Partition Hive table by existing field?

问题

Can I partition a Hive table upon insert by an existing field?

I have a 10 GB file with a date field and an hour of day field. Can I load this file into a table, then insert-overwrite into another partitioned table that uses those fields as a partition? Would something like the following work?

INSERT OVERWRITE TABLE tealeaf_event  PARTITION(dt=evt.datestring,hour=evt.hour) 
SELECT * FROM staging_event evt;

Thanks!

Travis

回答1:

I just ran across this trying to answer the same question and it was helpful but not quite complete. The short answer is yes, something like the query in the question will work but the syntax is not quite right.

Say you have three tables which were created using the following statements:

CREATE TABLE staging_unpartitioned (datestring string, hour int, a int, b int);

CREATE TABLE staging_partitioned (a int, b int) 
    PARTITIONED BY (datestring string, hour int);

CREATE TABLE production_partitioned (a int, b int) 
    PARTITIONED BY (dt string, hour int);

Columns a and b are just some example columns. dt and hour are the values we want to partition on once it gets to the production table. Moving the staging data to production from staging_unpartitioned and staging_partitioned looks exactly the same.

INSERT OVERWRITE TABLE production_partitioned PARTITION (dt, hour)
    SELECT a, b, datestring, hour FROM staging_unpartitioned;

INSERT OVERWRITE TABLE production_partitioned PARTITION (dt, hour)
    SELECT a, b, datestring, hour FROM staging_partitioned;

This uses a process called Dynamic Partitioning which you can read about here. The important thing to note is that which columns are associated with which partitions is determined by the SELECT order. All dynamic partitions must be selected last and in order.

There's a good chance when you try to run the code above you will hit an error due to the properties you have set. First, it will not work if you have dynamic partitioning disabled so make sure to:

set hive.exec.dynamic.partition=true;

Then you might hit an error if you aren't partitioning on at least one static partition before the dynamic partitions. This restriction would save you accidentally removing a root partition when you meant to overwrite its sub-partitions with dynamic partitions. In my experience this behavior has never been helpful and has often been annoying, but your mileage may vary. At any rate, it is easy to change:

set hive.exec.dynamic.partition.mode=nonstrict;

And that should do it.

回答2:

Maybe this is already answered... but yes, you can do exactly as you have stated. I have done it many times. Obviously your new table would need to be defined similar to the original one, but without the partition column, and with the partition specification. Also, I cannot remember if I had to explicitly list out the columns in the original table, or if the asterik was sufficient.

回答3:

I'm not super sure about this, but something like this might work

INSERT OVERWRITE TABLE tealeaf_event
SELECT col1 as tealeaf_col1, ..., datestring as ds;

回答4:

No. You will have to drop that field or, at least, rename it.

来源：https://stackoverflow.com/questions/6631464/partition-hive-table-by-existing-field