Skip to content

Event receiver for the Routemaster bus

License

Notifications You must be signed in to change notification settings

deliveroo/routemaster-drain

Repository files navigation

routemaster-drain Version Build Code Climate codecov Docs

A Rack-based event receiver for the Routemaster event bus.

routemaster-drain is a collection of Rack middleware to receive and parse Routemaster events, filter them, and preemptively cache the corresponding resources.

It provides prebuilt middleware stacks (Basic, Mapping, and Caching) for typical use cases, illustrated below, or you can easily roll your own by combining middleware.

Installation

Add this line to your application's Gemfile:

gem 'routemaster-drain'

Configuration

This gem is configured through the environment, making 12factor compliance easier.

Required:

  • ROUTEMASTER_DRAIN_TOKENS: a comma-separated list of valid authentication tokens, used by Routemaster to send you events.

Optional:

  • ROUTEMASTER_DRAIN_REDIS: the URL of the Redis instance used for filtering and dirty mapping. Required if you use either feature, ignored otherwise. A namespace can be specified. Example: redis://user:s3cr3t@myhost:1234/12/ns.
  • ROUTEMASTER_CACHE_REDIS: the URL of the Redis instance used for caching. Required if you use the feature, ignored otherwise. Formatted like ROUTEMASTER_DRAIN_REDIS.
  • ROUTEMASTER_CACHE_EXPIRY: if using the cache, for how long to cache entries, in seconds. Default 1 year (31,536,000).
  • ROUTEMASTER_CACHE_AUTH: if using the cache, specifies what username/password pairs to use to fetch resources. The format is a comma-separated list of colon-separate lists of regexp, username, password values. Example: server1:user:p4ss,server2:user:p4ass.
  • ROUTEMASTER_QUEUE_NAME: if using the cache, on which Resque queue the cache population jobs should be enqueued.
  • ROUTEMASTER_CACHE_TIMEOUT: if using the cache, how long before Faraday will timeout fetching the resource. Defaults to 1 second.
  • ROUTEMASTER_CACHE_VERIFY_SSL: if using the cache, whether to verify SSL when fetching the resource. Defaults to false.

Alternatively, it's possible to configure the gem to use pre-initialized Redis clients for the Cache Redis, the Drain Redis, or both.

The clients are expected to be instances of Redis::Distributed. When configured, the pre-built clients take precendence and the gem won't try to estabilish its own private connection to the Redis instances supplied in the ENV variables. This is helpful if you want to re-use connections in order to reduce the number of connected clients to your Redis servers.

Routemaster::RedisBroker.instance.inject(
  drain_redis: a_redis_object,
  cache_redis: another_redis_object
)

Upgrading

If upgrading from any version between 2.4.0 and 3.0.1, and are using caching, your cache may be corrupted by entries that lack a TTL (which will eventually cause your Redis storage to blow up).

We provide a tool to fix your data. With your environment loaded and configured (e.g. from a Rails console), run:

Routemaster::Tasks::FixCacheTTL.new.call

This will scan your cache and add TTLs where missing.

Illustrated use cases

Simply receive events from Routemaster

Provide a listener for events:

class Listener
  def on_events_received(batch)
    batch.each do |event|
      puts event.url
    end
  end
end

Each event is a Hashie::Mash and responds to type (one of create, update, delete, or noop), url (the resource), and t (the event timestamp, in milliseconds since the Epoch).

Create the app that will process events:

require 'routemaster/drain/basic'
$app = Routemaster::Drain::Basic.new

Bind the app to your listener:

$app.subscribe(Listener.new, prefix: true)

And finally, mount your app to your subscription path:

# typically in config.ru
map '/events' do
  run $app
end

You can use the .subscribe method multiple times to have your event batches go through multiple listeners. But bear in the mind any performance cost of multiple places of processing the event batches. For example, instead of having multiple listener classes that iterate over the events in a batch, you can have a single listener class that iterates over the batch only once and reacts to the batch's events accordingly.

This relies on the excellent event bus from the wisper gem.

