[ 
https://issues.apache.org/jira/browse/CASSANDRA-5514?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13645632#comment-13645632
 ] 

Sylvain Lebresne commented on CASSANDRA-5514:
---------------------------------------------

bq. So, we'll need to change that to always promote

I'm not sure that's necessary for that issue. The goal here would be to ignore 
sstable based on basically their metadata, so promoting stuff from the data 
file to the index file shouldn't have any impact here.

*But*, it's true that row tombstones complicate stuff because in theory, as 
soon as a sstable has a row tombstone, then the min/max column name should be 
empty (i.e. cover allthethings). But I think we can probably get away by 
involving the min/max timestamps. More precisely, we can ignore row tombstones 
in the min/max column computation and initially ignore sstable based on this 
min/max column name. But then, we would need to do a pass over the result to 
make sure no column from the result could be overridden by a row tombstone in 
one of the ignored sstable, which can be done using the sstable max timestamp.

Or was that something else you had in mind?
                
> Allow timestamp hints
> ---------------------
>
>                 Key: CASSANDRA-5514
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-5514
>             Project: Cassandra
>          Issue Type: New Feature
>          Components: API, Core
>            Reporter: Jonathan Ellis
>            Assignee: Marcus Eriksson
>             Fix For: 2.0
>
>
> Slice queries can't optimize based on timestamp except for rare cases 
> (CASSANDRA-4116).  However, many common queries involve an implicit time 
> component, where the application author knows that he is only interested in 
> data more recent than X, or older than Y.
> We could use the per-sstable max and min timestamps we track to avoid 
> touching cold data if we could pass a hint to Cassandra about the time range 
> we care about.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

Reply via email to