
    @team.team_players.create!(:player_id => 221)
    @team.active_players.increment
  end
end

Class methods work too, using the familiar ActiveRecord counter syntax:

Team.increment_counter :drafted_players, team_id
Team.decrement_counter :drafted_players, team_id, 2
Team.increment_counter :total_online_players  # no ID on global counter

Class-level atomic blocks can also be used. This may save a DB fetch, if you have a record ID and don't need any other attributes from the DB table:

Team.increment_counter(:drafted_players, team_id) do |val|
  TeamPitcher.create!(:team_id => team_id, :pitcher_id => 181)
  Team.increment_counter(:active_players, team_id)
end

Locks

Locks work similarly. On completion or exception the lock is released:

class Team < ActiveRecord::Base
  redis_lock :reorder # declare a lock
end

@team.reorder_lock.lock do
  @team.reorder_all_players
end

Class-level lock (same concept)

Team.obtain_lock(:reorder, team_id) do
  Team.reorder_all_players(team_id)
end

Lock expiration. Sometimes you want to make sure your locks are cleaned up should the unthinkable happen (server failure). You can set lock expirations to handle this. Expired locks are released by the next process to attempt lock. Just make sure you expiration value is sufficiently large compared to your expected lock time.

class Team < ActiveRecord::Base
  redis_lock :reorder, :expiration => 15.minutes
end

Keep in mind that true locks serialize your entire application at that point. As such, atomic counters are strongly preferred.

Expiration

Use :expiration and :expireat options to set default expiration.

value :value_with_expiration, :expiration => 1.hour
value :value_with_expireat, :expireat => lambda { Time.now + 1.hour }

⚠️ In the above example, expiration is evaluated at class load time. In this example, it will be one hour after loading the class, not after one hour after setting a value. If you want to expire one hour after setting the value, please use :expireat with lambda.

Custom serialization

You can customize how values are serialized by setting serializer: CustomSerializer. The default is Marshal from the standard lib, but it can be anything that responds to dump and load. JSON and YAML are popular options.

If you need to pass extra arguments to dump or load, you can set marshal_dump_args: { foo: 'bar' } and marshal_load_args: { foo: 'bar' } respectively.

class CustomSerializer
  def self.dump(value)
    # custom code for serializing
  end

  def self.load(value)
    # custom code for deserializing
  end
end

@account = Account.create!(params[:account])
@newest  = Redis::Value.new('custom_serializer', marshal: true, serializer: CustomSerializer)
@newest.value = @account.attributes

Under the Hood

Redis keys are prefixed to namespace keys with the same names. By default the prefix is generated from the embedded class's name. But you can also set a custom prefix using redis_prefix=.

If needed the redis key for a specific attribute of a class can be obtained by: MyClass.redis_field_key(attr_name, primary_id) or for globals MyClass.redis_field_key(attr_name)

Important 2.0 changes

Renaming of lock Method

New key naming method

Migrating old keys

Overview

Installation and Setup

Option 1: Model Class Include

Option 2: Standalone Usage

Counters

Locks

Values

Lists

Hashes

Sets

Sorted Sets

Locks

Expiration

Custom serialization

Under the Hood

Author

Renaming of `lock` Method