[
https://issues.apache.org/jira/browse/FLINK-23761?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
zhengjiewen updated FLINK-23761:
--------------------------------
Description:
When Run a Flink Sql which have Group By oprator, and filed type is INT
type，the result is incorrect.
I try to CAST the field to BIGINT type OR STRING type, the result is correct.
the SQL is folllow: 【receipt_year ，receipt_month {color:#172b4d}is INT
type{color} 】
{code:java}
" SELECT " +
" ware_id, " +
" supplier_id," +
" supplier_name," +
" receipt_year as years, " +
" receipt_month as months, " +
" SUM(COALESCE(receipt_count,0) * COALESCE(price,0.0)) /
SUM(COALESCE(receipt_count,0)) as supply_avg_price" +
" FROM " +
"
`kudu`.`default_database`.`impala::cube_kudu.dwd_storage_purchase_receipt` " +
" WHERE " +
" receipt_date BETWEEN TO_TIMESTAMP('"+ beginTime +"')
AND TO_TIMESTAMP('"+ endTime +"') " +
" GROUP BY " +
" ware_id, supplier_id, supplier_name, receipt_year ,
receipt_month "
{code}
{color:#172b4d}Here is the execution plan:{color}
{color:#172b4d}!image-2021-08-13-16-08-31-970.png!{color}
{color:#172b4d}When I CAST to another type like BIGINT, the plan is this:{color}
{color:#172b4d}!image-2021-08-13-16-09-19-951.png!{color}
DDL:
{code:java}
CREATE TABLE cube_kudu.dwd_storage_purchase_receipt (
receipt_date TIMESTAMP comment '采购入库日期（审核时间）',
ware_id STRING COMMENT '商品ID',
supplier_id STRING COMMENT '供应商ID',
supplier_name STRING COMMENT '供应商名称',
receipt_year INT COMMENT '入库年份',
receipt_month INT COMMENT '入库月份',
receipt_count DECIMAL(18, 4) COMMENT '入库数量',
price DECIMAL(18,4) COMMENT '采购商品单价',
PRIMARY key (receipt_date, ware_id)
}
PARTITION BY HASH(receipt_date) PARTITIONS 8
STORED AS KUDU;
{code}
was:
When Run a Flink Sql which have Group By oprator, and filed type is INT
type，the result is incorrect.
I try to CAST the field to BIGINT type OR STRING type, the result is correct.
the SQL is folllow: 【receipt_year ，receipt_month {color:#172b4d}is INT
type{color} 】
{code:java}
" SELECT " +
" ware_id, " +
" supplier_id," +
" supplier_name," +
" receipt_year as years, " +
" receipt_month as months, " +
" SUM(COALESCE(receipt_count,0) * COALESCE(price,0.0)) /
SUM(COALESCE(receipt_count,0)) as supply_avg_price" +
" FROM " +
"
`kudu`.`default_database`.`impala::cube_kudu.dwd_storage_purchase_receipt` " +
" WHERE " +
" receipt_date BETWEEN TO_TIMESTAMP('"+ beginTime +"')
AND TO_TIMESTAMP('"+ endTime +"') " +
" GROUP BY " +
" ware_id, supplier_id, supplier_name, receipt_year ,
receipt_month "
{code}
{color:#172b4d}Here is the execution plan:{color}
> FlinkSQL1.12 BatchMode Group By with INT type can not produce the correct
> result
> --------------------------------------------------------------------------------
>
> Key: FLINK-23761
> URL: https://issues.apache.org/jira/browse/FLINK-23761
> Project: Flink
> Issue Type: Bug
> Components: Table SQL / Planner
> Affects Versions: 1.12.2
> Reporter: zhengjiewen
> Priority: Major
> Attachments: image-2021-08-13-16-08-31-970.png,
> image-2021-08-13-16-09-19-951.png
>
>
> When Run a Flink Sql which have Group By oprator, and filed type is INT
> type，the result is incorrect.
> I try to CAST the field to BIGINT type OR STRING type, the result is correct.
> the SQL is folllow: 【receipt_year ，receipt_month {color:#172b4d}is INT
> type{color} 】
> {code:java}
> " SELECT " +
> " ware_id, " +
> " supplier_id," +
> " supplier_name," +
> " receipt_year as years, " +
> " receipt_month as months, " +
> " SUM(COALESCE(receipt_count,0) * COALESCE(price,0.0))
> / SUM(COALESCE(receipt_count,0)) as supply_avg_price" +
> " FROM " +
> "
> `kudu`.`default_database`.`impala::cube_kudu.dwd_storage_purchase_receipt` " +
> " WHERE " +
> " receipt_date BETWEEN TO_TIMESTAMP('"+ beginTime +"')
> AND TO_TIMESTAMP('"+ endTime +"') " +
> " GROUP BY " +
> " ware_id, supplier_id, supplier_name, receipt_year ,
> receipt_month "
> {code}
> {color:#172b4d}Here is the execution plan:{color}
> {color:#172b4d}!image-2021-08-13-16-08-31-970.png!{color}
>
> {color:#172b4d}When I CAST to another type like BIGINT, the plan is
> this:{color}
> {color:#172b4d}!image-2021-08-13-16-09-19-951.png!{color}
>
> DDL:
> {code:java}
> CREATE TABLE cube_kudu.dwd_storage_purchase_receipt (
> receipt_date TIMESTAMP comment '采购入库日期（审核时间）',
> ware_id STRING COMMENT '商品ID',
> supplier_id STRING COMMENT '供应商ID',
> supplier_name STRING COMMENT '供应商名称',
> receipt_year INT COMMENT '入库年份',
> receipt_month INT COMMENT '入库月份',
> receipt_count DECIMAL(18, 4) COMMENT '入库数量',
> price DECIMAL(18,4) COMMENT '采购商品单价',
> PRIMARY key (receipt_date, ware_id)
> }
> PARTITION BY HASH(receipt_date) PARTITIONS 8
> STORED AS KUDU;
> {code}
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
