CLUSTER RESET forgets other nodes, but they don't forget you

[zuiderkwast]

The CLUSTER RESET documentation says: "All the other nodes in the cluster are forgotten".

In reality, the node forgets the other nodes but the other nodes does not forget it. They will find it again soon.

It takes about 15 secs to re-join the cluster. So, we gonna have 15 seconds of errors which is pretty long for errors in such scenarios:

27859:M 19 Feb 2025 17:11:14.472 * Connection with replica 127.0.0.1:17383 lost.
27859:M 19 Feb 2025 17:11:29.523 * Sending MEET packet to node 096e0230bddddb11d32e36497adcc4235bffa7d1 () because there is no inbound link for it
27859:M 19 Feb 2025 17:11:29.593 * Successfully completed handshake with 096e0230bddddb11d32e36497adcc4235bffa7d1 ()

To make the rest of the cluster forget a node, we actually need to send CLUSTER FORGET to the other nodes in the cluster.

Idea: The CLUSTER RESET command could broadcast a forget message to the cluster.

Originally discussed by @skolosov-snap and others in #1674 (comment)

[zuiderkwast]

@enjoy-binbin @PingXie @madolson WDYT?

[PingXie]

Yep this is broken.

Idea: The CLUSTER RESET command could broadcast a forget message to the cluster.

That is what I am thinking too.

[skolosov-snap]

+1

[madolson]

Yeah, broadcasting the forget seems reasonable to me.

[enjoy-binbin]

This make sense to me, i had this idea a long time ago. So when a node receive a CLUSTER RESET, it broadcast a new FORGET message to tell the other nodes to forget about it?