[ 
https://issues.apache.org/jira/browse/CASSANDRA-13478?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16011822#comment-16011822
 ] 

Alex Petrov commented on CASSANDRA-13478:
-----------------------------------------

Currently, SPARSE mode would throw an exception if there're more than 5 items 
per term in the sstable. This results into the not being able to query the 
index and therefore not returning any results.

I'd argue that even though sparse mode is made for the cases when there are 
just a few tokens per sstable, we have to allow any amount. This will result 
into worse performance (because of the token tree reads) but there might be 
cases when giving a hard guarantee on 5 isn't possible to achieve (any user 
generated data). I'd say we have to change the current behaviour and test it 
well.

> SASI Sparse mode overflow corrupts the SSTable
> ----------------------------------------------
>
>                 Key: CASSANDRA-13478
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-13478
>             Project: Cassandra
>          Issue Type: Bug
>          Components: sasi
>         Environment: cqlsh 5.0.1 | Cassandra 3.10 | CQL spec 3.4.4 | Native 
> protocol v4 | ubuntu 14.04
>            Reporter: jack chen
>            Assignee: Alex Petrov
>            Priority: Minor
>         Attachments: schema
>
>
> I have a table, the schema can be seen in attach file
> I would like to search the data using the timestamp data type with lt gt eq 
> as a query condition,
> Ex:
> {code}
> CREATE TABLE XXX.userlist (
>     userid text PRIMARY KEY,
>     lastposttime timestamp
> )
> Select * from userlist where lastposttime> '2017-04-01 16:00:00+0000';
> {code}
> There are 2 conditions :
> If I insert the data and then select it, the result will be correct
> But in case I insert data and then the next day I restart Cassandra, and 
> after that select the data, there will be no data selected
> The difference is that there is no Service restart on th next day in the 
> first manner. Actually, the data are still living in Cassandra, but TimeStamp 
> can’t be used as the query condition



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org
For additional commands, e-mail: commits-h...@cassandra.apache.org

Reply via email to