cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Benjamin Lerer (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (CASSANDRA-9767) Allow the selection of columns together with aggregates
Date Thu, 16 Jul 2015 19:54:04 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-9767?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14630227#comment-14630227
] 

Benjamin Lerer commented on CASSANDRA-9767:
-------------------------------------------

Thanks for the review.

{quote}One questionable change: you've introduced new isSet fields in SimpleSelector and WritetimeOrTTLSelector.
I think the if (!isSet) test can safely changed to if (current==null). As far as I understood
we are returning a value from a "random" row anyway - so it wouldn't change that "contract"
- but we'd safe one field. If you think this is ok, free to change it on commit.{quote}

Returning a random value was not making a lot of sense in my opinion. Most of the databases
return the values of the first row. So, I decided to do the same. 
I initially started with {{if (current==null)}} but I faced a problem in the case where the
column value of the first row was {{null}}. 

> Allow the selection of columns together with aggregates
> -------------------------------------------------------
>
>                 Key: CASSANDRA-9767
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-9767
>             Project: Cassandra
>          Issue Type: Wish
>          Components: Core
>         Environment: Cassandra 2.0.16
> Ubuntu 15.04
>            Reporter: Ajay
>            Assignee: Benjamin Lerer
>            Priority: Minor
>
> Lets assume we have a column family as below:
> create table sample ( track_id int, user_id int, country varchar, primary key ((track_id),
user_id));
> where track_id is the partition key.
> Now to aggregate the number of rows for a single track_id, we can query using CQL as
below:
> select count(*) where track_id = 1 and user_id = 1;
> But that will return only the count. If we need the other columns along with the count,
we cannot query as below as it throws error:
>  select count(*), country  from sample where track_id = 1 and user_id = 1;
> Bad Request: line 1:15 mismatched input ',' expecting K_FROM.
> In this case, all rows for a given track_id and user_id will have the same value for
country. So we should be able to query as above.  Also in SQL, it is possible to select columns
along with aggregate functions.
> Though I know that Cassandra is not analytics (unlike Hadoop and Spark), we need some
basic aggregate functions like min, max, avg etc....Though performance wise it might not be
efficient, but it is better done in the cassandra side (as it uses native protocol) than we
getting all rows in the client and doing the basic aggregation.  It cannot used just as a
data store (as garbage-in garbage-out). In that context, currently CQL is pretty limited.
Just for getting data out of cassandra, we will have to spark though we will not be doing
much analytics on it.



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

Mime
View raw message