[
https://issues.apache.org/jira/browse/PHOENIX-2327?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
James Taylor updated PHOENIX-2327:
----------------------------------
Attachment: PHOENIX-2327_v1.patch
Thanks for the test, [~chunhui.L]. Would you mind trying with this patch and
letting me know if fixes the issue (your test and the other pass with it)?
FWIW, if you're implementing query more with RVCs (which is what we do at
Salesforce), you'll want to add an ORDER BY with the RVC expressions as well.
It'll be optimized out if the rows are already ordered in this way (either in
the data table or any index table).
> Row value constructors failed on the index, when len(table's pks) > 2 and
> table's 1st pk is index's last pk
> -----------------------------------------------------------------------------------------------------------
>
> Key: PHOENIX-2327
> URL: https://issues.apache.org/jira/browse/PHOENIX-2327
> Project: Phoenix
> Issue Type: Bug
> Affects Versions: 4.5.2
> Environment: cdh5.3.6
> Reporter: Chunhui Liu
> Attachments: PHOENIX-2327.patch, PHOENIX-2327_v1.patch
>
>
> 1. Table has more than 2 primary keys;
> 2. Table's 1st pk as index's last pk; eg. table's pks are (pk1, pk2, pk3),
> the failed index's pks are (pk2, pk3, pk1); table's pks are (1, 2, 3, 4),
> failed index's pks are (2, 3, 4, 5, 1);
> 3. Use row value constructors on index with another condition that use one
> pks(not the table's 1st pk);
> 4. You will get "DEGENERATE SCAN OVER TABLE_NAME"
> Here is the Test SQL
> {code:sql}
> DROP TABLE IF EXISTS T;
> CREATE TABLE IF NOT EXISTS T (
> PK1 VARCHAR not null,
> PK2 VARCHAR not null,
> PK3 VARCHAR not null,
> V1 VARCHAR,
> CONSTRAINT PK PRIMARY KEY (PK1, PK2, PK3)
> );
> CREATE INDEX IDX_T ON T
> (
> PK2, PK3, PK1
> );
> UPSERT INTO T VALUES('100', '200', '300', 'V');
> UPSERT INTO T VALUES('101', '201', '301', 'V');
> UPSERT INTO T VALUES('102', '202', '302', 'V');
> UPSERT INTO T VALUES('103', '203', '303', 'V');
> UPSERT INTO T VALUES('104', '204', '304', 'V');
> SELECT * FROM T;
> EXPLAIN
> SELECT PK1, PK2, PK3 FROM T WHERE
> (PK2, PK3, PK1) >= ('202', '302', '102')
> AND PK2 < '204'
> LIMIT 10;
> {code}
> I've tried 3 primary key, here is the results.
> 1. table's pks are (pk1, pk2, pk3);
> 2. 132 means (pk1, pk3, pk2);
> | index's pks order | result |
> | 132 | correct |
> | 213 | correct |
> | 231 | fail |
> | 312 | correct |
> | 321 | correct |
> I've also test this on table with 4, 5 pks
> | len(pks) | failed order |
> | 3 | 231 |
> | 4 | 2341 |
> | 5 | 23451 |
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
