Decouple ruby applications using SNS fanout with SQS processing.
A Circuitry publisher application can broadcast events which can be fanned out to any number of SQS queues. This technique is a common approach to implementing an enterprise message bus. For example, applications which care about billing or new user onboarding can react when a user signs up, without the origin web application being concerned with those domains. In this way, new capabilities can be connected to an enterprise system without change proliferation.
What circuitry provides:
- Decoupling: apps can send and receive messages to each other without explicitly coding destinations into your app.
- Fan-out: multiple queues (i.e.: multiple apps) can receive the same message by publishing it a single time.
- Reliability: if your app goes down (intentionally or otherwise), messages will be waiting in the queue whenever it starts up again.
- Speed: because it's built on AWS, message delivery and receipt is fast.
- Duplication: although SQS messages can be delivered multiple times, circuitry safeguards to ensure they're only received by your app once.
- Retries: if a received message fails to be processed, it will be retried (unless otherwise configured).
- Customization: configure your publisher and subscriber to behave the way each app independently expects.
What circuitry does not provide:
- Ordering: messages may not arrive in the order they were sent.
- Scheduling: messages are processed as they're received.
A circuitry-example app is available to quickly try the gem on your own AWS account.
Add this line to your application's Gemfile:
gem 'circuitry', '~> 3.1'
And then execute:
$ bundle
Or install it yourself as:
$ gem install circuitry
Circuitry is configured via its configuration object or via circuitry.yml config.
Circuitry.subscriber_config do |c|
c.queue_name = "#{Rails.env}-appname"
c.dead_letter_queue_name = "#{Rails.env}-appname-failures"
c.max_receive_count = 8
c.access_key = 'YOUR_AWS_ACCESS_KEY'
c.secret_key = 'YOUR_AWS_SECRET_KEY'
c.region = 'us-east-1'
c.logger = Rails.logger
c.error_handler = proc do |error|
HoneyBadger.notify(error)
HoneyBadger.flush
end
c.lock_strategy = Circuitry::Locks::Redis.new(url: 'redis://localhost:6379')
c.async_strategy = :thread
c.on_async_exit = proc { Mongoid.disconnect_sessions }
end
Circuitry.publisher_config do |c|
c.access_key = 'YOUR_AWS_ACCESS_KEY'
c.secret_key = 'YOUR_AWS_SECRET_KEY'
c.region = 'us-east-1'
c.logger = Rails.logger
c.error_handler = proc do |error|
HoneyBadger.notify(error)
HoneyBadger.flush
end
c.async_strategy = :batch
end
Many of the advanced options, such as error_handler
or async_strategy
require this initializer-
style configuration. A simpler option is available via config/circuitry.yml
(or config/circuitry.yml.erb):
---
access_key: "YOUR_AWS_ACCESS_KEY"
secret_key: "YOUR_AWS_SECRET_KEY"
region: "us-east-1"
development:
publisher:
topic_names:
- brandonc-appname-user-create
- brandonc-appname-user-destroy
subscriber:
queue_name: "brandonc-appname"
dead_letter_queue_name: "brandonc-appname-failures"
topic_names:
- brandonc-otherapp-content-create
- brandonc-otherapp-content-destroy
production:
publisher:
topic_names:
- production-appname-user-create
- production-appname-user-destroy
subscriber:
queue_name: "production-appname"
dead_letter_queue_name: "production-appname-failures"
topic_names:
- production-otherapp-content-create
- production-otherapp-content-destroy
Available configuration options for both subscriber and publisher applications include:
access_key
: The AWS access key ID that has access to SNS publishing and/or SQS subscribing. (required unless using iam profile)secret_key
: The AWS secret access key that has access to SNS publishing and/or SQS subscribing. (required unless using iam profile)use_iam_profile
: Whether or not to use an iam profile for authenticating to AWS. This will only work when running your application inside of an AWS instance. Acceptstrue
orfalse
Please refer to the AWS Docs for more details about using iam profiles for authentication. (required unless using access and secret keys, default:false
)region
: The AWS region that your SNS and/or SQS account lives in. (optional, default: "us-east-1")logger
: The logger to use for informational output, warnings, and error messages. (optional, default:Logger.new(STDOUT)
)error_handler
: An object that responds tocall
with two arguments: the deserialized message contents and the topic name used when publishing to SNS. (optional, default:nil
)async_strategy
: One of:fork
,:thread
, or:batch
that determines how asynchronous publish requests are processed. (optional, default::fork
):fork
: Forks a detached child process that immediately sends the request.:thread
: Creates a new thread that immediately sends the request. Because threads are not guaranteed to complete when the process exits, completion can be ensured by callingCircuitry.flush
.:batch
: Stores the request in memory to be submitted later. Batched requests must be manually sent by callingCircuitry.flush
. Only valid as a publishing strategy
on_async_exit
: An object that responds tocall
. This is useful for managing shared resources such as database connections that require closing. (optional, default:nil
)topic_names
: An array of topic names that your application will publish and/or subscribe to. This configuration is only used during provisioning.middleware
: A chain of middleware that messages must go through when sent or received. Please refer to the Middleware section for more details regarding this option.aws_options_overrides