Receive change notifications without duplicates

When reacting to changes of some resource, it's common to want to avoid receiving further change notifications until you've actually processed that resource.

Possibly you'll want to process changes in batches at regular time intervals.

For this purpose, use Routemaster::Drain::Mapping:

require 'routemaster/drain/mapping'
$app = Routemaster::Drain::Mapping.new

And mount it as usual:

# in config.ru
map('/events') { run $app }

Instead of processing events, you'll check for changes in the dirty map:

require 'routemaster/dirty/map'
$map = Routemaster::Dirty::Map.new

every_5_minutes do
  $map.sweep do |url|
    # do something about this changed resource
    true
  end
end

Until you've called #sweep and your block has returned true, you won't be bugged again — the dirty map acts as a buffer of changes (see below for internals).

Notes:

  • You can limit the number of resources to be swept ($map.sweep(123) { ... }).
  • You can count the number of resources to be swept with $map.count.
  • You're not told what is to be swept; entities won't be swept in the order of events received (much like Routemaster does not guarantee ordering).
  • If your sweeper fails, the dirty map will not be cleaned, so you can have leftovers. It's good practice to regularly run $map.sweep { ... } and perform cleanup regularly.
  • The map won't tell you if the resources has been changed, created, or deleted. You'll have to figure it out with an API call.
  • You can still attach a listener to the app to get all events.

Cache data for all notified resources

Another common use case is that you'll actually need the representation of the resources Routemaster tells you about.

The Caching prebuilt app can do that for you, using Resque to populate the cache as events are received.

For this purpose, use Routemaster::Drain::Caching:

require 'routemaster/drain/caching'
$app = Routemaster::Drain::Caching.new

And mount it as usual:

# in config.ru
map('/events') { run $app }

You can still attach a listener if you want the incoming events. Typically, what you'll want is the cache:

require 'routemaster/cache'
$cache = Routemaster::Cache.new

response = @cache.fget('https://example.com/widgets/123')
puts response.body.id

In this example, if your app was notified by Routemaster about Widget #123, the cache will be very likely to be hit; and it will be invalidated automatically whenever the drain gets notified about a change on that widget.

Note that Cache#fget is a future, so you can efficiently query many resources and have any HTTP GET requests (and cache queries) happen in parallel.

Resources are fetched through jobs. By default it uses Resque but can be changed to use Sidekiq.

ROUTEMASTER_QUEUE_ADAPTER=sidekiq

Finally do not forget the corresponding backend:

# config/initializers/...

require 'routemaster/jobs/backends/sidekiq'
# or
require 'routemaster/jobs/backends/resque'

See rubydoc for more details on Cache.

If you need to provide configure the APIClient used by Routemaster::Cache, you can configure it using client_options:

$cache = Routemaster::Cache.new(client_options: {source_peer: "<your user agent>"})

You can specify your user agent with the ROUTEMASTER_API_CLIENT_USER_AGENT environment variable as well.

Expire Cache data for all notified resources

You may wish to maintain a coherent cache, but don't need the cache to be warmed before you process incoming events. In that case, use the cache as detailed above but swap the Caching drain out for CacheBusting

require 'routemaster/drain/caching_busting'
$app = Routemaster::Drain::CachingBusting.new

HTTP Client

The Drain is using a Faraday http client for communication between services. The client comes with a convenient caching mechanism as a default and supports custom response materialization. The Drain itself has the concept of "HATEOAS"(see below) response that provides a common way of addressing resources.

** In order for the client to discover the resources that you are interested in, you need to call the #discover(service_url) method first

Example:

require 'routemaster/fetcher'
require 'routemaster/responses/hateoas_response'

client = Routemaster::APIClient.new(
  response_class: Routemaster::Responses::HateoasResponse,
  source_peer: "<your user agent>"
)

response = client.discover('https://identity.deliveroo.com.dev')
session_create_response = response.sessions.create(email: '[email protected]', password: 'sup3rs3cr3t')
session_create_response.user.show(1)

The index method returns an Enumerable response to fetch all items in a paginated collection with the options of passing filters.

