Apache Hive is a data warehouse system for Hadoop. Hive enables data summarization, querying, and analysis of data. Hive queries are written in HiveQL, which is a query language similar to SQL.
Hive allows you to project structure on largely unstructured data. After you define the structure, you can use HiveQL to query the data without knowledge of Java or MapReduce.
HDInsight provides several cluster types, which are tuned for specific workloads. The following cluster types are most often used for Hive queries:
Interactive Hive: A Hadoop cluster that provides Low Latency Analytical Processing (LLAP) functionality to improve response times for interactive queries. For more information, see the Start with Interactive Hive in HDInsight document.
Hadoop: A Hadoop cluster that is tuned for batch processing workloads. For more information, see the Start with Hadoop in HDInsight document.
Spark: Apache Spark has built-in functionality for working with Hive. For more information, see the Start with Spark on HDInsight document.
HBase: HiveQL can be used to query data stored in HBase. For more information, see the Start with HBase on HDInsight document.
How to use Hive
Use the following table to discover how to use Hive with HDInsight:
|Use this method if you want...||...an interactive shell||...batch processing||...with this cluster operating system||...from this client operating system|
|Hive View||✔||✔||Linux||Any (browser based)|
|Beeline client||✔||✔||Linux||Linux, Unix, Mac OS X, or Windows|
|REST API||✔||Linux or Windows*||Linux, Unix, Mac OS X, or Windows|
|HDInsight tools for Visual Studio||✔||Linux or Windows*||Windows|
|Windows PowerShell||✔||Linux or Windows*||Windows|
* Linux is the only operating system used on HDInsight version 3.4 or greater. For more information, see HDInsight retirement on Windows.
HiveQL language reference
HiveQL language reference is available in the language manual (https://cwiki.apache.org/confluence/display/Hive/LanguageManual).
Hive and data structure
Hive understands how to work with structured and semi-structured data. For example, text files where the fields are delimited by specific characters. The following HiveQL statement creates a table over space-delimited data:
CREATE EXTERNAL TABLE log4jLogs (t1 string, t2 string, t3 string, t4 string, t5 string, t6 string, t7 string) ROW FORMAT DELIMITED FIELDS TERMINATED BY ' ' STORED AS TEXTFILE LOCATION '/example/data/';
Hive also supports custom serializer/deserializers (SerDe) for complex or irregularly structured data. For more information, see the How to use a custom JSON SerDe with HDInsight document.
For more information on file formats supported by Hive, see the Language manual (https://cwiki.apache.org/confluence/display/Hive/LanguageManual)
Hive internal tables vs external tables
There are two types of tables that you can create with Hive:
Internal: Data is stored in the Hive data warehouse. The data warehouse is located at
/hive/warehouse/on the default storage for the cluster.
Use internal tables when:
- Data is temporary.
- You want Hive to manage the lifecycle of the table and data.
External: Data is stored outside the data warehouse. The data can be stored on any storage accessible by the cluster.
Use external tables when:
- The data is also used outside of Hive. For example, the data files are updated by another process (that does not lock the files.)
- Data needs to remain in the underlying location, even after dropping the table.
- You need a custom location, such as a non-default storage account.
- A program other than hive manages the data format, location, etc.
For more information, see the Hive Internal and External Tables Intro blog post.
User-defined functions (UDF)
Hive can also be extended through user-defined functions (UDF). A UDF allows you to implement functionality or logic that isn't easily modeled in HiveQL. For an example of using UDFs with Hive, see the following documents:
Hive on HDInsight comes pre-loaded with an internal table named
hivesampletable. HDInsight also provides example data sets that can be used with Hive. These data sets are stored in the
/HdiSamples directories. These directories exist in the default storage for your cluster.
