Five Hard-Won Lessons Using Hive

EDIT, 9/8/2016: Hive has come a long way in the two years since I’ve written this. While some of the code snippets might still work, it’s likely the case that this information is so out-of-date to be nothing more than a reflection of working with Hadoop in 2014.

I’ve been spending a ton of time lately on the data engineering side of ‘data science’, so I’ve been writing a lot of Hive queries. Hive is a great tool for querying large amounts of data, without having to know very much about the underpinnings of Hadoop. Unfortunately, there are a lot of things about Hive (version 0.12 and before) that aren’t quite the same as SQL and have caused me a bunch of frustration; here they are, in no particular order.

1. Set Hive Temp directory To Same As Final Output Directory

When doing a “Create Table As” (CTAS) statement in Hive, Hive allocates temp space for the Map and Reduce portions of the job. If you’re not lucky, the temp space for the job will be somewhere different than where your table actually ends up being saved, resulting in TWO I/O operations instead of just one. This can lead to a painful delay in when your Hive job says it is finished vs. when the table becomes available (one time, I saw a 30 hour delay writing 5TB of data).

If your Hive jobs seem to hang after the Job Tracker says they are complete, try this setting at the beginning of your session:

set hive.optimize.insert.dest.volume=true;

2. Column Aliasing In Group By/Order By

Not sure why this isn’t a default, but if you want to be able to reference your column names by position (i.e. group by 1,2) instead of by name (i.e. group by name, age), then run this at the beginning of your session:

set hive.groupby.orderby.position.alias=true;

3. Be Aware Of Predicate Push-Down Rules

In Hive, you can get great performance gains if you A) partition your table by commonly used columns/business concepts (i.e. Day, State, Market, etc.) and B) you use the partitions in a WHERE clause. These are known as partition-based queries. Otherwise, if you don’t use a partition in your WHERE clause, you will get a full table scan.

Unfortunately, when doing an OUTER JOIN, Hive will sometimes ignore the fact that your WHERE clause is on a partition and do a full table scan anyway. In order to get Hive to push your predicate down and avoid a full table scan, put your predicate on the JOIN instead of the WHERE clause:

--#### Assume sales Hive table partitioned by day_id ####--

--Full Table Scan
from employees
left join sales on ( = sales.employee_id)
where day_id between '2014-03-01' and '2014-05-31';

--Partitioned-based query
from employees
left join sales on ( = sales.employee_id and sales.day_id between '2014-03-01' and '2014-05-31');

If you don’t want to think about the different rules, you can generally put your limiting clauses inside your JOIN clause instead of on your WHERE clause. It should just be a matter of preference (until your query performance indicates it isn’t!)

4. Calculate And Append Percentiles Using CROSS JOIN

Suppose you want to calculate the top 10% of your customers by sales. If you try to do the following, Hive will complain about needing a GROUP BY, because percentile_approx() is a summary function:

--Hive expects that you want to calculate your percentiles by account_number and sales
--This code will generate an error about a missing GROUP BY statement
CASE WHEN sales > percentile_approx(sales, .9) THEN 1 ELSE 0 END as top10pct_sales

To get around the the need for a GROUP BY, we can use a CROSS JOIN. A CROSS JOIN is another name for a Cartesian Join, meaning all of the rows from the first table will be joined to ALL of the rows of the second table. Because the subquery only returns one row, the CROSS JOIN provides the desired affect of joining the percentile values back to the original table while keeping the same number of rows from the original table. Generally, you don’t want to do a CROSS JOIN (because relational data generally is joined on a key), but this is a good use case.

5.  Calculating a Histogram

Creating a histogram using Hive should be as simple as calling the histogram_numeric() function. However, the syntax and results of this function are just plain weird. To create a histogram, you can run the following:

histogram_numeric(salary, 20) as salary_hist


The results of this query comes back as a list, which is very un-SQL like! To get the data as a table, we can use LATERAL VIEW and EXPLODE:

   CAST(hist.x as int) as bin_center,
   CAST(hist.y as bigint) as bin_height
FROM (select
      histogram_numeric(salary, 20) as salary_hist
      sample_08) a
LATERAL VIEW explode(salary_hist) exploded_table as hist;

	bin_center	bin_height
0	23507	255
1	31881	340
2	39824	287
3	47615	181
4	55667	164
5	59952	8
6	66034	137
7	75642	82
8	82496	44
9	91431	60
10	100665	21
11	107326	15
12	121248	16
13	142070	2
14	153896	6
15	162310	6
16	169810	2
17	176740	2
18	193925	8
19	206770	2

However, now that we have a table of data, it’s still not clear how to create a histogram, as the center of variable-width bins is what is returned by Hive. The Hive documentation for histogram_numeric() references Gnuplot, Excel, Mathematica and MATLAB, which I can only assume can deal with plotting the centers?  Eventually I’ll figure out how to deal with this using R or Python, but for now, I just use the table as a quick gauge of what the data looks like.

