Customizing Pinot

There are a lot of places in Pinot which can be customized depending on the infrastructure or the use case. Below is a list of such customization points.


1. Generating Pinot segments

Typically, data files will be available on some offline data storage, such as HDFS, and a Hadoop job can be written to read the data and create the segment. The SegmentCreationJob class contains a hadoop job for creating segments. This is a map only job, and the mapper can be found in SegmentCreationMapper. You can override the SegmentCreationMapper with a custom mapper by overriding the SegmentCreationJob::getMapperClass() method. You can read more details about creating segments on Hadoop in Creating Pinot segments

Depending on how frequently new data is generated, you can schedule your jobs to run periodically using either cron or a scheduler such as Azkaban.

2. Pluggable storage

We expect the storage to be shared across controllers of the same cluster, such as NFS. You can write your own implementation of PinotFS to store segments in a data layer of your choice, for example Azure or S3. Please refer to the Pluggable Storage doc for more details.

3. Pluggable streams

We provide out of the box support for consumption from Kafka stream. You can write your own plugin in order to consume from another pub-sub stream such as Azure EventHubs or Amazon Kinesis. Refer to the Pluggable Streams doc for more details.

4. Encrypting segments

The PinotCrypter will encrypt and decrypt segments when they are uploaded to the controller or downloaded by the server. This class is especially useful in cases where segments cannot be stored unencrypted in storage. By default, we use NoOpPinotCrypter crypter which will not encrypt anything.

You can write your own implementation by extending the org.apache.pinot.core.crypt.PinotCrypter interface in a similar fashion. The crypter can be used by passing the crypter class name in the header for the upload request as the header string CRYPTER.

5. Segment assignment strategies

We have various strategies for assigning segments to the available servers. These can be found under the SegmentAssignmentStrategy interface. More details about which one to use depending on your usecase can be found in Tuning Pinot. By default, the BalanceNumSegmentAssignmentStrategy will be used.

You can also write your own by implementing the org.apache.pinot.controller.helix.core.sharding.SegmentAssignmentStrategy interface. The segment assignment strategy can be configured for a table by setting it in the table config as

    "segmentsConfig": {
        "segmentAssignmentStrategy": "BalanceNumSegmentAssignmentStrategy",

6. Data partitioning strategies

We have various algorithms to partition data during segment creation, as listed under the interface PartitionFunction. No partitioning is done by default. You can write your own partitioning function by implementing the interface for The partitioning function can be configured on the required columns in the table config as

     "segmentPartitionConfig": {
        "columnPartitionMap": {
            "userId": {
                "functionName": "murmur",
                "numPartitions": 10

7. Routing strategies

We have many routing strategies which you can find under the RoutingTableBuilder interface. More details about which one to use depending on your usecase can be found in Tuning Pinot. By default we will use DefaultOfflineRoutingTableBuilder or the DefaultRealtimeRoutingTableBuilder.

You can write your own routing table builder by implementing the interface. The routing table builder can be set in the table config as

    "routing": {
        "routingTableBuilderName": "PartitionAwareRealtime",
        "routingTableBuilderOptions": {}

8. Broker endpoint

If you setup a usecase to have multiple brokers, you will have to develop your restful service to accept queries and distribute them across the brokers.

9. Access Control

Access control can be setup at various points in Pinot, such as controller endpoints and broker query endpoints. By default we will use AllowAllAccessFactory and hence not be enforcing any access controls. You can add access control by implementing the AccessControlFactory interface.

The access control factory can be configured in the controller configs by setting the fully qualified class name of the AccessControlFactory in the property controller.admin.access.control.factory.class The access control factory can be configured in the broker configs by setting the fully qualified class name of the AccessControlFactory in the property Any other properties required for initializing the factory can be set in the broker configs as properties with the prefix

10. Minion tasks

Minion tasks can be configured for background activities such as purging data, adding indexes, merging segments. The existing minion tasks can be found under the PinotTaskExecutor interface. You can create your own minion task by implementing the PinotTaskExecutor and the PinotTaskExecutorFactory interfaces. The minion job can be configured to run for your table by setting the config in the table as

     "task": {
        "taskTypeConfigsMap": {
          "PurgeTask": {}

11. Custom configs

Custom configs can be injected into Pinot by adding the customConfigs field in the table config. This field accepts key value pairs.

    "customConfigs": {
        "specialConfig": "testValue",
        "anotherSpecialConfig": "value"

12. Metrics

We use yammer MetricsRegistry to collect metrics within our application components. These metrics can be published to a metrics server with the help of MetricsRegistryRegistrationListener interface. By default, we publish metrics to JMX using the JmxReporterMetricsRegistryRegistrationListener.

You can write a listener to publish metrics to another metrics server by implementing the MetricsRegistryRegistrationListener interface. This listener can be injected into the controller by setting the fully qualified name of the class in the controller configs for the property pinot.controller.metrics.metricsRegistryRegistrationListeners.

You would have to design your own systems to view and monitor these metrics. A list of all the metrics published for each component can be found in ControllerMeter, ControllerGauge, BrokerMeter, BrokerGauge, ServerMeter, ServerGauge, MinionMeter and MinionGauge.

13. Deployables

You can deploy pinot server, broker, controller and minion individually. You can either use out of the box jars and start the components via PinotAdministrator, or run via scripts as described in the Getting Started.

The starter classes for pinot controller, broker, server and minion are ControllerStarter, HelixBrokerStarter, HelixServerStarter and MinionStarter respectively.