consistent-hash: support reading by single vnode for specific states

[xx01cyx]

Currently, all states will maintain a set of vnodes, which are stored in the state's keyspace. However, every state of hash agg and hash join will only query data that belongs to one vnode. Therefore, we could narrow the range of vnodes when reading data from storage, i.e., only pass in one vnode instead of a set of vnodes.

The vnode that belongs to a state could be computed when the state is created.

[xx01cyx]

cc. @fuyufjh @hzxa21

[skyzh]

So we'll need to compute hash before every get and prefix scan in storage?

[xx01cyx]

So we'll need to compute hash before every get and prefix scan in storage?

For MV, maybe yes. For other states, we could directly use the pre-computed vnode.

[hzxa21]

So we'll need to compute hash before every get and prefix scan in storage?

For MV, maybe yes. For other states, we could directly use the pre-computed vnode.

How to pre-compute the VNode of a get/prefix_scan key beforehand? We don't know the key until we need to read it.

For internal state (i.e. non-MV state), I think compure hash before every get/prefix_scan is okay since we will only hit one VNode.

[fuyufjh]

I think this issue mainly focuses on the internal state of HashAgg & HashJoin. They both hold a bunch of ManagedStates for each join/agg key, where each of them only touches one vnode, determined by its join/agg key.