An alternative to the ring based, consistent hashing. This is a fast thread safe implementation of Rendezvous (Highest Random Weight, HRW) hashing. An algorithm that allows clients to achieve distributed agreement on which node (or proxy) a given key is to be placed in. This implementation has the following properties.
- Non-blocking reads : Determining which node a key belongs to is always non-blocking. Adding and removing nodes however blocks each other.
- Low overhead: providing using a hash function of low overhead. Throughput can be computed as (hashes computable per sec)/node count
- Load balancing: Since the hash function is randomizing, each of the n nodes is equally likely to receive the key K. Loads are uniform across the sites.
- High hit rate: Since all clients agree on placing an key K into the same node N , each fetch or placement of K into N yields the maximum utility in terms of hit rate. The key K will always be found unless it is evicted by some replacement algorithm at N.
- Minimal disruption: When a node is removed, only the keys mapped to that node need to be remapped and they will be distributed evenly
Source: https://en.wikipedia.org/wiki/Rendezvous_hashing
In comparison (source code) of Consistent hashing and Rendezvous hashing, consider the following load distribution after removing a couple nodes in a 5 node ring:
Only node4 takes the load of the 2 that were removed. However using HRW the distribution remains even
This example uses a rather simple ring for consistent hash implementation however and this extreme unbalance can be mitigated by adding the nodes many times (ie ~200) throughout the ring. These virtual nodes (or vnodes) are used in databases like Riak and Cassandra. Many libraries however do not implement vnodes.
Example:
private static final Funnel<CharSequence> strFunnel = Funnels.stringFunnel(Charset.defaultCharset());
// prepare 5 initial nodes "node1", "node2" ... "node5"
List<String> nodes = Lists.newArrayList();
for(int i = 0 ; i < 5; i ++) {
nodes.add("node"+i);
}
// create HRW instance
RendezvousHash<String, String> h = new RendezvousHash(Hashing.murmur3_128(), strFunnel, strFunnel, nodes);
String node = h.get("key"); // returns "node1"
// remove "node1" from pool
h.remove(node);
h.get("key"); // returns "node2"
// add "node1" back into pool
h.add(node);
h.get("key"); // returns "node1"