Get number of active publishers

[saikishor]

Description

I see that currently there is get_publisher_count() method for the subscribers to query the number of publishers, but I want my subscriber to reject operating when receiving data from 2 active publishers. There could be a case where 2 different nodes create the publishers, but only one active publisher publishes a message at a time and this can be allowed

Motivation

With the current get_publisher_count() there is no clear way to control the subscription callback data. For the robot control, you might be interested in having a reference from only one node at a time, instead of multiple nodes, and this should be somehow avoided. There can be multiple publishers, but only one can be active at all times. There is currently no way to handle this kind of behaviour. Please help out on how this can be implemented.

Design / Implementation Considerations

No response

Additional Information

No response

[fujitatomoya]

I may be missing some requirement that you could have.

but how about using content fileting on the subscription side? for example, having the publisher node name or identification in the message, and subscriber can call the content filtering API as it needs to choose the active publisher. means that, you have total number of publishers on get_publisher_count, but you can actually choose the publisher endpoint that you want to receive the data on each subscription.

[saikishor]

but how about using content filtering on the subscription side? for example, having the publisher node name or identification in the message, and subscriber can call the content filtering API as it needs to choose the active publisher. means that, you have total number of publishers on get_publisher_count, but you can actually choose the publisher endpoint that you want to receive the data on each subscription.

Thanks for talking about the content filtering API, I'm not aware of that, but it sounds useful. This means I'll have to add a field UID or node_name to the message and use that information to filter. However, how will the subscriber node know when there is a transition of references from one topic to another?

My problem is that I have a robotic arm to control, and I have a control architecture to control it via topic; however, the problem is that when 2 nodes try to control it, the robot cannot track references from both ends and it doesn't make sense too, so when there are more active publishers, I prefer that it prints warning and stops execution rather than continuing it. We can filter it somehow as you said, but the user wouldn't know implicitly which publisher is being used for the references.
Do you get my point?

[fujitatomoya]

@saikishor ah now i see your point.

how about taking advantage of https://docs.ros.org/en/rolling/Concepts/Intermediate/About-Quality-of-Service-Settings.html#matched-events then? with that, you application should be able to be notified by the event that tells you about the difference of the connectivity map for the topic. i guess this might work for your use case?

at the end, we do not have any identification for the endpoints such as publisher and subscription at least now, IMO pub/sub architecture this is intentionally decoupled for distributed system. but if this is the hard requirement here, i guess we need to start discussion on the new feature proposal to have the global identification for the endpoints?

[saikishor]

how about taking advantage of https://docs.ros.org/en/rolling/Concepts/Intermediate/About-Quality-of-Service-Settings.html#matched-events then? with that, you application should be able to be notified by the event that tells you about the difference of the connectivity map for the topic. i guess this might work for your use case?

Developers can subscribe to this event with a publisher or a subscription.

    publisher: this event happens when it finds a subscription which matches the topic and has compatible QoS or a connected subscription is disconnected

    subscription: this event happens when it finds a publisher which matches the topic and has compatible QoS or a connected publisher is disconnected

I think this only give you the event on the connection time but not on the publishing time, so there might be a case, where I have a node that created the publisher and not publishing the information until my action goal is received and at the same tine there is a node that is currently publishing the topic. So, In this case, if I understand correctly, The event happens only at the beginning when the compatibility exists. This won't the current use case, when the other one starts to publish at some random point.

at the end, we do not have any identification for the endpoints such as publisher and subscription at least now, IMO pub/sub architecture this is intentionally decoupled for distributed system. but if this is the hard requirement here, i guess we need to start discussion on the new feature proposal to have the global identification for the endpoints?

I understand that for the communication between nodes, in most cases, it might not be a necessity, but when you have a single resource that cannot process requests from everywhere just because contextually it doesn't make sense. It would be nice to have a way to have some identification of the endpoints, so the subscriber or the publisher can decide what to do in that particular case. The problem is that I cannot use an action goal, as I want to constantly stream trajectory references for the robot to perform a motion. This is very interesting for the new AI deployments, as you don't want trajectories from 2 different policies to be executed at the same time. We could discuss on the global identification of the endpoints

[saikishor]

How can we adding this global identification to the topics or where can we discuss about this?. I'm interested to know others point of view as well