Request:- How can I insert partition key pair into each parquet file while inserting data into Hive/Impala table.
Hive Table DDL
[
create external table db.tbl_name ( col1 string, col2 string)
Partitioned BY (date_col string)
STORED AS parquet
LOCATION 'hdfs_path/db/tbl_name'
]
Let's insert data into this hive table.
INSERT INTO db.tbl_name PARTITION (date_col=2020-07-26) VALUES ('test1_col1','test1_col2')
Once records get inserted, let's view data into parquet file using parquet-tools or any other tool.
parquet-tool cat hdfs_path/db/tbl_name/date_col=2020-07-26/parquet_file.parquet
Below would be the view.
**********************
col1 = test1_col1
col2 = test1_col2
**********************
However, if I fire following HQL query on Hive/Impala, then it will read partition value from metadata.
**Query**- select * from db.tbl_name
**Result** -
col1 col2 date_col
test1_col1 test1_col2 2020-07-26
Question- Is there any way, where we can view partition columnn name and value in parquet file like below.
col1 = test1_col1
col2 = test1_col2
date_col = 2020-07-26
Please use this -
INSERT INTO db.tbl_name PARTITION (date_col) VALUES ('test1_col1','test1_col2','2020-07-26');
Always mention partition name inside brackets() like above. And then in the values/select clause, order the partition column in the end.
Thats all you need to insert into hive/impala partitioned table.
Related
I'm not able to import data on partitioned table in Hive.
Here is how I create the table
CREATE TABLE IF NOT EXISTS title_ratings
(
tconst STRING,
averageRating DOUBLE,
numVotes INT
)
ROW FORMAT DELIMITED
FIELDS TERMINATED BY '\t'
STORED AS TEXTFILE
TBLPROPERTIES("skip.header.line.count"="1");
And then I load the data into it : LOAD DATA INPATH '/title.ratings.tsv.gz' INTO TABLE eval_hive_db.title_ratings;
It works fine till here. Now I want to create a dynamic partitioned table. First of all, I setup theses params:
SET hive.exec.dynamic.partition=true;
SET hive.exec.dynamic.partition.mode=nonstrict;
I now create my partitioned table:
CREATE TABLE IF NOT EXISTS title_ratings_part
(
tconst STRING,
numVotes INT
)
PARTITIONED BY (averageRating DOUBLE)
ROW FORMAT DELIMITED
FIELDS TERMINATED BY '\n'
STORED AS TEXTFILE;
insert into title_ratings_part partition(title_ratings) select tconst, averageRating, numVotes from title_ratings;
(I also tried with numVotes instead by the way)
And I receive this error: FAILED: ValidationFailureSemanticException eval_hive_db.title_ratings_part: Partition spec {title_ratings=null} contains non-partition columns
Someone can help me please?
Ideally, I want to partition my table by averageRating (less than 2, between 2 and 4, and greater than 4)
You can run this command to check if there are null values or not.
select count(averageRating) from title_ratings group by averageRating;
Now, if there are null values in this column then you will get the count, which you have to fill then apply partitioning again.
Partition column is stored as last column in a table so while inserting you need to maintain correct order in select statement.
Pls change order of columns in select.
insert into title_ratings_part partition(title_ratings)
Select
Tconst,
numVotes,
averageRating --orderwise this should always be last column
from title_ratings
I have orc table in hive I want to drop column from this table
ALTER TABLE table_name drop col_name;
but I am getting the following exception
Error occurred executing hive query: OK FAILED: ParseException line 1:35 mismatched input 'user_id1' expecting PARTITION near 'drop' in drop partition statement
Can any one help me or provide any idea to do this? Note, I am using hive 0.14
You cannot drop column directly from a table using command ALTER TABLE table_name drop col_name;
The only way to drop column is using replace command. Lets say, I have a table emp with id, name and dept column. I want to drop id column of table emp. So provide all those columns which you want to be the part of table in replace columns clause. Below command will drop id column from emp table.
ALTER TABLE emp REPLACE COLUMNS( name string, dept string);
There is also a "dumb" way of achieving the end goal, is to create a new table without the column(s) not wanted. Using Hive's regex matching will make this rather easy.