The following HiveQL statements project columns onto the
set hive.execution.engine=tez; DROP TABLE log4jLogs; CREATE EXTERNAL TABLE log4jLogs (t1 string, t2 string, t3 string, t4 string, t5 string, t6 string, t7 string) ROW FORMAT DELIMITED FIELDS TERMINATED BY ' ' STORED AS TEXTFILE LOCATION '/example/data/'; SELECT t4 AS sev, COUNT(*) AS count FROM log4jLogs WHERE t4 = '[ERROR]' AND INPUT__FILE__NAME LIKE '%.log' GROUP BY t4;
In the previous example, the HiveQL statements perform the following actions:
set hive.execution.engine=tez;: Sets the execution engine to use Tez. Using Tez instead of MapReduce can provide an increase in query performance. For more information on Tez, see the Use Apache Tez for improved performance section.
This statement is only required when using a Windows-based HDInsight cluster. Tez is the default execution engine for Linux-based HDInsight.
DROP TABLE: If the table already exists, delete it.
CREATE EXTERNAL TABLE: Creates a new external table in Hive. External tables only store the table definition in Hive. The data is left in the original location and in the original format.
ROW FORMAT: Tells Hive how the data is formatted. In this case, the fields in each log are separated by a space.
STORED AS TEXTFILE LOCATION: Tells Hive where the data is stored (the
example/datadirectory) and that it is stored as text. The data can be in one file or spread across multiple files within the directory.
SELECT: Selects a count of all rows where the column t4 contains the value [ERROR]. This statement returns a value of 3 because there are three rows that contain this value.
INPUT__FILE__NAME LIKE '%.log'- Hive attempts to apply the schema to all files in the directory. In this case, the directory contains files that do not match the schema. To prevent garbage data in the results, this statement tells Hive that we should only return data from files ending in .log.
External tables should be used when you expect the underlying data to be updated by an external source. For example, an automated data upload process, or MapReduce operation.
Dropping an external table does not delete the data, it only deletes the table definition.
To create an internal table instead of external, use the following HiveQL:
set hive.execution.engine=tez; CREATE TABLE IF NOT EXISTS errorLogs (t1 string, t2 string, t3 string, t4 string, t5 string, t6 string, t7 string) STORED AS ORC; INSERT OVERWRITE TABLE errorLogs SELECT t1, t2, t3, t4, t5, t6, t7 FROM log4jLogs WHERE t4 = '[ERROR]';
These statements perform the following actions:
CREATE TABLE IF NOT EXISTS: If the table does not exist, create it. Because the EXTERNAL keyword is not used, this statement creates an internal table. The table is stored in the Hive data warehouse and is managed completely by Hive.
STORED AS ORC: Stores the data in Optimized Row Columnar (ORC) format. ORC is a highly optimized and efficient format for storing Hive data.
INSERT OVERWRITE ... SELECT: Selects rows from the log4jLogs table that contains [ERROR], and then inserts the data into the errorLogs table.
Unlike external tables, dropping an internal table also deletes the underlying data.
Improve Hive query performance
Apache Tez is a framework that allows data intensive applications, such as Hive, to run much more efficiently at scale. Tez is enabled by default for Linux-based HDInsight clusters.
Tez is currently off by default for Windows-based HDInsight clusters and it must be enabled. To take advantage of Tez, the following value must be set for a Hive query:
Tez is the default engine for Linux-based HDInsight clusters.
The Hive on Tez design documents contains details about the implementation choices and tuning configurations.
To aid in debugging jobs ran using Tez, HDInsight provides the following web UIs that allow you to view details of Tez jobs:
Low Latency Analytical Processing (LLAP)
LLAP (sometimes known as Live Long and Process) is a new feature in Hive 2.0 that allows in-memory caching of queries. LLAP makes Hive queries much faster, up to 26x faster than Hive 1.x in some cases.
HDInsight provides LLAP in the Interactive Hive cluster type. For more information, see the Start with Interactive Hive document.
Hive jobs and SQL Server Integration Services
You can use SQL Server Integration Services (SSIS) to run a Hive job. The Azure Feature Pack for SSIS provides the following components that work with Hive jobs on HDInsight.
Learn more about the Azure Feature Pack for SSIS here.
Now that you've learned what Hive is and how to use it with Hadoop in HDInsight, use the following links to explore other ways to work with Azure HDInsight.