Consistent hashing (Effect's HashRing): nodes named by their
primary_key() placed on a ring of virtual points, base_weight points
per unit of weight::
ring = HashRing[Runner]()
ring.add(runner_a)
ring.add(runner_b, weight=2)
ring.get("user-1") # the node nearest the input's hash
ring.get_shards(300) # a balanced node for each of 300 shardsHashes are Effect's Hash.string, so the same nodes give the same
assignment as Effect's ring.
get is only as even as the node keys hash: with short keys such as a
and b, 600 inputs job-0 to job-599 split 97 / 503 (Effect's ring
gives the same counts). For an even split, hash each input to one of a fixed
number of shards and place the shards with get_shards, as the cluster
does.
Classes#
On This Page
Classes