  • Self-Service Adobe Analytics Data Feeds!
  • RSiteCatalyst Version 1.4.10 Release Notes
  • WordPress to Jekyll: A 30x Speedup
  • Bulk Downloading Adobe Analytics Data
  • Adobe Analytics Clickstream Data Feed: Calculations and Outlier Analysis
  • Adobe: Give Credit. You DID NOT Write RSiteCatalyst.
  • RSiteCatalyst Version 1.4.8 Release Notes
  • Adobe Analytics Clickstream Data Feed: Loading To Relational Database
  • Calling RSiteCatalyst From Python
  • RSiteCatalyst Version 1.4.7 (and 1.4.6.) Release Notes
  • RSiteCatalyst Version 1.4.5 Release Notes
  • Getting Started: Adobe Analytics Clickstream Data Feed
  • RSiteCatalyst Version 1.4.4 Release Notes
  • RSiteCatalyst Version 1.4.3 Release Notes
  • RSiteCatalyst Version 1.4.2 Release Notes
  • Destroy Your Data Using Excel With This One Weird Trick!
  • RSiteCatalyst Version 1.4.1 Release Notes
  • Visualizing Website Pathing With Sankey Charts
  • Visualizing Website Structure With Network Graphs
  • RSiteCatalyst Version 1.4 Release Notes
  • Maybe I Don't Really Know R After All
  • Building JSON in R: Three Methods
  • Real-time Reporting with the Adobe Analytics API
  • RSiteCatalyst Version 1.3 Release Notes
  • Adobe Analytics Implementation Documentation in 60 Seconds
  • RSiteCatalyst Version 1.2 Release Notes
  • Clustering Search Keywords Using K-Means Clustering
  • RSiteCatalyst Version 1.1 Release Notes
  • Anomaly Detection Using The Adobe Analytics API
  • (not provided): Using R and the Google Analytics API
  • My Top 20 Least Useful Omniture Reports
  • For Maximum User Understanding, Customize the SiteCatalyst Menu
  • Effect Of Modified Bounce Rate In Google Analytics
  • Adobe Discover 3: First Impressions
  • Using Omniture SiteCatalyst Target Report To Calculate YOY growth
  • Google Analytics Individual Qualification (IQ) - Passed!
  • Google Analytics SEO reports: Not Ready For Primetime?
  • An Afternoon With Edward Tufte
  • Google Analytics Custom Variables: A Page-Level Example
  • Xchange 2011: Think Tank and Harbor Cruise
  • Google Analytics for WordPress: Two Methods
  • WordPress Stats or Google Analytics? Yes!
  • Building a Data Science Workstation (2017)
  • JuliaCon 2015: Everyday Analytics and Visualization (video)
  • Vega.jl, Rebooted
  • Sessionizing Log Data Using data.table [Follow-up #2]
  • Sessionizing Log Data Using dplyr [Follow-up]
  • Sessionizing Log Data Using SQL
  • Review: Data Science at the Command Line
  • Introducing Twitter.jl
  • Code Refactoring Using Metaprogramming
  • Evaluating BreakoutDetection
  • Creating A Stacked Bar Chart in Seaborn
  • Visualizing Analytics Languages With VennEuler.jl
  • String Interpolation for Fun and Profit
  • Using Julia As A "Glue" Language
  • Five Hard-Won Lessons Using Hive
  • Using SQL Workbench with Apache Hive
  • Getting Started With Hadoop, Final: Analysis Using Hive & Pig
  • Quickly Create Dummy Variables in a Data Frame
  • Using Amazon EC2 with IPython Notebook
  • Adding Line Numbers in IPython/Jupyter Notebooks
  • Fun With Just-In-Time Compiling: Julia, Python, R and pqR
  • Getting Started Using Hadoop, Part 4: Creating Tables With Hive
  • Tabular Data I/O in Julia
  • Hadoop Streaming with Amazon Elastic MapReduce, Python and mrjob
  • A Beginner's Look at Julia
  • Getting Started Using Hadoop, Part 3: Loading Data
  • Innovation Will Never Be At The Push Of A Button
  • Getting Started Using Hadoop, Part 2: Building a Cluster
  • Getting Started Using Hadoop, Part 1: Intro
  • Instructions for Installing & Using R on Amazon EC2
  • Video: SQL Queries in R using sqldf
  • Video: Overlay Histogram in R (Normal, Density, Another Series)
  • Video: R, RStudio, Rcmdr & rattle
  • Getting Started Using R, Part 2: Rcmdr
  • Getting Started Using R, Part 1: RStudio
  • Learning R Has Really Made Me Appreciate SAS