[ 
https://issues.apache.org/jira/browse/CASSANDRA-14723?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16894032#comment-16894032
 ] 

sankalp kohli commented on CASSANDRA-14723:
-------------------------------------------

Another extension of this idea is to use IR to reduce the window in which we 
need this. Here is how it will work and I think deserve another Jira
 # Dont bump the consistency level or do anything special when a new machine is 
joining as part of expansion.
 # When the new machine has streamed all the data, we now bump consistency 
level to require 1 more ack or use this idea. 
 # We run IR on all nodes including new one(need majority to be up).
 # New node take over membership. 

Assuming IR to be faster than bootstrap, we will reduce the window in which we 
will need more acks. 

This idea will be possible only when IR is running as part of the database and 
is run regularly and is not backed up. 

> Increase write availability during range movements
> --------------------------------------------------
>
>                 Key: CASSANDRA-14723
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-14723
>             Project: Cassandra
>          Issue Type: Improvement
>          Components: Legacy/Coordination
>            Reporter: Benedict
>            Priority: Normal
>             Fix For: 4.x
>
>
> If pending ranges were to be paired with their ‘natural’ replicas, instead of 
> requiring QUORUM+#pending, we could instead require QUORUM, but count both 
> owners of the pending range as a combined 1 - where they are only counted if 
> both return success.
> As an example, in a cluster with RF=3, a pending range movement would 
> currently require 3 nodes to respond, so must include at least one of the 
> pending nodes.  Under the proposed scheme, the 2 non-participating nodes 
> could respond as normal to reach a QUORUM, not requiring either of the 
> pending nodes to respond.



--
This message was sent by Atlassian JIRA
(v7.6.14#76016)

---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org
For additional commands, e-mail: commits-h...@cassandra.apache.org

Reply via email to