Object.keys is expensive when called on thousands of keys. Gotta go fast

[jminarik-atl]

We use this on a client where our ring contains ~ 200 nodes. With 160 replicas, that is 32,000 items in the ring. CPU profiling has revealed that calling Object.keys(this.ring) turned out to be expensive when called frequently (called twice in getNode())

[jminarik-atl]

Benchmark results

ORIGINAL should get correct ring length (27082 ms)
FAST should get correct ring length (167 ms)

Benchmark code:

describe('Faster Consistent Hashing class', () => {
  const nodes = Array(200)
    .fill(1)
    .map((_, idx) => idx);
  const chash = new ConsistentHashing(nodes);
  const chashFast = new FastConsistentHashing(nodes);
  it('ORIGINAL should get correct ring length', async () => {
    await benchmark.record(
      () => {
        chash.getRingLength();
      },
      { iterations: 10000 },
    );
  });
  it('FAST should get correct ring length', async () => {
    await benchmark.record(
      () => {
        chashFast.getRingLength();
      },
      { iterations: 10000 },
    );
  });
});