Currently when trying to send CLUSTER FORGET node_id to all the nodes in a cluster there is the problem that some node will be able to re-add it into some other node via the gossip section of cluster pings.

When CLUSTER FORGET is received, even if not processed as there is no such node ID, the node should be set into a blacklist for a few seconds (a fixed time should be more than enough even with the max number of nodes, ~1000), so that it is not re-added before this time expires.

Comment From: madolson

Salvatore did this.

Comment From: zuiderkwast

For reference, it was done in 3e948970fe81d2a209c5f9731123ca5cebfe1458 and in the commits before it.