Message published by a new publisher does not reach an existing subscriber

[iche033]

Environment

• OS Version: Ubuntu 24.04
• Source or binary build? source build, gz-transport14 branch

Description

• Expected behavior: When a new publisher comes up, it fails to publish messages to an existing subscriber on the first try
• Actual behavior: Publish always fails on the first try for different reasons, see steps to reproduce below

Steps to reproduce

I have added an example test_publisher executable in the example directory in the iche033/pub_sub_no_msg branch to demonstrate the issue:

1. Checkout the iche033/pub_sub_no_msg branch and build the examples
2. In one terminal, run the subscriber executable
3. In another terminal, run test_publisher - you should see Enter any key to publish message printed is the console
4. Hit enter in the test_publisher terminal to publish a message.
   • Publisher fails due to no connections (issue 1)
5. Hit enter again in the test_publisher terminal. This time publish succeeds and a message in printed in the subscriber terminal.
6. Restart subscriber, e.g. Ctrl+C and run ./subscriber again
7. Go back to the test_publisher terminal and hit any key to publish a message
   • Publish call succeeds, however no messages are received by the subscriber (issue 2)

I have highlighted the issues in bold above.

• issue 1:
  • my hypothesis is that discovery takes time so when the publisher comes up, it has not had time to be notified of all existing subscribers.
    • This is less of a problem as we can keep retrying until HasConnections return true
• issue 2:
  • when the publisher comes up, the HasConnections call actually returned true so it attempts to publish a message. The Publish call also returned true so it thinks that the message is published successfully.
    • This is more of a problem as there is no indication that the message never reached the subscriber.

Note that if I add a std::this_thread::sleep_for(std::chrono::milliseconds(1000) right after node.Advertise in test_publisher.cc, everything works fine. Same thing if I just move the publisher out of the while loop (so that the publisher is alive for the whole duration), everything works as expected.