Remove the NONDETERMINISTIC_OUTPUT flag from most CLUSTER sub-commands.

[huangzhw]

TLDR: the CLUSTER command originally had the random flag,
so all the sub-commands initially got that new flag, but in fact many
of them don't need it.
The only effect of this change is on the output of COMMAND INFO.

Original text:

cluster keyslot is deterministic.
I doubt whether other cluster commands is NONDETERMINISTIC_OUTPUT.
cluster myid
cluster flushslots

What definition of NONDETERMINISTIC_OUTPUT really is?

[oranagra]

the definition is here: https://redis.io/docs/reference/command-tips/

i see that this flag was added to all these commands in #10104, i.e. in the initial PR that added this flag, and not some specific concern later realized, so chances are these are just an oversight. @guybe7 FYI.
considering some of these return a simple OK, it certainly doesn't make sense.
@huangzhw can you please go over them and fix the others?

[huangzhw]

Does different errors nondeterministic?
e.g. cluster flushslot may return "DB must be empty to perform CLUSTER FLUSHSLOTS.".
saveconfig may fail with "error saving the cluster node config: %s".
cluster forget may return different errors because of server role. If this is, role should also be.
cluster id I think it is deterministic.
So I think the definition is not enough.

[oranagra]

I don't think any of these count as non-deterministic for that matter.
any command in redis may result with different response depending on state (such as WRONGTYPE, MASTERDOWN, LOADING, etc).
the non-deterministic flag is for commands like SPOP, RANDOMKEY, TTL, and even INFO, which could result in different response every time they're called, even on the exact same data set and state.

[huangzhw]

Add more cluster commands.

[oranagra]

@madolson can you please make sure we're not missing anything.
@guybe7 do you recall how come they all got this flag?

[guybe7]

@oranagra before Redis 7 clusterCommand had the "random" flag, i think it just gave all subcommands NONDETERMINISTIC_OUTPUT, thinking to preserve the status quo (i don't know much about cluster commands so i preferred to leave it as-is rather than to do something wrong)

[madolson]

Yeah, this list seems right. I might also include countkeysinslot, which is arguably deterministic based on the dataset, and dbsize doesn't have the flag.