Inconsistency for FIRST_KEY_INDEX for scripts in cluster mode

[dtikhonov-iex]

Description

When using scripts in cluster mode, RedisCluster.extractFirstKey() is called with redisArgs the same as args. In other words, working off the example in documentation, when calling set('key', 1), redisArgs is set to ["SET", "key", "1"]; when calling add('key', 1), it's set to ["key", "1"].

Instead of setting FIRST_KEY_INDEX to the usual value of 1, for scripts, it should be set to 0 to get the correct key value.

This should be documented because it's not obvious. I happened to look at the packet capture and noticed this behavior. Because redirection works well, this inefficiency may go unnoticed by users.

[leibale]

@dtikhonov-iex TBH I found out about this a few days ago by mistake (I was changing the code around that for v5..), and I'm not sure how to proceed:

1. document that FIRST_KEY_INDEX is based on the reply from transformArguments, and not necessarily the whole command arguments.
2. "fix" it in v5

What do you think we should do?

@guyroyse @simonprickett what do you think?

[dtikhonov-iex]

The fix would be: if FIRST_KEY_INDEX is not specified, the library should do the right thing.

Otherwise, documentation should definitely be updated.

[johnhuffsmith]

FIRST_KEY_INDEX for EVAL (evalFirstKeyIndex) is also incorrect. In cluster mode EVAL with randomly pick nodes until it finds the correct slot or fails.