Highest random weight (aka rendezvous hashing) is an older, simpler, and more general algorithm for determining which subset of available replicas should have a copy of any given piece of content. This is done by hashing the content ID with each replica ID, then sorting the resulting hashes in descending order and choosing the replicas associated with the first k hashes.
This algorithm deterministically selects replicas in such a way that any number of nodes that have the same list of replica IDs will choose the same subset of replicas for each content ID, and each replica will have an equal chance of being selected for any given content ID. It is superior to consistent hashing by reducing the overhead for load balancing and from reindexing during replica churn: instead of every replica getting a large number of keys on a logical ring and passing their content "to the right" on the ring when they spin down, each piece of content is dynamically allotted replicas at the time that it is stored or accessed using only the IDs used for each replica and the ID of the content. Additionally, the load balancing proportions between replicas can be managed precisely by adding replica ID aliases for replicas to increase each replica's chance of being chosen by HRW; e.g. if one replica has twice the capability of any other replica, it would be allocated a second replica ID (e.g. the hash of its initial ID concatenated with a number).
pip install hrwThe simplest method for using this algorithm is to use the choose function
specifying just the content_id and the replica_ids. Doing so will return 2
deterministic, dynamically-sized lists of replica IDs: the first list is the
replicas that should store/cache the content, and the second list is the rest of
the replicas ordered by the likelihood they have the content. This second list
is useful as a fallback: in case none of the chosen replicas have the content,
possibly due to their recent entry into the network, the remaining replicas can
be queried in order of descending likelihood of success. For load balancing
purposes, the order of chosen replica IDs should be shuffled before querying,
but the order of fallback replicas should remain ordered for resilience.
from hashlib import sha256
from hrw import choose
# some list of bytes replica ids, which should be unique
replica_ids = [i.to_bytes(2) for i in range(256)]
# some content
content = b'Lorem ipsum dolor sit amet, something something darkside.'
# set the content_id as the sha256 hash of the content
content_id = sha256(content).digest()
# choose a subset of replicas for storage/caching of the content
chosen_replica_ids, remaining_replica_ids = choose(content_id, replica_ids)
# should print twelve
print([crid.hex() for crid in chosen_replica_ids])
expected = ['004c', '006d', '0047', '004e', '00ee', '008b', '00be', '0016', '0064', '00e2', '0055', '002f']
assert [crid.hex() for crid in chosen_replica_ids] == expected
# should print 244
print(len(remaining_replica_ids))To set the number of replicas to be chosen, supply the named int argument k:
choose(content_id, replica_ids, k=3)If you want simply to order the replicas, use the sort method:
from hrw import sort
replica_ids = [i.to_bytes(2) for i in range(12)]
content_id = b'0123456789abcdef'
ordered = sort(content_id, replica_ids)
print([rid.hex() for rid in ordered])
expected = ['0009', '000b', '0006', '0002', '0003', '0004', '0008', '000a', '0001', '0000', '0005', '0007']
assert [rid.hex() for rid in ordered] == expectedTo use a different hashing algorithm, supply the argument hash_function:
from hashlib import shake_256
from hrw import choose
hash_func = lambda preimage: shake_256(preimage).digest(20)
replica_ids = [i.to_bytes(2) for i in range(256)]
content = b'Lorem ipsum dolor sit amet, something something darkside.'
content_id = hash_func(content)
chosen_replica_ids, remaining_replica_ids = choose(content_id, replica_ids, hash_function=hash_func)
# should print twelve
print([crid.hex() for crid in chosen_replica_ids])
expected = ['00e2', '0061', '000a', '0099', '0024', '00aa', '00bd', '0017', '006b', '00cd', '0079', '00e1']
assert [crid.hex() for crid in chosen_replica_ids] == expected
# should print 244
print(len(remaining_replica_ids))To calculate the number of replicas at which a piece of content should be stored,
use the calculate_k function: k = calculate_k(replica_ids).
Raises TypeError or ValueError for invalid arguments.
To test, clone the repository and run the following:
python tests/test_functions.pyThere are currently 11 tests that document the behavior of the package.
Copyleft (c) 2023 k98kurz
Permission to use, copy, modify, and/or distribute this software for any purpose with or without fee is hereby granted, provided that the above copyleft notice and this permission notice appear in all copies.
Exceptions: this permission is not granted to Alphabet/Google, Amazon, Apple, Microsoft, Netflix, Meta/Facebook, Twitter, or Disney; nor is permission granted to any company that contracts to supply weapons or logistics to any national military; nor is permission granted to any national government or governmental agency; nor is permission granted to any employees, associates, or affiliates of these designated entities.
THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.