[
https://issues.apache.org/jira/browse/CASSANDRA-11401?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15206860#comment-15206860
]
DOAN DuyHai commented on CASSANDRA-11401:
-----------------------------------------
bq. If you ignore the order that Cassandra stores the rows on disk, you can see
that logically speaking, 3.3 returns incorrect results
The problem is that I have designed the table with DESC clustering order for
the {{date}} column. Also the {{SELECT * FROM entity_with_clusterings;}} is
showing the result as Cassandra stores data on disk, which is consistent with
the defined clustering order:
{noformat}
id | uuid | date
| value
---------------------+--------------------------------------+--------------------------+-------
3233835834146573312 | 00000000-0000-0000-0000-000000000000 | 2015-10-03
00:00:00+0000 | val3
3233835834146573312 | 00000000-0000-0000-0000-000000000000 | 2015-10-02
00:00:00+0000 | val2
3233835834146573312 | 00000000-0000-0000-0000-000000000000 | 2015-10-01
00:00:00+0000 | val1
3233835834146573312 | 00000000-0000-0000-0000-000000000001 | 2015-10-05
00:00:00+0000 | val5
3233835834146573312 | 00000000-0000-0000-0000-000000000001 | 2015-10-04
00:00:00+0000 | val4
{noformat}
So if the SELECT returns the above result, logically, all clustering tuples
(uuid, date) >= {{(00....000, 2015-10-02 00:00:00+0000)}} should be {{val2}},
{{val1}}, {{val5}} and {{val4}} and exactly in this order ...
The trick is that the ordering is done on the couple {{(uuid,date)}} and
neither {{uuid}} nor {{date}} alone ...
My problem here is that if you say that *Cassandra 3.4* is returning the
correct result, then it is not consistent with what is displayed by {{SELECT *
...}}
> [Regression] Incorrect results for clustering tuples query
> ----------------------------------------------------------
>
> Key: CASSANDRA-11401
> URL: https://issues.apache.org/jira/browse/CASSANDRA-11401
> Project: Cassandra
> Issue Type: Bug
> Components: CQL
> Environment: *Cassandra 3.4*
> Reporter: DOAN DuyHai
> Priority: Critical
>
> There is a regression since *Cassandra 3.4* with query using clustering tuple
> values:
> {noformat}
> cqlsh:test> CREATE TABLE IF NOT EXISTS entity_with_clusterings(
> id bigint,
> uuid uuid,
> date timestamp,
> value text,
> PRIMARY KEY(id, uuid, date))
> WITH CLUSTERING ORDER BY(uuid ASC, date DESC);
> cqlsh:test> INSERT INTO entity_with_clusterings(id, uuid, date, value)
> VALUES(3233835834146573312, 00000000-0000-0000-0000-000000000000, '2015-10-01
> 00:00:00+0000', 'val1');
> cqlsh:test> INSERT INTO entity_with_clusterings(id, uuid, date, value)
> VALUES(3233835834146573312, 00000000-0000-0000-0000-000000000000, '2015-10-02
> 00:00:00+0000', 'val2');
> cqlsh:test> INSERT INTO entity_with_clusterings(id, uuid, date, value)
> VALUES(3233835834146573312, 00000000-0000-0000-0000-000000000000, '2015-10-03
> 00:00:00+0000', 'val3');
> cqlsh:test> INSERT INTO entity_with_clusterings(id, uuid, date, value)
> VALUES(3233835834146573312, 00000000-0000-0000-0000-000000000001, '2015-10-04
> 00:00:00+0000', 'val4');
> cqlsh:test> INSERT INTO entity_with_clusterings(id, uuid, date, value)
> VALUES(3233835834146573312, 00000000-0000-0000-0000-000000000001, '2015-10-05
> 00:00:00+0000', 'val5');
> cqlsh:test> SELECT * FROM entity_with_clusterings;
> id | uuid | date
> | value
> ---------------------+--------------------------------------+--------------------------+-------
> 3233835834146573312 | 00000000-0000-0000-0000-000000000000 | 2015-10-03
> 00:00:00+0000 | val3
> 3233835834146573312 | 00000000-0000-0000-0000-000000000000 | 2015-10-02
> 00:00:00+0000 | val2
> 3233835834146573312 | 00000000-0000-0000-0000-000000000000 | 2015-10-01
> 00:00:00+0000 | val1
> 3233835834146573312 | 00000000-0000-0000-0000-000000000001 | 2015-10-05
> 00:00:00+0000 | val5
> 3233835834146573312 | 00000000-0000-0000-0000-000000000001 | 2015-10-04
> 00:00:00+0000 | val4
> (5 rows)
> cqlsh:test > SELECT uuid,date,value
> FROM entity_with_clusterings
> WHERE id=3233835834146573312
> AND (uuid,date)>=(00000000-0000-0000-0000-000000000000,'2015-10-02
> 00:00:00+0000')
> AND (uuid,date)<(00000000-0000-0000-0000-000000000001, '2015-10-04
> 00:00:00+0000');
> uuid | date |
> value
> --------------------------------------+---------------------------------+-------
> 00000000-0000-0000-0000-000000000000 | 2015-10-03 00:00:00.000000+0000 |
> val3
> 00000000-0000-0000-0000-000000000000 | 2015-10-02 00:00:00.000000+0000 |
> val2
> {noformat}
> The same query with *Cassandra 3.3* returns correct answer:
> {noformat}
> cqlsh:test> SELECT uuid,date,value
> FROM entity_with_clusterings
> WHERE id=3233835834146573312
> AND (uuid,date)>=(00000000-0000-0000-0000-000000000000,'2015-10-02
> 00:00:00+0000')
> AND (uuid,date)<(00000000-0000-0000-0000-000000000001, '2015-10-04
> 00:00:00+0000');
> uuid | date | value
> --------------------------------------+--------------------------+-------
> 00000000-0000-0000-0000-000000000000 | 2015-10-02 00:00:00+0000 | val2
> 00000000-0000-0000-0000-000000000000 | 2015-10-01 00:00:00+0000 | val1
> 00000000-0000-0000-0000-000000000001 | 2015-10-05 00:00:00+0000 | val5
> {noformat}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
