Skip to content
This repository

HTTPS clone URL

Subversion checkout URL

You can clone with HTTPS or Subversion.

Download ZIP

Wraps memcached, redis, memcache-client, dalli and handles their weirdnesses, including forking.

This branch is 0 commits ahead and 0 commits behind mcmire

README.md

cache

Wraps memcached, redis, memcache-client, dalli and handles their weirdnesses, including forking.

Aims to let other libraries be cache-agnostic in return for a performance hit.

Real world usage

Used by lock_method and cache_method so that you can use them with memcached, redis, etc.

In production use at carbon.brighterplanet.com and data.brighterplanet.com.

Quick example

require 'memcached' # a really fast memcached client gem by Evan Weaver, one of the lead engineers at Twitter
require 'cache'     # this gem, which wraps the client to provide a standard interface

client = Memcached.new('127.0.0.1:11211', :binary_protocol => true)
@cache = Cache.wrap(client)

# don't worry, even though it's memcached gem, this won't raise Memcached::NotFound
@cache.get('hello')

# fetch is not provided by the memcached gem, the wrapper adds it
@cache.fetch('hello') { 'world' }

# don't worry, the wrapper will automatically clone the Memcached object after forking (or threading for that matter)
Kernel.fork { @cache.get('hello') }

If you can't use the memcached gem (because you're on heroku, for example) then just wrap a dalli or a redis client. You still get exactly the same interface.

Rationale

I wanted a common interface to a bunch of great Ruby cache clients so I can develop gems (lock_method, cache_method) that accept any of them.

  • I'm tired of rescuing from Memcached::NotFound
  • I'm tired of forgetting whether it's :expires_in or :ttl
  • I don't know why we ever started using read/write instead of get/set.
  • I don't like how you have to manually handle after_fork for Redis, Memcached, etc.
  • I don't know why Memcached::Rails doesn't act like a ActiveRecord::Cache::Store
  • Why are you asking me about :raw or whatever? Just marshal it

Speed

It's more than 50% slower than raw Memcached and about the same as raw Dalli

# raw dalli versus wrapped

set: cache:dalli:bin                   2.150000   0.840000   2.990000 (  3.752008) <- Cache.wrap(Dalli::Client.new)
set: dalli:bin                         2.120000   0.830000   2.950000 (  3.734024) <- Dalli::Client.new

get: cache:dalli:bin                   2.040000   0.910000   2.950000 (  3.646148)
get: dalli:bin                         2.040000   0.900000   2.940000 (  3.632840)

delete: cache:dalli:bin                1.830000   0.880000   2.710000 (  3.381917)
delete: dalli:bin                      1.790000   0.880000   2.670000 (  3.327514)

get-missing: cache:dalli:bin           1.780000   0.880000   2.660000 (  3.344041)
get-missing: dalli:bin                 1.760000   0.880000   2.640000 (  3.337539)

set-large: cache:dalli:bin             2.750000   0.880000   3.630000 (  4.474265)
set-large: dalli:bin                   2.720000   0.870000   3.590000 (  4.436163)

get-large: cache:dalli:bin             2.420000   0.990000   3.410000 (  4.135326)
get-large: dalli:bin                   2.410000   0.990000   3.400000 (  4.119832)

# raw memcached versus wrapped

set: cache:libm:bin                    0.860000   0.640000   1.500000 (  3.033145) <- Cache.wrap(Memcached.new(:binary_protocol => true))
set: libm:bin                          0.200000   0.480000   0.680000 (  1.907099) <- Memcached.new(:binary_protocol => true)

get: cache:libm:bin                    0.800000   0.680000   1.480000 (  2.700458)
get: libm:bin                          0.260000   0.660000   0.920000 (  1.974025)

delete: cache:libm:bin                 1.000000   0.600000   1.600000 (  2.968057)
delete: libm:bin                       0.600000   0.560000   1.160000 (  2.375070)

get-missing: cache:libm:bin            0.980000   0.800000   1.780000 (  2.850947)
get-missing: libm:bin                  0.640000   0.710000   1.350000 (  2.520733)

set-large: cache:libm:bin              1.220000   0.590000   1.810000 (  3.404739)
set-large: libm:bin                    0.230000   0.520000   0.750000 (  2.111738)

get-large: cache:libm:bin              3.780000   0.870000   4.650000 (  6.073208)
get-large: libm:bin                    0.340000   0.830000   1.170000 (  2.304408)

