HiveRunner

Welcome to HiveRunner - Zero installation open source unit testing of Hive applications

Welcome to the open source project HiveRunner. HiveRunner is a unit test framework based on JUnit4 and enables TDD development of HiveQL without the need of any installed dependencies. All you need is to add HiveRunner to your pom.xml as any other library and you're good to go.

HiveRunner is under constant development. We use it extensively in all our Hive projects. Please feel free to suggest improvements both as Pull requests and as written requests.

A word from the inventors

HiveRunner enables you to write Hive SQL as releasable tested artifacts. It will require you to parametrize and modularize HiveQL in order to make it testable. The bits and pieces of code should then be wired together with some orchestration/workflow/build tool of your choice, to be runnable in your environment (e.g. Oozie, pentaho, Talend, maven, etc…)

So, even though your current Hive SQL probably won't run off the shelf within HiveRunner, we believe the enforced testability and enabling of a TDD workflow will do as much good to the scripting world of SQL as it has for the Java community.

Cook Book

Include HiveRunner

Clone this repos and build with

mvn install

Add the dependency of HiveRunner to your pom file. If you prefer Ivy, then you're currently on your own, but then again, since you're using Ivy, you probably know what to do anyway.

<dependency>
    <groupId>com.klarna</groupId>
    <artifactId>hiverunner</artifactId>
    <version>[HIVERUNNER VERSION]</version>
    <scope>test</scope>
</dependency>

Also explicitly add the surefire plugin and configure forkMode=always to avoid OutOfMemory when building big test suites.

<plugin>
    <groupId>org.apache.maven.plugins</groupId>
    <artifactId>maven-surefire-plugin</artifactId>
    <version>2.17</version>
    <configuration>
        <forkMode>always</forkMode>
    </configuration>
</plugin>

as an alternative if this does not solve the OOM issues, try increase the -Xmx and -XX:MaxPermSize settings. For example:

<plugin>
    <groupId>org.apache.maven.plugins</groupId>
    <artifactId>maven-surefire-plugin</artifactId>
    <version>2.17</version>
    <configuration>
        <forkCount>1</forkCount>
        <reuseForks>false</reuseForks>
        <argLine>-Xmx2048m -XX:MaxPermSize=512m</argLine>
    </configuration>
</plugin>

(please note that the forkMode option is deprecated and you should use forkCount and reuseForks instead)

With forkCount and reuseForks there is a possibility to reduce the test execution time drastically, depending on your hardware. A plugin configuration which are using one fork per CPU core and reuse threads would look like:

<plugin>
    <groupId>org.apache.maven.plugins</groupId>
    <artifactId>maven-surefire-plugin</artifactId>
    <version>2.17</version>
    <configuration>
        <forkCount>1C</forkCount>
        <reuseForks>true</reuseForks>
        <argLine>-Xmx2048m -XX:MaxPermSize=512m</argLine>
    </configuration>
</plugin>

By default, HiveRunner uses mapreduce (mr) as the execution engine for hive. If you wish to run using tez, set the System property hiveconf_hive.execution.engine to 'tez'.

(Any hive conf property may be overridden by prefixing it with 'hiveconf_')

    <plugin>
        <groupId>org.apache.maven.plugins</groupId>
        <artifactId>maven-surefire-plugin</artifactId>
        <version>2.17</version>
        <configuration>
            <systemProperties>
                <hiveconf_hive.execution.engine>tez</hiveconf_hive.execution.engine>
                <hiveconf_hive.exec.counters.pull.interval>1000</hiveconf_hive.exec.counters.pull.interval>
            </systemProperties>
        </configuration>
    </plugin>

Timeout - It's possible to configure HiveRunner to make tests time out after some time and retry those tests a couple of times.. This is to cover for the bug https://issues.apache.org/jira/browse/TEZ-2475 that at times causes test cases to not terminate due to a lost DAG reference. The timeout feature can be configured via the 'enableTimeout', 'timeoutSeconds' and 'timeoutRetries' properties. A configuration which enables timeouts after 30 seconds and allows 2 retries would look like

<plugin>
    <groupId>org.apache.maven.plugins</groupId>
    <artifactId>maven-surefire-plugin</artifactId>
    <version>2.17</version>
    <configuration>
        <systemProperties>
            <enableTimeout>true</enableTimeout>
            <timeoutSeconds>30</timeoutSeconds>
            <timeoutRetries>2</timeoutRetries>
        </systemProperties>
    </configuration>
</plugin>

Logging

src/main/resources/log4j.properties confogures the log levels. Log level is default set to WARN. Some traces remain due to the fact that Hive logs to stdout.

Look at the examples

Look at the com.klarna.hiverunner.HelloHiveRunner reference test case to get a feeling for how a typical test case looks like. If you're put off by the verbosity of the annotations, there's always the possibility to use HiveShell in a more interactive mode. The com.klarna.hiverunner.SerdeTest adds a resources (test data) interactively with HiveShell instead of using annotations.