Here is what I would do:
-- make a copy of the old table
ALTER TABLE table RENAME TO table_to_dump;
-- make the new table without the columns to be deleted
CREATE TABLE table AS
SELECT `(col_to_remove_1|col_to_remove_2)?+.+`
FROM table_to_dump;
-- dump the table
DROP TABLE table_to_dump;
If the table in question is not too big, this should work just well.
suppose you have an external table viz. organization.employee as: (not including TBLPROPERTIES)
hive> show create table organization.employee;
OK
CREATE EXTERNAL TABLE `organization.employee`(
`employee_id` bigint,
`employee_name` string,
`updated_by` string,
`updated_date` timestamp)
ROW FORMAT SERDE
'org.apache.hadoop.hive.ql.io.orc.OrcSerde'
STORED AS INPUTFORMAT
'org.apache.hadoop.hive.ql.io.orc.OrcInputFormat'
OUTPUTFORMAT
'org.apache.hadoop.hive.ql.io.orc.OrcOutputFormat'
LOCATION
'hdfs://getnamenode/apps/hive/warehouse/organization.db/employee'
You want to remove updated_by, updated_date columns from the table. Follow these steps:
create a temp table replica of organization.employee as:
hive> create table organization.employee_temp as select * from organization.employee;
drop the main table organization.employee.
hive> drop table organization.employee;
remove the underlying data from HDFS (need to come out of hive shell)
[nameet#ip-80-108-1-111 myfile]$ hadoop fs -rm hdfs://getnamenode/apps/hive/warehouse/organization.db/employee/*
create the table with removed columns as required:
hive> CREATE EXTERNAL TABLE `organization.employee`(
`employee_id` bigint,
`employee_name` string)
ROW FORMAT SERDE
'org.apache.hadoop.hive.ql.io.orc.OrcSerde'
STORED AS INPUTFORMAT
'org.apache.hadoop.hive.ql.io.orc.OrcInputFormat'
OUTPUTFORMAT
'org.apache.hadoop.hive.ql.io.orc.OrcOutputFormat'
LOCATION
'hdfs://getnamenode/apps/hive/warehouse/organization.db/employee'
insert the original records back into original table.
hive> insert into organization.employee
select employee_id, employee_name from organization.employee_temp;
finally drop the temp table created
hive> drop table organization.employee_temp;
ALTER TABLE emp REPLACE COLUMNS( name string, dept string);
Above statement can only change the schema of a table, not data.
A solution of this problem to copy data in a new table.
Insert <New Table> Select <selective columns> from <Old Table>
ALTER TABLE is not yet supported for non-native tables; i.e. what you get with CREATE TABLE when a STORED BY clause is specified.
check this https://cwiki.apache.org/confluence/display/Hive/StorageHandlers
After a lot of mistakes, in addition to above explanations, I would add simpler answers.
Case 1: Add new column named new_column
ALTER TABLE schema.table_name
ADD new_column INT COMMENT 'new number column');
Case 2: Rename a column new_column to no_of_days
ALTER TABLE schema.table_name
CHANGE new_column no_of_days INT;
Note that in renaming, both columns should be of same datatype like above as INT
For external table its simple and easy.
Just drop the table schema then edit create table schema , at last again create table with new schema.
example table: aparup_test.tbl_schema_change and will drop column id
steps:-
------------- show create table to fetch schema ------------------
spark.sql("""
show create table aparup_test.tbl_schema_change
""").show(100,False)
o/p:
CREATE EXTERNAL TABLE aparup_test.tbl_schema_change(name STRING, time_details TIMESTAMP, id BIGINT)
ROW FORMAT SERDE 'org.apache.hadoop.hive.ql.io.parquet.serde.ParquetHiveSerDe'
WITH SERDEPROPERTIES (
'serialization.format' = '1'
)
STORED AS
INPUTFORMAT 'org.apache.hadoop.hive.ql.io.parquet.MapredParquetInputFormat'
OUTPUTFORMAT 'org.apache.hadoop.hive.ql.io.parquet.MapredParquetOutputFormat'
LOCATION 'gs://aparup_test/tbl_schema_change'
TBLPROPERTIES (
'parquet.compress' = 'snappy'
)
""")
------------- drop table --------------------------------
spark.sql("""
drop table aparup_test.tbl_schema_change
""").show(100,False)
------------- edit create table schema by dropping column "id"------------------
CREATE EXTERNAL TABLE aparup_test.tbl_schema_change(name STRING, time_details TIMESTAMP)
ROW FORMAT SERDE 'org.apache.hadoop.hive.ql.io.parquet.serde.ParquetHiveSerDe'
WITH SERDEPROPERTIES (
'serialization.format' = '1'
)
STORED AS
INPUTFORMAT 'org.apache.hadoop.hive.ql.io.parquet.MapredParquetInputFormat'
OUTPUTFORMAT 'org.apache.hadoop.hive.ql.io.parquet.MapredParquetOutputFormat'
LOCATION 'gs://aparup_test/tbl_schema_change'
TBLPROPERTIES (
'parquet.compress' = 'snappy'
)
""")
------------- sync up table schema with parquet files ------------------
spark.sql("""
msck repair table aparup_test.tbl_schema_change
""").show(100,False)
==================== DONE =====================================
Even below query is working for me.
