[
https://issues.apache.org/jira/browse/CASSANDRA-6814?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13942604#comment-13942604
]
Vijay edited comment on CASSANDRA-6814 at 3/21/14 12:56 AM:
Hi Brandon,
If we have to add the assert then we have to force the DS to give priority for
localhost and not the remote one.
{quote} dsnitch should be able to override reading from the local machine{quote}
I am kind of neutral on this because, I have been in projects where i had to
disable DS, since we manage the latency info elsewhere and route to the right
node (cost of a n/w hop).
Having said this there is definitely a case where the local node has a bad disk
or a lot queued up where DS can help reduce the load.
was (Author: vijay2...@yahoo.com):
Hi Brandon,
If we have to add the assert then we have to force the DS to give priority for
localhost and not the remote one.
{quote} dsnitch should be able to override reading from the local machine{quote}
I am kind of neutral on this because, I have been in projects where i had to
disable DS, since we manage the latency info elsewhere and route to the right
node (costing of a n/w hop).
Having said this there is definitely a case where the local node has a bad disk
or a lot queued up where DS can help reduce the load.
We should ensure that AbstractReadExecutor.makeDigestRequests never calls the
local machine
---
Key: CASSANDRA-6814
URL: https://issues.apache.org/jira/browse/CASSANDRA-6814
Project: Cassandra
Issue Type: Improvement
Reporter: Benedict
Assignee: Vijay
Priority: Minor
Fix For: 3.0
Attachments: 0001-CASSANDRA-6814.patch
It should always opt to perform the actual read locally, when possible, and
this code path should assert that it does not read locally.
--
This message was sent by Atlassian JIRA
(v6.2#6252)