How to find the most recent partition in HIVE table

17
votes

I have a partitioned table - with 201 partitions. I need to find latest partition in this table and use it to post process my data. The query to find list of all partitions is :

use db;
show partitions table_name;

I need a query to find the latest of these partitions. The partitions are in format 

ingest_date=2016-03-09

I tried using max() which gave me a wrong result. I do not want to traverse through entire table by doing 

select max(ingest_date) from db.table_name;

This would give me the expected output.. but kill the whole point of having partitions in the 1st place.

Is there a more efficient query to get the latest partition for HIve table ?

4
hadoopapache-sparkhive
Possible duplicate of Hive: how to show all partitions of a table? - zero323
Alas, Hive does not expose the metadata as virtual tables - nothing like a SYS or INFORMATION_SCHEMA database. The only efficient way to list the partitions of a Hive table is to use the Metastore java API. Or to reverse-engineer the way partitions are organized in HDFS (quite straightforward actually) then list the directories, hoping HDFS and the Metastore are 100% in sync. - Samson Scharfrichter
Or, if you can query the Metastore database directly (probably in MySQL or PostgreSQL), reverse-engineer the data model and run any SQL you like. - Samson Scharfrichter

4 Answers

10
votes

You can use "show partitions":

hive -e "set hive.cli.print.header=false;show partitions table_name;" | tail -1 | cut -d'=' -f2

This will give you "2016-03-09" as output.

8
votes

If you want to avoid running the "show partitions" in hive shell as suggested above, you can apply a filter to your max() query. That will avoid doing a fulltable scan and results should be fairly quick!

select max(ingest_date) from db.table_name where ingest_date>date_add(current_date,-3) will only scan 2-3 partitions.

2
votes

if you know your table location in hdfs. This is the most quick way without even opening the hive shell.

You can check you table location in hdfs using command;

show create table <table_name>

then

hdfs dfs -ls <table_path>| sort -k6,7 | tail -1

It will show latest partition location in hdfs

0
votes

It looks like there is no way to query for the last partition via Hive (or beeline) CLI that checks only metadata (as one should expect).

For the sake of completeness, the alternative I would propose to the bash parsing answer is the one directly querying the metastore, which can be easily extended to more complex functions of the ingest_date rather than just taking the max. For instance, for a MySQL metastore I've used:

SELECT MAX(PARTITIONS.PART_NAME) FROM
DBS
INNER JOIN
TBLS ON DBS.DB_ID = TBLS.DB_ID
INNER JOIN
PARTITIONS ON TBLS.TBL_ID = PARTITIONS.TBL_ID
PARTITIONS DBS.NAME = 'db'
PARTITIONS TBLS.TBL_NAME = 'my_table'

Then the output will be in format partition_name=partition_value.