Hive Developer FAQ
Hive is using Maven as its build tool. Versions prior to 0.13 were using Ant.
How do I move some files?
Post a patch for testing purposes which simply does add and deletes. SVN will not understand these patches are actually moves, therefore you should actually upload the following, in order so the last upload is the patch for testing purposes:
- A patch which has only the non-move changes for commit e.g. HIVE-XXX-for-commit.patch
- A script of of commands required to make the moves HIVE-XXX-moves.sh
- A patch for testing purposes HIVE-XXX.patch
The script should be a set of
svn mv commands along with any
perl commands required for find/replace. For example:
How do I add a new MiniDriver test?
See MiniDriver Tests for information about MiniDriver and Beeline tests.
For FAQ about how to run tests, see Testing below.
- See Getting Started: Building Hive from Source for detailed information about building Hive releases 0.13 and later with Maven.
- See Installing from Source Code (Hive 0.12.0 and Earlier) for detailed information about building Hive 0.12 and earlier with Ant.
You might have to set the following Maven options on certain systems to get build working: Set MAVEN_OPTS to "-Xmx2g -XX:MaxPermSize=256M".
How to build all source?
The way Maven is set up differs between the master branch and branch-1. In branch-1, since both Hadoop 1.x and 2.x are supported, you need to specify whether you want to build Hive against Hadoop 1.x or 2.x. This is done via Maven profiles. There is a profile for each version of Hadoop,
hadoop-2. For most Maven operations one of these profiles needs to be specified or the build will fail.
In master, only Hadoop 2.x is supported, thus there is no need to specify a Maven profile for most build operations.
In master, MVN:
In branch-1, MVN:
To build against Hadoop 1.x, switch the above to
For the remainder of this page we will assume master and not show the profiles. However, if you are working on branch-1 remember that you will need to add in the appropriate profile.
How do I import into Eclipse?
Build and generate Eclipse files (the conservative method).
For master – without using a local Maven repository:
NOTE: The presence of the package target is crucial in successfully generating the project files.
NOTE: It is recommended to wipe out or rename the org/apache/hive subtree of the local m2 repository prior to generating the project files; and you may want to check after its execution that it does not contain any files – if it does contain something, then your project files may not be fully independent of downloaded Maven artifacts.
This will work on all branches; but keep in mind that in this case you are installing the Hive artifacts into your local repository.
In Eclipse define M2_REPO in Preferences -> Java -> Build Path -> Classpath Variables to either:
Then import the workspaces. If you get an error about "restricted use of Signal" for Beeline and CLI, follow these instructions.
Note that if you use the Hive git base directory as the Eclipse workspace, then it does not pick the right project names (for example, picks 'ant' instead of 'hive-ant'). Therefore it's recommended to have the workspace directory one up from the git directory. For example workspaces/hive-workspace/hive where hive-workspace is the Eclipse workspace and hive is the git base directory.
How to generate tarball?
It will then be located in the packaging/target/ directory.
How to generate protobuf code?
How to generate Thrift code?
How to run findbugs after a change?
Note: Available in Hive 1.1.0 onward (see HIVE-8327).
How to compile ODBC?
How do I publish Hive artifacts to my local Maven repository?
For general information, see Unit Tests and Debugging in the Developer Guide.
How do I run precommit tests on a patch?
Hive precommit testing is triggered automatically when a file is uploaded to the JIRA ticket:
- Attach the patch file to a JIRA ticket: in the ticket's "More" tab, select "Attach Files" and use "Choose File" to upload the file, then add a descriptive comment.
- Put the patch in the review queue: click the "Submit Patch" button. The button name will change to "Cancel Patch" and the ticket status will change to Patch Available.
See Hive PreCommit Patch Testing for more detail.
How do I rerun precommit tests over the same patch?
For patch updates, our convention is to number them like HIVE-1856.1.patch, HIVE-1856.2.patch, etc. And then click the "Submit Patch" button again when a new one is uploaded; this makes sure it gets back into the review queue.
How do I run a single test?
Note that any test in the itests directory needs to be executed from within the itests directory. The pom is disconnected from the parent project for technical reasons.
Single test class:
Single test method:
Note that a pattern can also be supplied to -Dtests to run multiple tests matching the pattern:
For more usage see the documentation for the Maven Surefire Plugin.
How do I run all of the unit tests?
Note that you need to have previously built and installed the jars:
Legacy information for the Ant build
Make sure that your JAVA_HOME is appropriately set (some tests need this), and set ANT_OPTS to increase the size allocated to the Permanent Generation as per the following:
Then, for a clean build, run
Note that running
ant test will not work;
ant package does some setup work that is required for the testcases to run successfully.
How do I run all of the unit tests except for a certain few tests?
Similar to running all tests, but define test.excludes.additional to specify a test/pattern to exclude from the test run. For example the following will run all tests except for the CliDriver tests:
How do I run the clientpositive/clientnegative unit tests?
All of the below require that you have previously run
To run clientpositive tests
To run a single clientnegative test alter1.q
To run all of the clientpositive tests that match a regex, for example the partition_wise_fileformat tests
To run a single contrib test alter1.q and overwrite the result file
To run a single test groupby1.q and output detailed information during execution
As of Hive 0.11.0+ you can cut down the total build time by specifying that only the ql module needs to rebuild. For example, run all the partition_wise_fileformat tests
How do I modify the init script when testing?