Alter table tbl_name drop col_name
Suppose I have a table definition as follows in Hive(the actual table has around 65 columns):
CREATE EXTERNAL TABLE S.TEST (
COL1 STRING,
COL2 STRING
)
PARTITIONED BY (extract_date STRING)
ROW FORMAT DELIMITED FIELDS TERMINATED BY '\007'
LOCATION 'xxx';
Once the table is created, when I run hive -e "describe s.test", I see extract_date as being one of the columns on the table. Doing a select * from s.test also returns extract_date column values. Is it possible to exclude this virtual(?) column when running select queries in Hive.
Change this property
set hive.support.quoted.identifiers=none;
and run the query as
SELECT `(extract_date)?+.+` FROM <table_name>;
I tested it working fine.
I have an external partitioned Hive table. One of its columns is a string named OLDDATE that has the date in a different format(DD-MM-YY). I want to update the column and store dates in YYYY-MM-DD format. All years are 20XX.
So I thought of this
select CONCAT('20',SPLIT(OLDDATE ,'-')[2],'-',SPLIT(OLDDATE ,'-')[1],'-',SPLIT(OLDDATE ,'-')[0]) from table
This gives me the dates in the format I want. Now how do I overwrite the old date with this new date?
You can effect an update by overwriting the table with its own contents, just with the date field changed according to your transformation, like this pseudo-code:
INSERT OVERWRITE table
SELECT
col1
, col2
...
, CONCAT('20',SPLIT(OLDDATE ,'-')[2],'-',SPLIT(OLDDATE ,'-')[1],'-',SPLIT(OLDDATE ,'-')[0]) AS olddate
...
, coln
FROM table;
#user2441441
To overwrite a partitioned table:
INSERT OVERWRITE table PARTITION (p_col)
SELECT
col1
, col2
...
, CONCAT('20',SPLIT(OLDDATE ,'-')[2],'-',SPLIT(OLDDATE ,'-')[1],'-
',SPLIT(OLDDATE ,'-')[0]) AS olddate
...
, coln
, p_col
FROM table;
Since its an partitioned table, the folder names must be created with the date values.
Hence you are not able to update the values.
One work around for this would be create a new table and run your above query and insert data into the new table.
After that you can drop your existing table and treat this new table as your required table.
I have sqoopd data from Netezza table and output file is in HDFS, but one column is a timestamp and I want to load it as a date column in my hive table. Using that column I want to create partition on date. How can i do that?
Example: in HDFS data is like = 2013-07-30 11:08:36
In hive I want to load only date (2013-07-30) not timestamps. I want to partition on that column DAILY.
How can I pass partition by column as dynamically?
I have tried with loading data into one table as source. In final table I will do insert overwrite table partition by (date_column=dynamic date) select * from table1
Set these 2 properties -
SET hive.exec.dynamic.partition=true;
SET hive.exec.dynamic.partition.mode=nonstrict;
And the Query can be like -
INSERT OVERWRITE TABLE TABLE PARTITION (DATE_STR)
SELECT
:
:
-- Partition Col is the last column
to_date(date_column) DATE_STR
FROM table1;
You can explore the two options of hive-import - if it is an incremental import you will be able to get the current day's partition.
--hive-partition-key
--hive-partition-value
You can just load the EMP_HISTORY table from EMP by enabling dynamic partition and converting the timestamp to date using to_date date function
The code might look something like this....
SET hive.exec.dynamic.partition=true;
SET hive.exec.dynamic.partition.mode=nonstrict;
INSERT OVERWRITE TABLE EMP_HISTORY PARTITION (join_date)
SELECT e.name as name, e.age as age, e.salay as salary, e.loc as loc, to_date(e.join_date) as join_date from EMP e ;