Annotations and interactive mode can be mixed and matched, however you'll always need to include the com.klarna.hiverunner.annotations.HiveSQL annotation e.g:

     @HiveSQL(files = {"serdeTest/create_table.sql", "serdeTest/hql_custom_serde.sql"}, autoStart = false)
     public HiveShell hiveShell;

Note that the autostart = false is needed for the interactive mode. It can be left out when running with only annotations.

Sequence files

If you work with sequence files (Or anything else than regular text files) make sure to take a look at ResourceOutputStreamTest for an example of how to use the new method HiveShell#getResourceOutputStream to manage test input data.

Understand a little bit of the order of execution

HiveRunner will in default mode setup and start the HiveShell before the test method is invoked. If autostart is set to false, the HiveShell must be started manually from within the test method. Either way, HiveRunner will do the following steps when start is invoked.

Merge any @HiveProperties from the test case with the hive conf
Start the HiveServer with the merged conf
Copy all @HiveResource data into the temp file area for the test
Execute all fields annotated with @HiveSetupScript
Execute the script files given in the @HiveSQL annotation

The HiveShell field annotated with @HiveSQL will always be injected before the test method is invoked.

Hive version compatibility

This version of HiveRunner is built for hive 14.

Future work and Limitations

HiveRunner does not allow the add jar statement. It is considered bad practice to keep environment specific code together with the business logic that targets HiveRunner. Keep environment specific stuff in separate files and use your build/orchestration/workflow tool to run the right files in the right order in the right environment. When running HiveRunner, all SerDes available on the classpath of the IDE/maven will be available.
HiveRunner runs Hive and Hive runs on top of hadoop, and hadoop has limited support for windows machines. Installing Cygwin might help out.
Some of the HiveRunner annotations should probably be rebuilt to be more test method specific. E.g. Resources may be described on a test method basis instead as for a whole test case. Feedback is always welcome!
The interactive API of HiveShell should support as much as the annotated DSL. Currently some features are not implemented in the HiveShell but only available via annotations.

DONE: HiveShell now includes all the features of the annotations and then some!
Clean up the logs

DONE: There's now a working log4j.properties here: src/main/resources/log4j.properties
HiveRunner currently uses in-memory Derby as metastore. It seems to be a real performance bottleneck. We are looking to replace it with hsqldb in the near future.

DONE: This version of HiveRunner uses an in-memory hsqldb instead of derby. It's faster and more reliant!
Currently the HiveServer spins up and tears down for every test method. As a performance option it should be possible to clean the HiveServer and metastore between each test method invocation. The choice should probably be exposed to the test writer. By switching between different strategies, side effects/leakage can be ruled out during test case debugging.
Redirect derby.log. It currently ends up in the build root dir.

DONE: Derby is gone -> derby.log is gone!

Change Log (From version 2.2.0 and onwards)

2.4.0

Enabled any hiveconf variables to be set as System properties by using the naming convention hiveconf_[HiveConf property name]. E.g: hiveconf_hive.execution.engine

Fixed bug: Results sets bigger than 100 rows only returned the first 100 rows.

2.3.0

Merged tez and mr context into the same context again. Now, the same test suite may alter between execution engines by doing E.g:

 hive> set hive.execution.engine=tez;
 hive> [some query]
 hive> set hive.execution.engine=mr;
 hive> [some query]

2.2.0

Added support for setting hivevar:s via HiveShell

Known Issues

IOException in Hive 0.14.0

Described in this issue: HiveRunner#3

This is a known bug in hive. Try setting hive.exec.counters.pull.interval to 1000 millis. It has worked for some projects. You can do this in the surefire plugin:

      <plugin>
          <groupId>org.apache.maven.plugins</groupId>
          <artifactId>maven-surefire-plugin</artifactId>
          <version>2.17</version>
          <configuration>
              <systemProperties>
                  <hiveconf_hive.exec.counters.pull.interval>1000</hiveconf_hive.exec.counters.pull.interval>
              </systemProperties>
          </configuration>
      </plugin>

Also you can try to use the retry functionality in Surefire: https://maven.apache.org/surefire/maven-surefire-plugin/examples/rerun-failing-tests.html

Tez queries do not terminate

Tez will at times forget the process id of a random DAG. This will cause the query to never terminate. To get around this there is a timeout and retry functionality implemented in HiveRunner:

     <plugin>
         <groupId>org.apache.maven.plugins</groupId>
         <artifactId>maven-surefire-plugin</artifactId>
         <version>2.17</version>
         <configuration>
             <systemProperties>
                 <enableTimeout>true</enableTimeout>
                 <timeoutSeconds>30</timeoutSeconds>
                 <timeoutRetries>2</timeoutRetries>
                 </systemProperties>
         </configuration>
     </plugin>

Make sure to set the timeoutSeconds to that of your slowest test in the test suite and then add some padding.

mpeter28 / HiveRunner