Developing Applications With Apache Kudu

Kudu provides C++, Java and Python client APIs, as well as reference examples to illustrate their use.

Use of server-side or private interfaces is not supported, and interfaces which are not part of public APIs have no stability guarantees.

Viewing the API Documentation

C++ API Documentation

You can view the C++ client API documentation online. Alternatively, after building Kudu from source, you can additionally build the doxygen target (e.g., run make doxygen if using make) and use the locally generated API documentation by opening docs/doxygen/client_api/html/index.html file in your favorite Web browser.

In order to build the doxygen target, it’s necessary to have doxygen with Dot (graphviz) support installed at your build machine. If you installed doxygen after building Kudu from source, you will need to run cmake again to pick up the doxygen location and generate appropriate targets.
Java API Documentation

You can view the Java API documentation online. Alternatively, after building the Java client, Java API documentation is available in java/kudu-client/target/apidocs/index.html.

Working Examples

Several example applications are provided in the examples directory of the Apache Kudu git repository. Each example includes a README that shows how to compile and run it. The following list includes some of the examples that are available today. Check the repository itself in case this list goes out of date.


A simple C++ application which connects to a Kudu instance, creates a table, writes data to it, then drops the table.


A simple Java application which connects to a Kudu instance, creates a table, writes data to it, then drops the table.


A small Java application which listens on a TCP socket for time series data corresponding to the Collectl wire protocol. The commonly-available collectl tool can be used to send example data to the server.


A Java application that generates random insert load.


An example program that shows how to use the Kudu Python API to load data into a new / existing Kudu table generated by an external program, dstat in this case.


An example plugin for using graphite-web with Kudu as a backend.

These examples should serve as helpful starting points for your own Kudu applications and integrations.

Maven Artifacts

The following Maven <dependency> element is valid for the Apache Kudu public release (since 1.0.0):


Convenience binary artifacts for the Java client and various Java integrations (e.g. Spark, Flume) are also available via the ASF Maven repository and Maven Central repository.

Example Impala Commands With Kudu

See Using Impala With Kudu for guidance on installing and using Impala with Kudu, including several impala-shell examples.

Kudu Integration with Spark

Kudu integrates with Spark through the Data Source API as of version 1.0.0. Include the kudu-spark dependency using the --packages option:

Use the kudu-spark_2.10 artifact if using Spark with Scala 2.10. Note that Spark 1 is no longer supported in Kudu starting from version 1.6.0. So in order to use Spark 1 integrated with Kudu, version 1.5.0 is the latest to go to.

spark-shell --packages org.apache.kudu:kudu-spark_2.10:1.5.0

Use kudu-spark2_2.11 artifact if using Spark 2 with Scala 2.11.

kudu-spark versions 1.8.0 and below have slightly different syntax. See the documentation of your version for a valid example. Versioned documentation can be found on the releases page.
spark-shell --packages org.apache.kudu:kudu-spark2_2.11:1.9.0

then import kudu-spark and create a dataframe:

import org.apache.kudu.client._
import collection.JavaConverters._

// Read a table from Kudu
val df =
  .options(Map("kudu.master" -> "kudu.master:7051", "kudu.table" -> "kudu_table"))

// Query using the Spark API..."id").filter("id >= 5").show()

// ...or register a temporary table and use SQL
val filteredDF = spark.sql("select id from kudu_table where id >= 5").show()

// Use KuduContext to create, delete, or write to Kudu tables
val kuduContext = new KuduContext("kudu.master:7051", spark.sparkContext)

// Create a new Kudu table from a dataframe schema
// NB: No rows from the dataframe are inserted into the table
    "test_table", df.schema, Seq("key"),
    new CreateTableOptions()
        .addHashPartitions(List("key").asJava, 3))

// Insert data
kuduContext.insertRows(df, "test_table")

// Delete data
kuduContext.deleteRows(filteredDF, "test_table")

// Upsert data
kuduContext.upsertRows(df, "test_table")

// Update data
val alteredDF ="id", $"count" + 1)
kuduContext.updateRows(filteredRows, "test_table")

// Data can also be inserted into the Kudu table using the data source, though the methods on
// KuduContext are preferred
// NB: The default is to upsert rows; to perform standard inserts instead, set operation = insert
// in the options map
// NB: Only mode Append is supported
  .options(Map("kudu.master"-> "kudu.master:7051", "kudu.table"-> "test_table"))

