Node hosting a stream replica performs no peer discovery after reset

[maestroes]

Problem description

A node hosting a stream replica fails to rejoin its cluster after reset. There is no peer discovery attempt because the node is not considered virgin in case the stream directory is created before is_virgin_node() check.

Proposal

Probably the stream directory should be added to the ignored files list in is_virgin_node() check: https://github.com/rabbitmq/rabbitmq-server/blob/main/deps/rabbit/src/rabbit_mnesia.erl#L992

Environment

RabbitMQ 3.11.3
Erlang 25.1.2

Steps to reproduce

1. Create a three-node cluster with a consul-based peer discovery
2. Create a stream with default settings
3. Stop and reset any node
4. Start the node

[michaelklishin]

Streams are entirely orthogonal to peer discovery. Nodes perform peer discovery when they are blank, that is, their data store directory is blank.

Unless reset does not remove stream data — which I doubt — streams cannot possibly affect peer discovery.

[mkuratczyk]

Hi. I've tried, but was unable to. Earlier today I asked the team how they feel about just adding the folder to ignored as you suggested. Let's see if there are any objections. My main questions is whether stream replication should just resume as you described. Perhaps that's the real issue - it's a "fresh" node that is already a stream member. But streams put replica on all nodes so maybe that's the correct, if unexpected, behaviour.

[maestroes]

Ok, I see, thanks for sharing the details

[mkuratczyk]

Your suggestion has been implemented and merged. It'll be available in the next 3.10 and 3.11 patch releases.
#7222

[maestroes]

Thanks, I've just tried 3.11.9. Peer discovery works fine.

[lukebakken]

@maestroes we appreciate you following up!

[mkuratczyk]

Implemented in #7222