Multi-Master - Removing a down node

[jameswestover]

When running KeyDB in a multi-master setup, if a master fails and is unable to reacquire the same address upon startup, the node moves into a master_link_status:down state. It would be nice to be able to remove the bad replica without having to restart the KeyDb process.

REPLICAOF NO ONE removes all nodes, in which case the healthy nodes would have to be added back in, causing a lapse in replication / availability.

It would be nice to have a command something like REPLICAOF REMOVE IP PORT which would remove a master from the replication group.

I haven't found an alternative that allows removing nodes in a zero down time way, while still retaining the other health replication members.

[JohnSully]

Yes I agree that's a major missing feature.

I've implemented it exactly as you described in change: 70b80aa

[jameswestover]

Amazing! Thank you!

[darinspivey]

Wow, I'm super glad I found this issue as I was getting ready to implement replicaof no one to remove dead replicas. Thank you for this fix, but the documentation desperately needs to be updated to reflect this feature.

I opened Snapchat/KeyDB-docs#49 for this, and THANK YOU for this fix which I can immediately use!