[
https://issues.apache.org/jira/browse/CASSANDRA-14777?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Benedict updated CASSANDRA-14777:
---------------------------------
Component/s: (was: Legacy/Coordination)
Consistency/Coordination
> ReadExecutor cannot upgrade digest to data request
> --------------------------------------------------
>
> Key: CASSANDRA-14777
> URL: https://issues.apache.org/jira/browse/CASSANDRA-14777
> Project: Cassandra
> Issue Type: Improvement
> Components: Consistency/Coordination
> Reporter: Benedict
> Priority: Minor
>
> ReadExecutor cannot send a request to any node it has already contacted, as
> the accounting cannot handle multiple responses from the same host. In fact,
> any bug wherein a host somehow responded twice anyway would lead to a
> consistency level breakage.
> The main problem incurred here is that if we receive no data response, we may
> not be able to speculate to obtain another one, particularly if only
> transient replicas remain to be contacted. We might also be able to avoid
> speculating two data requests in ‘always’ speculating, also, as presumably
> this was to ensure the same problem was not encountered (wherein by ‘always’
> speculating upfront, we could not recover from our only data response not
> arriving).
> Ideally, we would maintain a map of endpoint->responses, and simply use the
> ‘best’ response we have for each endpoint, and only count the first response
> towards our consistency counter.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
