[ 
https://issues.apache.org/jira/browse/CASSANDRA-10341?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14939040#comment-14939040
 ] 

Yuki Morishita commented on CASSANDRA-10341:
--------------------------------------------

We don't need to read SSTables again, since we already obtain partition key 
when reading from network.
So we can construct set of partition keys to invalidate, or if those are too 
many, we can constuct ranges of partition keys.
Iterate over keys after adding SSTables in row cache and check if key is in 
ranges, we can drop cache.


> Streaming does not guarantee cache invalidation
> -----------------------------------------------
>
>                 Key: CASSANDRA-10341
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-10341
>             Project: Cassandra
>          Issue Type: Bug
>          Components: Core
>            Reporter: Benedict
>            Assignee: Paulo Motta
>
> Looking at the code, we attempt to invalidate the row cache for any rows we 
> receive via streaming, however we invalidate them immediately, before the new 
> data is available. So, if it is requested (which is likely if it is "hot") in 
> the interval, it will be re-cached and not invalidated.



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

Reply via email to