The option to skip the init script or supply a custom init script was added in Hive 2.0 (see HIVE-11538).
To skip initialization:
To supply a custom script:
How do I update the output of a CliDriver testcase?
As of Hive 0.11.0+ you can cut this time in half by specifying that only the ql module needs to rebuild
How do I update the results of many test cases?
Assume that you have a file like below which you'd like to re-generate output files for. Such a file could be created by copying the output from the precommit tests.
You can re-generate all those output files in batches of 20 with the command below
Where is the log output of a test?
Logs are put in a couple locations:
From the root of the source tree: find . -name hive.log
/tmp/$USER/ (Linux) or $TMPDIR/$USER/ (MacOS)
See Hive Logging for details about log files, including alternative configurations.
How do I add a test case?
First, add the test case to the qfile test suite:
- Copy the test to a new file under
/clientnegativeif it is a negative test).
- Add the
itests/src/test/resources/testconfiguration.propertiesto the appropriate variable (ex.
Next, generate the golden (output) file the first time you run the test case:
Compile the Hive source from the top level Hive directory:
Compile the itests:
Run the test and generate the output file using the appropriate
itests/qtest/pom.xmlfor the names of other test suites):
- Add your output file to
/clientnegativeif it is a negative test).
With the above steps, you can create a patch which has a
.java file, a
.q file and a
Why isn't the itests pom connected to the root pom?
It would be great to have it connected, but it would make it harder to use mvn test locally. The best option would be to utilize the failsafe plugin for integration testing; but it needs a bit different setup, and it's harder to use for now.... If you'd like to give that a try, by all means, go ahead.
There is an option to attach all the itest subprojects to the main project by enabling this with -Pitests (HIVE-13490).
There are some good and bad sides using this, it's introduced as a profile to clearly communicate that the integration tests are attached to the main project.
The good side is that you may freely use -Pitests to run integration tests from the root project without the need of mvn install.
The bad side is that a simple mvn test -Pitests will start executing all integration tests.
Why does a test fail with a NullPointerException in MiniDFSCluster?
If any test fails with the error below it means you have an inappropriate umask setting. It should be set to 0022.
Why do Spark unit tests fail with a SecurityException?
If you get the following errors in the unit tests:
2016-01-07T09:51:49,365 ERROR [Driver]: spark.SparkContext (Logging.scala:logError(96)) - Error initializing SparkContext. java.lang.SecurityException: class "javax.servlet.FilterRegistration"'s signer information does not match signer information of other classes in the same package
It happens because there are two conflicting versions of the same classes "javax.servlet:servlet-api" and "org.eclipse.jetty.orbit:javax-servlet". Spark requires the eclipse version, but most tools including Hadoop and Jetty depend on the javax one. To avoid this problem, we need to exclude the javax version everywhere it comes up. Fortunately, maven has a tool to do that with:
mvn dependency:tree -Dverbose
which prints out the dependency tree. Go to each directory with the failing unit tests and search the dependency tree for "javax.servlet:servlet-api" and use exclusions in the pom.xml to remove it. See HIVE-12783 for an example.
How do I debug into a single test in Eclipse?
You can debug into a single JUnit test in Eclipse by first making sure you've built the Eclipse files and imported the project into Eclipse as described here. Then set one or more breakpoints, highlight the method name of the JUnit test method you want to debug into, and do
Another useful method to debug these tests is to attach a remote debugger. When you run the test, enable the debug mode for surefire by passing in "
-Dmaven.surefire.debug". Additional details on how to turning on debugging for surefire can be found here. Now when you run the tests, it will wait with a message similar to
Note that this assumes that you are still using the default port 5005 for surefire. Otherwise you might see a different port. Once you see this message, in Eclipse right click on the project you want to debug, go to "Debug As -> Debug Configurations -> Remote Java Application" and hit the "+" sign on far left top. This should bring up a dialog box. Make sure that the host is "localhost" and the port is "5005". Before you start debugging, make sure that you have set appropriate debug breakpoints in the code. Once ready, hit "Debug". Now if you go back to the terminal, you should see the tests running and they will stop at the breakpoints that you set for debugging.
How do I debug my queries in Hive?
You can also interactively debug your queries in Hive by attaching a remote debugger. To do so, start Beeline with the "
Once you see this message, in Eclipse right click on the project you want to debug, go to "Debug As -> Debug Configurations -> Remote Java Application" and hit the "+" sign on far left top. This should bring up a dialog box. Make sure that the host is the host on which the Beeline CLI is running and the port is "8000". Before you start debugging, make sure that you have set appropriate debug breakpoints in the code. Once ready, hit "Debug". The remote debugger should attach to Beeline and proceed.
At this point, run your queries as normal and it should stop at the breakpoints that you set so that you can start debugging.
This method should work great if your queries are simple fetch queries that do not kick off MapReduce jobs. If a query runs in a distributed mode, it becomes very hard to debug. Therefore, it is advisable to run in a "local" mode for debugging. In order to run Hive in local mode, do the following:
At this point, attach the remote debugger as mentioned before to start debugging your queries.