: A key/value hash of option overrides and additions passed through to theAWS::SQS::Client
(for subscriber config) orAWS::SNS::Client
(for publisher config)
Available configuration options for subscriber applications include:
queue_name
: The name of the SQS queue that your subscriber application will listen to. This queue will be created or configured during provisioning.dead_letter_queue_name
: The name of the SQS dead letter queue that will be used after all retries fail. This configuration value is only used during provisioning. (optional, default:<subscriber_queue_name>-failures
)lock_strategy
- The store used to ensure that no duplicate messages are processed. Please refer to the Lock Strategies section for more details regarding this option. (default:Circuitry::Locks::Memory.new
)max_receive_count
- The number of times a message will be received by the queue after unsuccessful attempts to process it before it is discarded or added to thedead_letter_queue_name
queue. This configuration value is only used during provisioning. (optional, default: 8)visibility_timeout
- A period of time during which Amazon SQS prevents other subscribers from receiving and processing that message (before it is deleted by circuitry after being processed successfully.) This configuration value is only used during provisioning. (optional, default: 1800)
You can automatically provision SQS queues, SNS topics, and the subscriptions between them using
two methods: the circuitry CLI or the rake circuitry:setup
task. The rake task will provision the
subscriber queue and publishing topics that are configured within your application.
require 'circuitry/tasks'
Circuitry.subscriber_config do |c|
c.queue_name = 'myapp-production-events'
c.topic_names = ['theirapp-production-stuff-created', 'theirapp-production-stuff-deleted']
end
When provisioning, a dead letter queue is also created using the name "<queue_name>-failures". You can customize the dead letter queue name in your configuration.
Run circuitry help provision
for help using CLI provisioning.
Publishing is done via the Circuitry.publish
method. It accepts a topic name
that represents the SNS topic along with any non-nil object, representing the
data to be serialized. Whatever object is called will have its to_json
method
called for serialization.
obj = { foo: 'foo', bar: 'bar' }
Circuitry.publish('any-topic-name', obj)
The publish
method also accepts options that impact instantiation of the
Publisher
object, which currently includes the following options.
:async
- Whether or not publishing should occur in the background. Accepts one of:fork
,:thread
,:batch
,true
, orfalse
. Passingtrue
uses theasync_strategy
value from the gem configuration. Please refer to the Asynchronous Support section for more details regarding this option. (default:false
):timeout
- The maximum amount of time in seconds that publishing a message will be attempted before giving up. If the timeout is exceeded, an exception will raised to be handled by your application orerror_handler
. (default: 15)
obj = { foo: 'foo', bar: 'bar' }
Circuitry.publish('my-topic-name', obj, async: true, timeout: 20)
Alternatively, if your options hash will remain unchanged, you can build a single
Publisher
object to use for all publishing.
options = { ... }
publisher = Circuitry::Publisher.new(options)
publisher.publish('my-topic-name', obj)
Subscribing is done via the Circuitry.subscribe
method. It accepts a block for processing each
message. This method indefinitely blocks, processing messages as they are enqueued.
Circuitry.subscribe do |message, topic_name|
puts "Received #{topic_name} message: #{message.inspect}"
end
The subscribe
method also accepts options that impact instantiation of the
Subscriber
object, which currently includes the following options.
:lock
- The strategy used to ensure that no duplicate messages are processed. Acceptstrue
,false
, or an instance of a class inheriting fromCircuitry::Locks::Base
. Passingtrue
uses thelock_strategy
value from the gem configuration. Passingfalse
uses the NOOP strategy. Please refer to the Lock Strategies section for more details regarding this option. (default:true
):async
- Whether or not subscribing should occur in the background. Accepts one of:fork
,:thread
,true
, orfalse
. Passingtrue
uses theasync_strategy
value from the gem configuration. Passing an asynchronous value will cause messages to be handled concurrently. Please refer to the Asynchronous Support section for more details regarding this option. (default:false
):timeout
- The maximum amount of time in seconds that processing a message will be attempted before giving up. If the timeout is exceeded, an exception will raised to be handled by your application orerror_handler
. (default: 15):wait_time
- The number of seconds to wait for messages while connected to SQS. Anything above 0 results in long-polling, while 0 results in short-polling. (default: 10):batch_size
- The number of messages to retrieve in a single SQS request. (default: 10)
options = {
lock: true,
async: true,
timeout: 20,
wait_time: 60,
batch_size: 20
}
Circuitry.subscribe(options) do |message, topic_name|
# ...
end
Alternatively, if your options hash will remain unchanged, you can build a single
Subscriber
object to use for all subscribing.
options = { ... }
subscriber = Circuitry::Subscriber.new(options)
subscriber.subscribe do |message, topic_name|
# ...
end
Publishing supports three asynchronous strategies (forking, threading, and batching) while subscribing supports two (forking and threading).
