membership questions

[mbj4668]

Suppose I have a cluster with 3 members, and I want to move this cluster to 3 other nodes. I can call add_member with all 3 new nodes, one at the time.
Is there a way to know that it is safe to delete the 3 old members?

[kjnilsson]

ra:remove_member/2|3 does this

…

On Mon, 14 Nov 2022 at 07:52, Martin Bjorklund ***@***.***> wrote: Suppose I have a cluster with 3 members, and I want to move this cluster to 3 other nodes. I can call add_member with all 3 new nodes, one at the time. Is there a way to know that it is safe to delete the 3 old members? — Reply to this email directly, view it on GitHub <#332>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAJAHFFTVWOTKZJKSU3Q5ATWIHVTDANCNFSM6AAAAAAR7QT7C4> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

-- *Karl Nilsson*

[mbj4668]

Do you mean that remove_member returns with an error if not all state is available somewhere in the rest of the cluster?

Is there are way to test this before calling remove_member?

[michaelklishin]

A Ra cluster member can find out the most recent version of machine state by executing a consistent query. This also makes sure that a majority of nodes is online. There should be nothing else to verify besides, perhaps, any application-specific aspects of your state that Ra has no knowledge of.

[michaelklishin]

Actually, there is one more thing to verify: that the node pending removal is not the quorum-critical one. There is a health check that does this in RabbitMQ that you can use example.

[mbj4668]

Ok, thanks, I'll take a closer look at that code. From just a glance it's not obvious how to use it ;)

Just to be clear, is it always safe to call remove_node? I.e., will it return an error if the node I try to remove is quorum-critical?