Class: RuoteAMQP::ParticipantProxy
- Inherits:
-
Object
- Object
- RuoteAMQP::ParticipantProxy
- Includes:
- Ruote::LocalParticipant
- Defined in:
- lib/ruote-amqp/participant.rb
Overview
AMQP Participants
The RuoteAMQP::ParticipantProxy allows you to send workitems (serialized as JSON) or messages to any AMQP queues right from the process definition. When combined with the RuoteAMQP::Receiver you can easily leverage an extremely powerful local/remote participant combinations.
For local/remote participants The local part of the RuoteAMQP::ParticipantProxy relies on the presence of a RuoteAMQP::Receiver. Workitems are sent to the remote participant and the local part does not normally reply to the engine. Instead the engine will continue when a reply is received on the ‘ruote_workitems’ queue (see RuoteAMQP::Receiver).
Of course, the standard :forget => true format can be used even with remote particpants and :forget can even be set as a default in the options.
A simple way to create a remote participant to act upon workitems is to use the daemon-kit ruote responder.
Simple AMQP messages are treated as ‘fire and forget’ and the flow will continue when the local participant has queued the message for sending. (As there is no meaningful way to receive a workitem in reply).
Configuration
AMQP configuration is handled by directly manipulating the values of the AMQP.settings
hash, as provided by the AMQP gem. No AMQP defaults are set by the participant.
Usage
Define the queue used by an AMQP participant :
engine.register_participant(
:delete_user, RuoteAMQP::ParticipantProxy, 'queue' => 'user_manager')
Sending a workitem to the remote participant defined above:
Ruote.process_definition do
sequence do
delete_user
end
end
Let the local participant reply to the engine without involving the receiver
Ruote.process_definition do
sequence do
delete_user :forget => true
end
end
Setting up the participant in a slightly more ‘raw’ way:
engine.register_participant(
:amqp, RuoteAMQP::ParticipantProxy )
Sending a workitem to a specific queue:
Ruote.process_definition do
sequence do
amqp :queue => 'test', 'command' => '/run/regression_test'
end
end
Setup a ‘fire and forget’ participant that always replies to the engine:
engine.register_participant(
:jfdi, RuoteAMQP::ParticipantProxy, 'forget' => true )
Sending a message example to a specific queue (both steps are equivalent):
Ruote.process_definition do
sequence do
amqp :queue => 'test', :message => 'foo'
amqp :queue => 'test', :message => 'foo', :forget => true
end
end
AMQP notes
The participant currently only makes use of direct exchanges. Possible future improvements might see use for topic and fanout exchanges as well.
The direct exchanges are always marked as durable by the participant, and messages are marked as persistent by default (see #RuoteAMQP)
Direct Known Subclasses
Instance Method Summary collapse
- #cancel(fei, flavour) ⇒ Object
-
#consume(workitem) ⇒ Object
Process the workitem at hand.
-
#do_not_thread ⇒ Object
The current AMQP (0.6.7) has 1 queue per thread.
-
#initialize(options) ⇒ ParticipantProxy
constructor
The following parameters are used in the process definition.
-
#stop ⇒ Object
(Stops the underlying queue subscription).
Constructor Details
#initialize(options) ⇒ ParticipantProxy
The following parameters are used in the process definition.
An options hash with the same keys to provide defaults is accepted at registration time (see above).
-
:queue => (string) The AMQP queue used by the remote participant. nil by default.
-
:forget => (bool) Whether the flow should block until the remote participant replies. false by default
121 122 123 124 125 126 127 128 129 130 131 132 133 |
# File 'lib/ruote-amqp/participant.rb', line 121 def initialize() RuoteAMQP.start! @options = { 'queue' => nil, 'forget' => false, }.merge(.inject({}) { |h, (k, v)| h[k.to_s] = v; h }) # # the inject is here to make sure that all options have String keys end |
Instance Method Details
#cancel(fei, flavour) ⇒ Object
174 175 176 177 178 179 |
# File 'lib/ruote-amqp/participant.rb', line 174 def cancel(fei, flavour) # # TODO : sending a cancel item is not a bad idea, especially if the # job done over the amqp fence lasts... # end |
#consume(workitem) ⇒ Object
Process the workitem at hand. By default the workitem will be published to the direct exchange specified in the queue
workitem parameter. You can specify a message
workitem parameter to have that sent instead of the workitem.
140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 |
# File 'lib/ruote-amqp/participant.rb', line 140 def consume(workitem) target_queue = determine_queue(workitem) raise 'no queue specified (outbound delivery)' unless target_queue q = MQ.queue(target_queue, :durable => true) forget = determine_forget(workitem) opts = { :persistent => RuoteAMQP., :content_type => 'application/json' } if = workitem.fields['message'] || workitem.params['message'] forget = true # sending a message implies 'forget' => true q.publish(, opts) else q.publish(encode_workitem(workitem), opts) end reply_to_engine(workitem) if forget end |
#do_not_thread ⇒ Object
The current AMQP (0.6.7) has 1 queue per thread. If you let the default “one thread per participant consume call” kick in, you’ll end up with 1 queue per consume call (and…)
So, by returning true here, we force the queue to be always the same.
Many thanks to github.com/weifeng365 for reporting this issue and suggesting the fix.
TODO : should we have something to close queues when the engine / worker
shuts down ?
or is it already covered in the #stop ?
194 195 196 197 |
# File 'lib/ruote-amqp/participant.rb', line 194 def do_not_thread true end |