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

Posted to dev@gossip.apache.org by "Edward Capriolo (JIRA)" <ji...@apache.org> on 2017/01/08 16:19:58 UTC

[jira] [Created] (GOSSIP-39) User controlled Gossip

Edward Capriolo created GOSSIP-39:
-------------------------------------

             Summary: User controlled Gossip
                 Key: GOSSIP-39
                 URL: https://issues.apache.org/jira/browse/GOSSIP-39
             Project: Gossip
          Issue Type: New Feature
            Reporter: Edward Capriolo
             Fix For: 0.1.2


There are two interesting usecases:

* Cassandra does this 'trick' where nodes gossip more often to seeds so that the cluster converge faster
* It would be ideal if we could gossip more to some nodes than others. For example if we have a cluster in two data centers if they were more likely to gossip inside the wan

The challenge here is flexibility vs complexity. Cassandra has a concept of Datacenter and Rack, do we make these 1st class parameters or do we let the logic consider ANY user data?




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