RegisterOnMemberUp() does not fire immediately if cluster already up

[ayoung]

If the cluster member is already up, calling cluster.RegisterOnMemberUp(callback) does not immediately call the callback.

According to this code here

[Danthar]

Hmm. @Horusiath does the JVM work like this as well ?
Im hesitant to mark this as an bug. Because the same issue would apply to all places where you register status callbacks on the cluster extension. Which is due to the async nature of the whole system of course.

I would say this is an problem that needs to be solved in user space. By having the user check the cluster readview for this scenario, should it apply.
Thoughts? @Aaronontheweb

[ayoung]

It seems like an inconsistency with how RegisterOnMemberRemoved() works. That method immediately calls the callback if the member is terminated.

[Aaronontheweb]

Actually, I wonder if this is true... let me run an experiment real quick.

[Aaronontheweb]

@ayoung so it turns out that this isn't an issue, and the test I added in #2446 verifies it.

TL;DR;, the reason for the different code in the RegisterOnMemberRemoved is because the actors responsible for executing RegisterOnMember[X] callbacks are terminated by that point in time. In the event of RegisterOnMemberUp, if a callback is added after the cluster has started the actor will check to see if the current member is already in that state (MemberUp), and if so will proceed to invoke the callback immediately rather than wait for another event.