// Check for the existence of a Kudu table

// Delete a Kudu table

Upsert option in Kudu Spark

The upsert operation in kudu-spark supports an extra write option of ignoreNull. If set to true, it will avoid setting existing column values in Kudu table to Null if the corresponding dataframe column values are Null. If unspecified, ignoreNull is false by default.

val dataDF =
  .options(Map("kudu.master" -> "kudu.master:7051", "kudu.table" -> simpleTableName))
// Below is the original data in the table 'simpleTableName'
|  0|foo|

// Upsert a row with existing key 0 and val Null with ignoreNull set to true
val nullDF = spark.createDataFrame(Seq((0, null.asInstanceOf[String]))).toDF("key", "val")
val wo = new KuduWriteOptions
wo.ignoreNull = true
kuduContext.upsertRows(nullDF, simpleTableName, wo)
// The val field stays unchanged
|  0|foo|

// Upsert a row with existing key 0 and val Null with ignoreNull default/set to false
kuduContext.upsertRows(nullDF, simpleTableName)
// Equivalent to:
// val wo = new KuduWriteOptions
// wo.ignoreNull = false
// kuduContext.upsertRows(nullDF, simpleTableName, wo)
// The val field is set to Null this time
|key| val|
|  0|null|

Using Spark with a Secure Kudu Cluster

The Kudu Spark integration is able to operate on secure Kudu clusters which have authentication and encryption enabled, but the submitter of the Spark job must provide the proper credentials. For Spark jobs using the default 'client' deploy mode, the submitting user must have an active Kerberos ticket granted through kinit. For Spark jobs using the 'cluster' deploy mode, a Kerberos principal name and keytab location must be provided through the --principal and --keytab arguments to spark2-submit.

Spark Integration Best Practices

Avoid multiple Kudu clients per cluster.

One common Kudu-Spark coding error is instantiating extra KuduClient objects. In kudu-spark, a KuduClient is owned by the KuduContext. Spark application code should not create another KuduClient connecting to the same cluster. Instead, application code should use the KuduContext to access a KuduClient using KuduContext#syncClient.

To diagnose multiple KuduClient instances in a Spark job, look for signs in the logs of the master being overloaded by many GetTableLocations or GetTabletLocations requests coming from different clients, usually around the same time. This symptom is especially likely in Spark Streaming code, where creating a KuduClient per task will result in periodic waves of master requests from new clients.

Spark Integration Known Issues and Limitations

  • Spark 2.2+ requires Java 8 at runtime even though Kudu Spark 2.x integration is Java 7 compatible. Spark 2.2 is the default dependency version as of Kudu 1.5.0.

  • Kudu tables with a name containing upper case or non-ascii characters must be assigned an alternate name when registered as a temporary table.

  • Kudu tables with a column name containing upper case or non-ascii characters may not be used with SparkSQL. Columns may be renamed in Kudu to work around this issue.

  • <> and OR predicates are not pushed to Kudu, and instead will be evaluated by the Spark task. Only LIKE predicates with a suffix wildcard are pushed to Kudu, meaning that LIKE "FOO%" is pushed down but LIKE "FOO%BAR" isn’t.

  • Kudu does not support every type supported by Spark SQL. For example, Date and complex types are not supported.

  • Kudu tables may only be registered as temporary tables in SparkSQL. Kudu tables may not be queried using HiveContext.

JVM-Based Integration Testing

As of version 1.9.0, Kudu ships with an experimental feature called the binary test JAR. This feature gives people who want to test against Kudu the capability to start a Kudu "mini cluster" from Java or another JVM-based language without having to first build Kudu locally. This is possible because the Kudu binary JAR contains relocatable Kudu binaries that are used by the KuduTestHarness in the kudu-test-utils module. The KuduTestHarness contains logic to search the classpath for the Kudu binaries and to start a mini cluster using them.

Important: The kudu-binary module should only be used to run Kudu for integration testing purposes. It should never be used to run an actual Kudu service, in production or development, because the kudu-binary module includes native security-related dependencies that have been copied from the build system and will not be patched when the operating system on the runtime host is patched.

System Requirements

The binary test JAR must be run on one of the supported Kudu platforms, which includes:

  • macOS El Capitan (10.11) or later;

  • CentOS 6.6+, Ubuntu 14.04+, or another recent distribution of Linux

The related Maven integration using os-maven-plugin requires Maven 3.1 or later.