When forking a child process, that child is detached so that your application does not need to worry about waiting for the process to finish. Forked requests begin processing immediately and do not have any overhead in terms of waiting for them to complete.
There are two important notes regarding forking in general as it relates to asynchronous support:
-
Forking is not supported on all platforms (e.g.: Windows and NetBSD 4), requiring that your implementation use synchronous requests or an alternative asynchronous strategy in such circumstances.
-
Forking results in resources being copied from the parent process to the child process. In order to prevent database connection errors and the like, you should properly handle closing and reopening resources before and after forking, respectively. For example, if you are using Rails with Unicorn, you may need to add the following code to your
unicorn.rb
configuration:before_fork do |server, worker| if defined?(ActiveRecord::Base) ActiveRecord::Base.connection.disconnect! end end after_fork do |server, worker| if defined?(ActiveRecord::Base) ActiveRecord::Base.establish_connection( Rails.application.config.database_configuration[Rails.env] ) end end
Refer to your adapter's documentation to determine how resources are handled with regards to forking.
Threaded publish and subscribe requests begin processing immediately. Unlike
forking, it's up to you to ensure that all threads complete before your
application exits. This can be done by calling Circuitry.flush
.
Batched publish and subscribe requests are queued in memory and do not begin
processing until you explicit flush them. This can be done by calling
Circuitry.flush
.
The Amazon SQS FAQ includes the following important point:
Amazon SQS is engineered to provide “at least once” delivery of all messages in its queues. Although most of the time each message will be delivered to your application exactly once, you should design your system so that processing a message more than once does not create any errors or inconsistencies.
Given this, it's up to the user to ensure messages are not processed multiple times in the off chance that Amazon does not recognize that a message has been processed.
The circuitry gem handles this by caching SQS message IDs: first via a "soft lock" that denotes the message is about to be processed, then via a "hard lock" that denotes the message has finished processing.
The soft lock has a default TTL of 5 minutes (a seemingly sane amount of time
during which processing most queue messages should certainly be able to
complete), while the hard lock has a default TTL of 24 hours (based upon
a suggestion by an AWS employee).
The soft and hard TTL values can be changed by passing a :soft_ttl
or
:hard_ttl
value to the lock initializer, representing the number of seconds
that a lock should persist. For example:
Circuitry.subscriber_config.lock_strategy = Circuitry::Locks::Memory.new(
soft_ttl: 10 * 60, # 10 minutes
hard_ttl: 48 * 60 * 60 # 48 hours
)
If not specified in your circuitry configuration, the memory store will be used by default. This lock strategy is provided as the lowest barrier to entry given that it has no third-party dependencies. It should be avoided if running multiple subscriber processes or if expecting a high throughput that would result in a large amount of memory consumption.
Circuitry::Locks::Memory.new
Using the redis lock strategy requires that you add gem 'redis'
to your
Gemfile
, as it is not included bundled with the circuitry gem by default.
There are two ways to use the redis lock strategy. The first is to pass your
redis connection options to the lock in the same way that you would when building
a new Redis
object.
Circuitry::Locks::Redis.new(url: 'redis://localhost:6379')
The second way is to pass in a :client
option that specifies either the redis
client itself or a ConnectionPool
of redis clients. This is useful for more advanced usage such as sharing an
existing redis connection, connection pooling, utilizing
Redis::Namespace, or utilizing
hiredis.
client = Redis.new(url: 'redis://localhost:6379')
Circuitry::Locks::Redis.new(client: client)
client = ConnectionPool.new(size: 5) { Redis.new }
Circuitry::Locks::Redis.new(client: client)
Using the memcache lock strategy requires that you add gem 'dalli'
to your
Gemfile
, as it is not included bundled with the circuitry gem by default.
There are two ways to use the memcache lock strategy. The first is to pass your
dalli connection host and options to the lock in the same way that you would when
building a new Dalli::Client
object. The special host
option will be treated
as the memcache host, just as the first argument to Dalli::Client
.
Circuitry::Locks::Memcache.new(host: 'localhost:11211', namespace: '...')
The second way is to pass in a :client
option that specifies the dalli client
itself. This is useful for sharing an existing memcache connection.
client = Dalli::Client.new('localhost:11211', namespace: '...')
Circuitry::Locks::Memcache.new(client: client)
Using the noop lock strategy permits you to continue to treat SQS as a distributed queue in a true sense, meaning that you might receive duplicate messages. Please refer to the Amazon SQS documentation pertaining to the Properties of Distributed Queues.