users = response.users
user_index_response = users.index(filters: {first_name: 'Jeff'})
total_users = user_index_response.total_users

puts "printing names of all #{total_users} users"
user_index_response.each do |user|
  puts user.full_name
end

HATEOAS materialisation

The client comes with optional HATEOAS response capabilities. They are optional, because drain itself doesn't need to use the HATEOAS response capabilities. Whenever the client is used outside of the drain it is strongly advised to be used with the HATEOAS response capabilities. The HATEOAS response will materialize methods based on the keys found under the _links key on the payload. The semantics are the following:

# Given the following payload
{
  "_links" : {
    "users" : { "href" : "https://identity.deliveroo.com.dev/users" },
    "user"  : { "href" : "https://identity.deliveroo.com.dev/users/{id}", "templated" : true }
  }
}

client = Routemaster::APIClient.new(response_class: Routemaster::Responses::HateoasResponse)
response = client.discover('https://identity.deliveroo.com.dev')

response.users.create(username: 'roo')
#=> HateoasResponse
response.users.index
#=> HateoasResponse
response.user.show(1)
#=> HateoasResponse

Circuit Breaker

The client ships with a circuit breaker for get requests, this can be enabled by setting ROUTEMASTER_ENABLE_API_CLIENT_CIRCUIT. The Circuit Breaker is powered by Circuitbox.

The following parameters can be used:

  • ROUTEMASTER_CIRCUIT_BREAKER_SLEEP_WINDOW - The time in seconds the circuit will remain open once it's triggered
  • ROUTEMASTER_CIRCUIT_BREAKER_TIME_WINDOW - The time in seconds over which it calculates the error rate
  • ROUTEMASTER_CIRCUIT_BREAKER_VOLUME_THRESHOLD - The minimum amount of requests that can trigger the circuit to open
  • ROUTEMASTER_CIRCUIT_BREAKER_ERROR_THRESHOLD - The % of errors required for the circuit to open in the sleep window

Each of these settings can be controlled on a per domain basis. To set ROUTEMASTER_CIRCUIT_BREAKER_SLEEP_WINDOW for example.com set example.com.ROUTEMASTER_CIRCUIT_BREAKER_SLEEP_WINDOW

Internals

The more elaborate drains are built with two components which can also be used independently, Dirty::Map, Dirty::Filter and Middleware::Siphon.

Dirty map

A dirty map collects entities that have been created, updated, or deleted (or rather, their URLs). It can be used to delay your service's reaction to events, for instance combined with Resque.

A dirty map map gets marked when an event about en entity gets processed that indicates a state change, and swept to process those changes.

Practically, instances of Routemaster::Dirty::Map will emit a :dirty_entity event when a URL is marked as dirty, and can be swept when an entity is "cleaned". If a URL is marked multiple times before being swept (e.g. for very volatile entities), the event will only by broadcast once.

To sweep the map, you can for instance listen to this event and call #sweep_one.

If you're not in a hurry and would rather run through batches you can call #sweep which will yield URLs until it runs out of dirty resources.

Filter

Routemaster::Dirty::Filter is a simple event filter that performs reordering. It ignores events older than the latest known information on an entity.

It stores transient state in Redis and will emit :entity_changed events whenever an entity has changed. This event can usefully be fed into a dirty map, as in Receiver::Filter for instance.

Siphon

Middleware::Siphon extracts payloads from the middleware chain, allowing them to be processed separately. This is useful for event topics where the update frequency is not well suited to frequent caching. For example, a location update event which you'd expect to receive every few seconds.

Contributing

  1. Fork it ( http://github.com/deliveroo/routemaster-drain/fork )
  2. Create your feature branch (git checkout -b my-new-feature)
  3. Commit your changes (git commit -am 'Add some feature')
  4. Push to the branch (git push origin my-new-feature)
  5. Create new Pull Request

Do not bump version numbers on branches (a maintainer will do this when cutting a release); but please do describe your changes in the CHANGELOG (at the top, without a version number).

If you have changed dependencies, you need to run appraisal update to make sure the various version specific gemfiles are updated.