Thanks to https://github.com/evan/memcached/blob/master/test/profile/benchmark.rb

So: hopefully it makes it easier to get started with caching and hit the low-hanging fruit. Then you can move on to a raw client!

Features

Forking/threading

When you use a Cache object to wrap Memcached or Redis, you don't have to worry about forking or threading.

For example, you don't have to set up unicorn or PhusionPassenger's after_fork.

TTL

0 means don't expire.

The default ttl is 60 seconds.

Marshalling

Everything gets marshalled. No option to turn it into "raw" mode. If you need that kind of control, please submit a patch or just use one of the other gems directly.

Methods

It will translate these methods to whatever Redis, Memcached, etc. client you're using:

@cache.get 'hello'
@cache.set 'hello', 'world', 5.minutes
@cache.delete 'hello'
@cache.flush
@cache.exist? 'hello'
@cache.reset
@cache.fetch 'hello' { 'world' }
@cache.cas 'hello' { |current| 'world' }
@cache.increment 'high-fives'
@cache.decrement 'high-fives'
@cache.get_multi 'hello', 'privyet', 'hallo'

Also provided for Rails compatibility:

@cache.write 'hello', 'world', :expires_in => 5.minutes
@cache.read 'hello'
@cache.clear
@cache.compare_and_swap
@cache.read_multi 'hello', 'privyet', 'hallo'

Supported clients

Supported memcached clients:

  • memcached ~> 1.4 (native C extensions, super fast!)
  • dalli ~> 2.1 (pure ruby, recommended if you're on heroku)
  • memcache-client ~> 1.8 (not recommended. the one that comes with Rails.)

Supported Redis clients:

How you might use it

  Super-fast memcached Pure Ruby memcached (works on Heroku) Redis
Rails
config.cache_store = Cache.wrap(Memcached.new)
config.cache_store = Cache.wrap(Dalli::Client.new)
config.cache_store = Cache.wrap(Redis.new)
Your own library
# Accept any client, let Cache take care of it
def cache=(raw_client)
  @cache = Cache.wrap(raw_client)
end
# Accept any client, let Cache take care of it
def cache=(raw_client)
  @cache = Cache.wrap(raw_client)
end
# Accept any client, let Cache take care of it
def cache=(raw_client)
  @cache = Cache.wrap(raw_client)
end
CacheMethod (already uses Cache internally)
CacheMethod.config.storage = Memcached.new
CacheMethod.config.storage = Dalli::Client.new
CacheMethod.config.storage = Redis.new
LockMethod (already uses Cache internally)
LockMethod.config.storage = Memcached.new
LockMethod.config.storage = Dalli::Client.new
LockMethod.config.storage = Redis.new

Other examples

It defaults to an in-process memory store:

@cache = Cache.new
@cache.set 'hello'
@cache.get 'hello', 'world'

You can specify a more useful cache client:

require 'redis'     # the redis key-value store
require 'cache'     # this gem, which provides a standard interface
raw_client = Redis.new
@cache = Cache.wrap(raw_client)

or

require 'dalli'     # the dalli memcached client used by heroku
require 'cache'     # this gem, which provides a standard interface
raw_client = Dalli::Client.new
@cache = Cache.wrap(raw_client)

Or you could piggyback off the default rails cache:

@cache = Cache.wrap(Rails.cache)

Copyright

Copyright 2011 Seamus Abshere

Something went wrong with that request. Please try again.