Using the Kudu Binary Test Jar

Take the following steps to start a Kudu mini cluster from a Java project.

1. Add build-time dependencies. The kudu-binary artifact contains the native Kudu (server and command-line tool) binaries for specific operating systems. In order to download the right artifact for the running operating system, use the os-maven-plugin to detect the current runtime environment. Finally, the kudu-test-utils module provides the KuduTestHarness class, which runs a Kudu mini cluster.

Maven example for Kudu 1.9.0:

    <!-- Used to find the right kudu-binary artifact with the Maven
         property ${os.detected.classifier} -->


2. Write a test that starts a Kudu mini cluster using the KuduTestHarness. It will automatically find the binary test JAR if Maven is configured correctly.

The recommended way to start a Kudu mini cluster is by using the KuduTestHarness class from the kudu-test-utils module, which also acts as a JUnit Rule. Here is an example of a Java-based integration test that starts a Kudu cluster, creates a Kudu table on the cluster, and then exits:

import org.apache.kudu.ColumnSchema;
import org.apache.kudu.Schema;
import org.apache.kudu.Type;
import org.apache.kudu.test.KuduTestHarness;
import org.junit.Rule;
import org.junit.Test;

import java.util.ArrayList;
import java.util.Collections;
import java.util.List;

public class MyKuduTest {
  // The KuduTestHarness automatically starts and stops a real Kudu cluster
  // when each test is run. Kudu persists its on-disk state in a temporary
  // directory under a location defined by the environment variable TEST_TMPDIR
  // if set, or under /tmp otherwise. That cluster data is deleted on
  // successful exit of the test. The cluster output is logged through slf4j.
  public KuduTestHarness harness = new KuduTestHarness();

  public void test() throws Exception {
    // Get a KuduClient configured to talk to the running mini cluster.
    KuduClient client = harness.getClient();

    // Create a new Kudu table.
    List<ColumnSchema> columns = new ArrayList<>();
        new ColumnSchema.ColumnSchemaBuilder(
            "key", Type.INT32).key(true).build());
    Schema schema = new Schema(columns);
    CreateTableOptions opts =
        new CreateTableOptions().setRangePartitionColumns(
    client.createTable("table-1", schema, opts);

    // Now we may insert rows into the newly-created Kudu table using 'client',
    // scan the table, etc.

For more examples of using the KuduTestHarness, including how to pass configuration options to the Kudu cluster being managed by the harness, see the java-example project in the Kudu source code repository, or look at the various Kudu integration tests under java in the Kudu source code repository.

Kudu Python Client

The Kudu Python client provides a Python friendly interface to the C++ client API. The sample below demonstrates the use of part of the Python client.

import kudu
from kudu.client import Partitioning
from datetime import datetime

# Connect to Kudu master server
client = kudu.connect(host='kudu.master', port=7051)

# Define a schema for a new table
builder = kudu.schema_builder()
builder.add_column('ts_val', type_=kudu.unixtime_micros, nullable=False, compression='lz4')
schema =

# Define partitioning schema
partitioning = Partitioning().add_hash_partitions(column_names=['key'], num_buckets=3)

# Create new table
client.create_table('python-example', schema, partitioning)

# Open a table
table = client.table('python-example')

# Create a new session so that we can apply write operations
session = client.new_session()

# Insert a row
op = table.new_insert({'key': 1, 'ts_val': datetime.utcnow()})

# Upsert a row
op = table.new_upsert({'key': 2, 'ts_val': "2016-01-01T00:00:00.000000"})

# Updating a row
op = table.new_update({'key': 1, 'ts_val': ("2017-01-01", "%Y-%m-%d")})

# Delete a row
op = table.new_delete({'key': 2})

# Flush write operations, if failures occur, capture print them.
except kudu.KuduBadStatus as e:

# Create a scanner and add a predicate
scanner = table.scanner()
scanner.add_predicate(table['ts_val'] == datetime(2017, 1, 1))

# Open Scanner and read all tuples
# Note: This doesn't scale for large scans
result =

Integration with MapReduce, YARN, and Other Frameworks

Kudu was designed to integrate with MapReduce, YARN, Spark, and other frameworks in the Hadoop ecosystem. See and for examples which you can model your own integrations on. Stay tuned for more examples using YARN and Spark in the future.