[jira] [Resolved] (CASSANDRA-7854) Unable to select partition keys directly using IN keyword (no replacement for multi row multiget in thrift)

["Sylvain Lebresne (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/CASSANDRA-7854?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Sylvain Lebresne resolved CASSANDRA-7854.
-----------------------------------------
    Resolution: Duplicate

bq. The issue is not really a duplicate of 7855.

You're right, it's more of a duplicate of CASSANDRA-6875. At least the intent expressed by Todd is.

What Todd was really asking about here was to be able use a IN on the partition keys and a IN on the clustering columns (which was allowed by CASSANDRA-6875).  It is true that the made-up syntax in the description uses a IN on both partition key and clustering columns, which is not stricly equivalent to the former, but Todd clearly didn't intented that generality since the ticket is in the context of replacing thrift multiget and the thrift multiget has never supported the more general form implied by an IN on the full primary key.

So I'm closing again as duplicate to avoid the confusion of reusing it for something it wasn't intented for.

We can create a separate issue for IN on full primary key *but* I'm not so sure it's a good idea because it's yet another form of multi-partition query and we discourage those (as doing multiple one-partition queries concurrently is a better idea in practice). We kind of had to support as much as what multiget was giving us with thrift for political reasons, but adding a new form that we'll spend our time discouraging while that form was never supported and never asked for (to the best of my knowledge) doesn't sound too compeling.


> Unable to select partition keys directly using IN keyword (no replacement for multi row multiget in thrift)
> -----------------------------------------------------------------------------------------------------------
>
>                 Key: CASSANDRA-7854
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-7854
>             Project: Cassandra
>          Issue Type: Bug
>            Reporter: Todd Nine
>            Assignee: Benjamin Lerer
>
> We're converting some old thrift CF's to CQL.  We aren't looking to change the underlying physical structure, since this has proven effective in production.  In order to migrate, we need full select via multi equivalent.  In thrift, the format was as follows.
> (scopeId, scopeType, nodeId, nodeType){ 0x00, timestamp }
> Where we have deliberately designed only 1 column per row.  To translate this to CQL, I have defined the following table.
> {code}
> CREATE TABLE Graph_Marked_Nodes ( 
>  scopeId uuid,
>  scopeType varchar,
>  nodeId uuid,
>  nodeType varchar,
>  timestamp bigint,
>  PRIMARY KEY(scopeId, scopeType, nodeId, nodeType)
> )
> {code}
> I then try to select using the IN keyword.
> {code}
> select timestamp from Graph_Marked_Nodes WHERE (scopeId , scopeType , nodeId , nodeType)  IN ( (5a391596-3181-11e4-a87e-600308a690e2, 'organization', 5a3a2708-3181-11e4-a87e-600308a690e2 ,'test' ),(5a391596-3181-11e4-a87e-600308a690e2, 'organization', 5a3a2709-3181-11e4-a87e-600308a690e2 ,'test' ),(5a391596-3181-11e4-a87e-600308a690e2, 'organization', 5a39fff7-3181-11e4-a87e-600308a690e2 ,'test' ) )
> {code}
> Which results in the following stack trace
> {code}
> Caused by: com.datastax.driver.core.exceptions.InvalidQueryException: Multi-column relations can only be applied to clustering columns: scopeid
> 	at com.datastax.driver.core.Responses$Error.asException(Responses.java:97)
> 	at com.datastax.driver.core.DefaultResultSetFuture.onSet(DefaultResultSetFuture.java:110)
> 	at com.datastax.driver.core.RequestHandler.setFinalResult(RequestHandler.java:235)
> 	at com.datastax.driver.core.RequestHandler.onSet(RequestHandler.java:367)
> 	at com.datastax.driver.core.Connection$Dispatcher.messageReceived(Connection.java:584)
> {code}
> This is still possible via the thrift API.  Apologies in advance if I've filed this erroneously.  I can't find any examples of this type of query anywhere.
> Note that our size grows far too large to fit in a single physical partition (row) if we use only scopeId and scopeType, so we need all 4 data elements to be part of our partition key to ensure we have the distribution we need.



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