Sankar Hariappan created HIVE-21564:
---------------------------------------
Summary: Load data into a bucketed table is ignoring partitions
specs and loading data into default partition.
Key: HIVE-21564
URL: https://issues.apache.org/jira/browse/HIVE-21564
Project: Hive
Issue Type: Bug
Components: Query Planning
Affects Versions: 4.0.0
Reporter: Sankar Hariappan
Assignee: Sankar Hariappan
When running below command to load data into bucketed tables it is not loading
into specified partition instead loaded into default partition.
LOAD DATA INPATH '/tmp/files/000000_0' OVERWRITE INTO TABLE call
PARTITION(year_partition=2012, month=12);
SELECT * FROM call WHERE year_partition=2012 AND month=12; --> returns 0 rows.
{code}
CREATE TABLE call(
date_time_date date,
ssn string,
name string,
location string)
PARTITIONED BY (
year_partition int,
month int)
CLUSTERED BY (
date_time_date)
SORTED BY (
date_time_date ASC)
INTO 1 BUCKETS
STORED AS ORC;
{code}
If set hive.exec.dynamic.partition to false, it fails with below error.
{code}
Error: Error while compiling statement: FAILED: SemanticException 1:18 Dynamic
partition is disabled. Either enable it by setting
hive.exec.dynamic.partition=true or specify partition column values. Error
encountered near token 'month' (state=42000,code=40000)
{code}
When we "set hive.strict.checks.bucketing=false;", the load works fine.
This is a behaviour imposed by HIVE-15148 to avoid incorrectly named data files
being loaded to the bucketed tables. In customer use case, if the files are
named properly with bucket_id (00000_0, 00000_1 etc), then it is safe to set
this flag to false.
However, current behaviour of loading into default partitions when
hive.strict.checks.bucketing=true and partitions specified, was a bug injected
by HIVE-19311 where the given query is re-written into a insert query (to
handle incorrect file names and Orc versions) but missed to incorporate the
partitions specs to it.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
