You are viewing a plain text version of this content. The canonical link for it is here.

Posted to commits@cassandra.apache.org by "Drew Kutcharian (JIRA)" <ji...@apache.org> on 2015/02/24 21:59:04 UTC

[jira] [Created] (CASSANDRA-8861) HyperLogLog Collection Type

Drew Kutcharian created CASSANDRA-8861:
------------------------------------------

             Summary: HyperLogLog Collection Type
                 Key: CASSANDRA-8861
                 URL: https://issues.apache.org/jira/browse/CASSANDRA-8861
             Project: Cassandra
          Issue Type: Wish
            Reporter: Drew Kutcharian


Considering that HyperLogLog and its variants have become pretty popular in analytics space and Cassandra has "read-before-write" collections (Lists), I think it would not be too painful to add support for HyperLogLog "collection" type. They would act similar to CQL 3 Sets, meaning you would be able to "set" the value and "add" an element, but you won't be able to remove an element. Also, when getting the value of a HyperLogLog collection column, you'd get the cardinality.

There are a couple of good attributes with HyperLogLog which fit Cassandra pretty well.
- Adding an element is idempotent (adding an existing element doesn't change the HLL)
- HLL can be thought of as a CRDT, since we can safely merge them. Which means we can merge two HLLs during read-repair. But if that's too much work, I guess we can even live with LWW since these counts are "estimates" after all.

There is already a proof of concept at:
http://vilkeliskis.com/blog/2013/12/28/hacking_cassandra.html




--
This message was sent by Atlassian JIRA
(v6.3.4#6332)