@redis/core-team The major change here is that CLUSTER FORGET will now attempt to gossip the deletion to all nodes in the cluster. You can still use the old style of calling CLUSTER FORGET on all nodes without errors.

LGTM on high level.
i understand that the implication on older versions (mixed cluster) is that older nodes will print a log warning, and will require an explicit FORGET.
so the feature can be relied on, only if all nodes supports this, but will not cause harm in mixed clusters.

I understand redis-cli still sends that command to all nodes (and we can probably never change that), so is this change just to resolve a race condition (mentioned in the related issue?).
let's copy the relevant justification text to the top comment here (for the purpose of release notes).

while on that topic, considering that it's gonna take a long time before we actually release it, i wanna propose a new guideline:
whenever we tag an issue to be mentioned in the release notes, we also add a line in the top comment of the PR with a short text to be copied to the release notes.
something like:

Release notes: 
<some short text that's enough to draw the reader attention if this might affect them, or cause them to skip otherwise>

WDYT?

@oranagra Since you are the one that is generating the release notes most of the time, I'm happy to follow/enforce that.

@zuiderkwast do you want to rebase?

tests sporadically failed on ARM + TLS
https://github.com/redis/redis-extra-ci/runs/7935419226?check_suite_focus=true

*** [err]: Set cluster hostnames and verify they are propagated in tests/unit/cluster/hostnames.tcl
cluster config did not reach a consistent state
*** [err]: Update hostnames and make sure they are all eventually propagated in tests/unit/cluster/hostnames.tcl
cluster config did not reach a consistent state
*** [err]: Remove hostnames and make sure they are all eventually propagated in tests/unit/cluster/hostnames.tcl
cluster config did not reach a consistent state
*** [err]: Verify cluster-preferred-endpoint-type behavior for redirects and info in tests/unit/cluster/hostnames.tcl
cluster config did not reach a consistent state

tests sporadically failed on ARM + TLS https://github.com/redis/redis-extra-ci/runs/7935419226?check_suite_focus=true

cluster config did not reach a consistent state

Any idea why and what we can do about it? Shall we increase the time to wait to reach a consistent cluster state?

Increasing the timing seems like one such option, we can see if it helps. An immediate change is we should probably print out the delta.

i see it's usually fast, so i suppose increase the time won't help much

[ok]: Set cluster hostnames and verify they are propagated (311 ms)
[ok]: Update hostnames and make sure they are all eventually propagated (309 ms)
[ok]: Remove hostnames and make sure they are all eventually propagated (312 ms)
[ok]: Verify cluster-preferred-endpoint-type behavior for redirects and info (307 ms)

It might be a failover causing some havoc? I have seen that happening on some of the other ARM tests. We should also probably clean up the logging a bit, it's hard to follow.