[
https://issues.apache.org/jira/browse/DRILL-5064?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Khurram Faraaz updated DRILL-5064:
----------------------------------
Attachment: typeall_r.parquet
typeall_l.parquet
> wrong results - equijoin involving interval day type columns
> ------------------------------------------------------------
>
> Key: DRILL-5064
> URL: https://issues.apache.org/jira/browse/DRILL-5064
> Project: Apache Drill
> Issue Type: Bug
> Components: Execution - Flow
> Affects Versions: 1.9.0
> Reporter: Khurram Faraaz
> Attachments: typeall_l.parquet, typeall_r.parquet
>
>
> The problem exists on Drill 1.9.0 git commit ID: 4312d65 (i.e difference in
> results) even without the nullability check ... OR ( t1.col_intrvl_day IS
> NULL AND t2.col_intrvl_day IS NULL ). A simple equi-join involving interval
> type columns returns wrong results.
> {noformat}
> postgres=# SELECT t1.col_intrvl_dy , t2.col_intrvl_dy
> postgres-# FROM typeall_l t1 INNER JOIN typeall_r t2
> postgres-# ON t1.col_intrvl_dy= t2.col_intrvl_dy;
> col_intrvl_dy | col_intrvl_dy
> --------------------------+--------------------------
> 9 years 10 mons 9 days | 9 years 10 mons 9 days
> 25 years 11 mons 22 days | 25 years 11 mons 22 days
> 23 years 10 mons 25 days | 23 years 10 mons 25 days
> 26 years 8 mons 12 days | 26 years 8 mons 12 days
> 13 years 6 mons 31 days | 13 years 7 mons 1 day
> 27 years 11 mons 17 days | 27 years 11 mons 17 days
> 27 years 11 mons 17 days | 27 years 11 mons 17 days
> 23 years 5 mons 20 days | 23 years 5 mons 20 days
> 14 years 7 mons 12 days | 14 years 7 mons 12 days
> 4 years 10 mons 11 days | 4 years 10 mons 11 days
> 21 years 31 days | 21 years 1 mon 1 day
> 23 years 10 mons 25 days | 23 years 10 mons 25 days
> (12 rows)
> {nformat}
> {noformat}
> 0: jdbc:drill:schema=dfs.tmp> SELECT t1.col_intrvl_day , t2.col_intrvl_day
> . . . . . . . . . . . . . . > FROM typeall_l t1 INNER JOIN typeall_r t2
> . . . . . . . . . . . . . . > ON t1.col_intrvl_day= t2.col_intrvl_day;
> +-----------------+------------------+
> | col_intrvl_day | col_intrvl_day0 |
> +-----------------+------------------+
> | P9DT33156S | P9DT33156S |
> | P22DT25501S | P22DT25501S |
> | P25DT10592S | P25DT10592S |
> | P17DT30024S | P17DT30024S |
> | P17DT30024S | P17DT30024S |
> +-----------------+------------------+
> 5 rows selected (0.515 seconds)
> {noformat}
> Parquet schema details of the two files used in the join query
> [root@centos-01 intrvl]# ~/parquet-tools/parquet-schema
> typeall_l/0_0_0.parquet
> message root {
> optional int32 col_int;
> optional binary col_chr (UTF8);
> optional binary col_vrchr1 (UTF8);
> optional binary col_vrchr2 (UTF8);
> optional int32 col_dt (DATE);
> optional int32 col_tim (TIME_MILLIS);
> optional int64 col_tmstmp (TIMESTAMP_MILLIS);
> optional float col_flt;
> optional fixed_len_byte_array(12) col_intrvl_yr (INTERVAL);
> optional fixed_len_byte_array(12) col_intrvl_day (INTERVAL);
> optional boolean col_bln;
> }
> [root@centos-01 intrvl]# ~/parquet-tools/parquet-schema
> typeall_r/0_0_0.parquet
> message root {
> optional int32 col_int;
> optional binary col_chr (UTF8);
> optional binary col_vrchr1 (UTF8);
> optional binary col_vrchr2 (UTF8);
> optional int32 col_dt (DATE);
> optional int32 col_tim (TIME_MILLIS);
> optional int64 col_tmstmp (TIMESTAMP_MILLIS);
> optional float col_flt;
> optional fixed_len_byte_array(12) col_intrvl_yr (INTERVAL);
> optional fixed_len_byte_array(12) col_intrvl_day (INTERVAL);
> optional boolean col_bln;
> }
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
