The assumption is that the master isn't flaky. You can achieve this in multiple ways. One of them you mention. If the software is bad, you can't help it. As f

The assumption is that the master isn't flaky. You can achieve this in multiple ways. One of them you mention. If the software is bad, you can't help it. As far as network failures are concerned, I think those are handled. If your network fails, you basically won't be able to send a message or won't get confirmation of it being sent. Either ways, if the queuing system works in at least once mode, you are guaranteed message delivery. If there is an error, the library reports an error and the application must re-try.