The rosmultimaster package is designed to allow you to connect to multiple ros cores from within a single process. This is achieved via the rosmultimaster.Adaptor class.


ROS does not have a built-in system to allow a single node to connect to multiple ROS cores. Since entire processes are declared as ros nodes, there is therefore no way to access multiple ros cores from a single process.

API Description

rosmultimaster.Adaptor objects are designed to replicate the rospy API as closely as possible in order to ensure ease of use for programmers who are familiar with using rospy to instantiate ROS nodes and use them to publish, offer services, subscribe, and get service proxies. This is best illustrated by a simple example:

   1 import rosmultimaster
   2 from std_msgs.msg import String
   3 from std_srvs.srv import Empty, EmptyResponse
   4 import time
   6 ip = ''
   7 port = 8675
   9 def sub_test_callback(string_request):
  10   print
  11 def handle_srv_test(empty_request):
  12   print "I received an empty request!"
  13   return EmptyResponse()
  15 adaptor = rosmultimaster.Adaptor( host = ip, port = port, name = 'arbitrary_name', anonymous = True)
  16 pub = adaptor.Publisher('pub_test', String)
  17 adaptor.Subscriber('sub_test', String, sub_test_callback)
  18 adaptor.Service('srv_test', Empty, handle_srv_test)
  19 adaptor.ServiceProxy('proxy_test', Empty)
  20 adaptor.start()
  22 for x in range(20):
  23   pub.Publish("Hello %s! I'm talking to you on port %s!" %( ip, repr(port) ) )
  24   time.sleep(1)

As you can see, this works almost exactly the same way as rospy except for the initialization of the Adaptor and the need for the adaptor.start() call. For simplicity, this example only illustrates connecting to a single core with a single adaptor, but hopefully you can easily see how you could use the same system to connect to as many cores as you like. Just initialize the Adaptors with the correct hosts and ports, create the Subscribers, Services, Publishers, and Service proxies for those adaptors, and then call start() on the adaptors. They operate completely independently of each other. You can even connect two to the same core, as long as they have different "names" or are anonymous (as declared in the initialization).

Below is a more detailed breakdown of the differences between rospy and a rosmultimaster.Adaptor object.

Differences for rospy API

Major Differences

There are two major differences between the rospy system and the rosmultimaster system:

  1. In rosmultimaster you must initialize an Adaptor object with the host and port of the roscore to which you wish to connect. The "name" and "anonymous" parameters affect the name of the rosnode that will appear in the specified roscore.
  2. You must call adaptor.start() before the adaptor will actually be able to publish or subscribe to topics. All declarations of Publishers, Subscribers, etc. must occur before adaptor.start() is called.

The issues below are more minor and should not affect most users of the rosmultimaster system.

Some missing features

Many of the more obscure features of rospy have not yet been implemented in rosmultimaster. This includes things that modify the behavior of publishers, subscribers, etc (e.g. queue_size, callback_args in rospy.Subscriber() ) and whole features like the ros Parameter Server. Most of these should be very easy (but somewhat time consuming) to implement without any major changes to the rosmultimaster backend.

No control over order of initialization

All of your Subscribers, Publishers, Services, and ServiceProxies are initialized in that order as soon as you call adaptor.start(), regardless of the order in which you created them. If this order of initialization really matters for your application, you can get around this problem by using several adaptors connected to the same ip and port and then starting them in the desired order.

No duplicate topics

You can't subscribe to the same topic twice, have two publishers to the same topic, offer two services with the same names, or get two service proxies for the same service using the same adaptor. This may be true of rospy too, but it is definitely true of Adaptor.

No similar Services

You can't offer two services with the same service_class using the same adaptor. This is because since rospy.Service does not allow you to pass arguments into the service callback in the same way that rospy.Subscriber and rospy.Publisher do, which causes problems in the implementation of rosmultimaster.

No wait_for_service

Issue:When using ServiceProxy's in rosmultimaster, it is unnecessary (and currently impossible) to make any kind of "wait_for_service" call. Simply calling adaptor.ServiceProxy('topic', ServiceClass) will automatically cause a call to rospy.wait_for_service('topic') to be made internally when adaptor.start() is called. This is not ideal, since the goal of Adaptor is to mimic the rospy API as much as possible.

Rationale: Why this break with the rospy API is necessary:

  • I assume that rospy.wait_for_service blocks until the service becomes available. Since no ROS stuff is initialized until adaptor.start() is called, adaptor.wait_for_service will have no way of knowing when the service becomes available, so it therefore cannot block in the same way as rospy.wait_for_service. The best we can do is set a flag that tells adaptor to make the blocking call to rospy.wait_for_service in adaptor.start.
  • Finally, I assume that if one attempts to create a rospy.ServiceProxy without first calling rospy.wait_for_service, a ServiceException is raised. Normally this can be handled gracefully by wrapping the rospy.ServiceProxy call in a try/except block. However, adaptor.start() actually spawns separate processes, and it is in these processes that rospy.ServiceProxy is actually called. Therefore, if a ServiceException is raised, it will be raised in a separate process from the process in which adaptor.start() is called and will not be able to be caught by wrapping adaptor.start() in a try/except block.

In short, it is not possible to mimic the API for rospy perfectly, and adding an analogue to wait_for_service may be more confusing than making the assumption that you will generally want to wait for services to become available.


In order to connect to a roscore without actually declaring the main process as a node, Adaptor must start a child process, which is connected to the main process through pipes. This child process becomes a ROS node and sets up all of the publishers, subscribers, services, and service proxies that are declared before adaptor.start() is called. After this set-up is done, data flows as follows:

main process <--> pipes <--> child process/ROS node <--> ros protocols (pub/sub/serv) <--> ROS system

The main process is sometimes called the "Adaptor" process and the child process is sometimes called the _AdaptorNode process, since each process hold a reference to an object of that type.

Wiki: rosmultimaster (last edited 2010-10-14 21:32:11 by KenConley)