Feature request: Cluster mode: A way to group keys by cluster node

[cbatson]

Given a set of arbitrary keys, presently, there does not appear to be a straightforward way to group keys by cluster node, for use with mget for instance.

I humbly request a documented public method which, when given a key as input, returns the node ID associated with that key (taking into account any presence of a hash tag, of course).

This will permit the caller to group arbitrary keys by node, and then issue the needed number of (for instance) mget commands to the cluster to retrieve the desired values.

[supercaracal]

Redis Gem has a client-side-consistent-hashing feature. It looks like similar feature is implemented. #687

The above feature might be useful for cluster mode too. Too many hash tags sometimes occur imbalances of shards.

[cbatson]

Thank you @supercaracal. I'm actually going to backpeddle on this request. I misunderstood Redis, but now know it's a limitation that queries involving multiple slots are not permitted. This severely limits the use of multi-key operations like mget, and so this feature request is significantly less valuable as a result.

Here's an example:

The keys aa and cc map to different slots:

irb(main):001:0> key1 = 'aa'
=> "aa"
irb(main):002:0> slot1 = Redis::Cluster::KeySlotConverter.convert(key1)
=> 1180
irb(main):005:0> key2 = 'cc'
=> "cc"
irb(main):006:0> slot2 = Redis::Cluster::KeySlotConverter.convert(key2)
=> 700

On my test cluster, a single node handles slots 0 through 1605, which includes the above to keys. Yet

$ nc xxx.cache.amazonaws.com 6379
MGET aa bb
-CROSSSLOT Keys in request don't hash to the same slot

So my dream of grouping keys by node for mget is shattered.

I will close this but feel free to re-open if you desire.