It's also possible to roll your own lock strategy. Simply create a class that
includes (or module that extends) Circuitry::Locks::Base
and implements the
following methods:
lock
: Accepts thekey
andttl
as parameters. If the key is already locked, this method must return false. If the key is not already locked, it must lock the key forttl
seconds and return true. It is important that the check and update are atomic in order to ensure the same message isn't processed more than once.lock!
: Accepts thekey
andttl
as parameters. Must lock the key forttl
seconds regardless of whether or not the key was previously locked.unlock!
: Accepts thekey
as a parameter. Must unlock (delete) the key if it was previously locked.
For example, a database-backed solution might look something like the following:
class DatabaseLockStrategy
include Circuitry::Locks::Base
def initialize(options = {})
super(options)
self.connection = options.fetch(:connection)
end
protected
def lock(key, ttl)
connection.exec("INSERT INTO locks (key, expires_at) VALUES ('#{key}', '#{Time.now + ttl}')")
end
def lock!(key, ttl)
connection.exec("UPSERT INTO locks (key, expires_at) VALUES ('#{key}', '#{Time.now + ttl}')")
end
def unlock!(key)
connection.exec("DELETE FROM locks WHERE key = '#{key}'")
end
private
attr_reader :connection
end
To use, simply create an instance of the class with your necessary options, and
pass your lock instance to the configuration as the :lock_strategy
.
connection = PG.connect(...)
Circuitry.subcriber_config.lock_strategy = DatabaseLockStrategy.new(connection: connection)
Circuitry middleware can be used to perform additional processing around a message being sent by a publisher or received by a subscriber. Some examples of processing that belong here are monitoring or encryption specific to your application.
Middleware can be added to the publisher, the subscriber, or both. A middleware
class is defined by an (optional) #initialize
method that accepts any number of
arguments, as well as a #call
method that accepts the topic
string, message
string, and a block for continuing processing.
For example, a simple logging middleware might look something like the following:
class LoggerMiddleware
attr_reader :namespace, :logger
def initialize(namespace:, logger: Logger.new(STDOUT))
self.namespace = namespace
self.logger = logger
end
def call(topic, message)
logger.info("#{namespace} (start): #{topic} - #{message}")
yield
ensure
logger.info("#{namespace} (done): #{topic} - #{message}")
end
private
attr_writer :namespace, :logger
end
Adding the middleware to the stack happens through the Circuitry config.
Circuitry.subscriber_config do |config|
# single-line format
config.middleware.add LoggerMiddleware, namespace: 'subscriber_app', logger: Rails.logger
# block format
config.middleware do |chain|
chain.add LoggerMiddleware, namespace: 'subscriber_app', logger: Rails.logger
end
end
Circuitry.publisher_config do |config|
# single-line format
config.middleware.add LoggerMiddleware, namespace: 'publisher_app'
# block format
config.middleware do |chain|
chain.add LoggerMiddleware, namespace: 'publisher_app'
end
end
config.middleware
responds to a handful of methods that can be used for configuring
your middleware:
#add
: Appends a middleware class to the end of the chain. If the class already exists, it is replaced.middleware.add NewMiddleware, arg1, arg2, ...
#prepend
: Prepends a middleware class to the beginning of the chain. If the class already exists, it is replaced.middleware.prepend NewMiddleware, arg1, arg2, ...
#remove
: Removes a middleware class from anywhere in the chain.middleware.remove NewMiddleware
#insert_before
: Injects a middleware class before another middleware class in the chain. If the other class does not exist in the chain, this behaves the same as#prepend
.middleware.insert_before ExistingMiddleware, NewMiddleware, arg1, arg2...
#insert_after
: Injects a middleware class after another middleware class in the chain. If the other class does not exist in the chain, this behaves the same as#add
.middleware.insert_after ExistingMiddleware, NewMiddleware, arg1, arg2...
#clear
: Removes all middleware classes from the chain.middleware.clear
Circuitry provides a simple option for testing publishing and subscribing without actually hitting
Amazon services. Inside your test suite (e.g.: spec_helper.rb
), just make sure you include the
following line:
require 'circuitry/testing'
After checking out the repo, run bin/setup
to install dependencies. Then, run
bin/console
for an interactive prompt that will allow you to experiment.
To install this gem onto your local machine, run bundle exec rake install
. To
release a new version, update the version number in version.rb
, and then run
bundle exec rake release
to create a git tag for the version, push git commits
and tags, and push the .gem
file to rubygems.org.
- Fork it (https://github.com/kapost/circuitry/fork)
- Create your feature branch (
git checkout -b my-new-feature
) - Update the changelog
- Commit your changes (
git commit -am 'Add some feature'
) - Push to the branch (
git push origin my-new-feature
) - Create a new Pull Request