Data Pipeline Load of JSON Data from HDFS

The data pipeline functionality enables the loading of data from Apache® Hadoop® Distributed File System (HDFS). You can load various file types stored in HDFS into the Ocient® System. The Ocient System uses data pipelines to transform each document into rows in one or more different tables. The loading and transformation capabilities use a simple SQL-like syntax for transforming data. This tutorial guides you through a simple example of loading data in the JSON format.

HDFS Advanced Source Options

Besides the specified options in the HDFS Source Options section, you can also specify these options that are available for file system sources:

COMPRESSION_METHOD
START_FILENAME
END_FILENAME
START_CREATED_TIMESTAMP
END_CREATED_TIMESTAMP
START_MODIFIED_TIMESTAMP
END_MODIFIED_TIMESTAMP
SORT_BY
SORT_DIRECTION
SORT_REWRITE

For Kerberos authentication, you can also specify these options that are available only for HDFS sources:

PRINCIPAL
KEYTAB_FILE
TOKEN
TOKEN_FILE
RENEW_INTERVAL

For details, see Kerberos Authentication for HDFS Sources. For the CONFIG source option, the most useful properties are the ones with the dfs.client prefix. For the full property reference, see HDFS Default and Core Default. Only properties that affect client connections and read operations apply to data pipeline loading. This table describes some common properties.

Property Name	Property Description
`dfs.client.use.datanode.hostname`	Set this property to `true` when you have datanodes and namenodes in the HDFS cluster configured to connect to each other using local IP addresses, but loading occurs externally and requires externally valid hostnames instead. The default value is `false`.
`dfs.client.retry.max.attempts`	Configure to change the number of times the loading process retries network requests to namenodes before throwing an error. You can increase this value when network connections are unstable. The default value is 10.
`dfs.client.socket-timeout`	Configure to change the timeout (in milliseconds) for all sockets used internally in HDFS loading. You can increase this value when network connections are unstable. The default value is 60000 (60 seconds).

Kerberos Authentication for HDFS Sources

HDFS sources support Kerberos authentication with these two methods:

Keytab file — Specify the PRINCIPAL and KEYTAB_FILE options.
Delegation token — Specify the TOKEN or TOKEN_FILE option and, optionally, the RENEW_INTERVAL option.

You can specify only one authentication method for each data pipeline. When you create a data pipeline that specifies Kerberos authentication, the Ocient System authenticates to the HDFS cluster during pipeline compilation. If authentication fails, the CREATE PIPELINE SQL statement fails. In most cases, you must also specify the dfs.namenode.kerberos.principal and dfs.data.transfer.protection configuration properties in the CONFIG option so that the loading process can identify the Kerberos principal of the namenode server. HDFS only supports one Key Distribution Center (KDC) path for each JVM®. Therefore, the KDC path affects all HDFS pipelines within the same system instance. If multiple HDFS pipelines are running, they must share the krb5.conf file by listing the principals and realms in it, or you must run them sequentially. A valid Kerberos configuration file (krb5.conf) must exist on all Loader Nodes regardless of the authentication method. For help locating the file, see Locating the krb5.conf Configuration File. If you do not configure the file location, the system searches the default operating system locations, such as the /etc/krb5.conf file on Linux®. To specify a different file location, or to refresh the file after you update its contents, set the streamloader.extractorEngineParameters.configurationOption.kerberosConfig configuration setting. This setting affects all data pipelines that run on the Loader Node. For details, see Configuration Settings for Data Pipelines. This example sets the Kerberos configuration file location to the /etc/krb5.conf file.

SQL

ALTER SYSTEM ALTER CONFIG SET
'streamloader.extractorEngineParameters.configurationOption.kerberosConfig'
'/etc/krb5.conf';

All HDFS data pipelines that run concurrently on the same Loader Node share the same Kerberos configuration file. Include the principals and realms for all data pipelines in that file. Keytab file authentication is the most reliable method. The loading process requests new credentials from the key distribution center (KDC) and automatically logs in again as needed during the load. With delegation token authentication, the loading process renews the token at the frequency that the RENEW_INTERVAL option specifies. The KDC cannot renew a token after the token reaches the renewal lifetime (renew_lifetime) that you configure in the Kerberos configuration. If the credentials expire before the load finishes, such as during a long-running or continuous data pipeline, the pipeline fails, and the load cannot progress with the expired credentials. For these cases, use keytab file authentication instead. To resume a failed load with a new delegation token, update the token using the ALTER PIPELINE SQL statement and restart the pipeline. For examples, see “Load Data from HDFS with Kerberos Keytab Authentication” and “Load Data from HDFS with Kerberos Delegation Token Authentication” on the Data Pipelines reference page.

HDFS Loading Example

Follow these steps to load JSON data from an HDFS source into the Ocient System.

Step 1: Create a New Database

Connect to a SQL Node using the Commands Supported by the Ocient JDBC CLI Program. Then, execute the CREATE DATABASE SQL statement to create the geo database.

SQL

CREATE DATABASE geo;

Step 2: Create a New Table in the Database

Create the locations table in the public schema to store location data with a name, zip code, and a point with latitude and longitude:

name — Location name as a not nullable string
zipcode — Zip code as an integer
location — Location latitude and longitude as a point

SQL

CREATE TABLE public.locations (
    name VARCHAR(255) NOT NULL,
    zipcode INT NOT NULL,
    location POINT
);

Step 3: Preview and Create a Data Pipeline

Preview the locations data pipeline to load JSON data from HDFS. This data pipeline uses the HDFS endpoint hdfs-namenode:9000, which consists of the namenode and port number, and the /locations/2026/**/*.json filter to load data from JSON files. The pipeline selects the name, zip code, and point data. For the point data construction, see ST_POINT.

SQL

PREVIEW PIPELINE locations
  SOURCE hdfs
    ENDPOINT 'hdfs-namenode:9000'
    FILTER '/locations/2026/**/*.json'
  EXTRACT
    FORMAT json
  INTO locations
  SELECT
    $name AS name,
    $zipcode AS zipcode,
    ST_POINT($longitude, $latitude) AS location;

Create the locations data pipeline to load JSON data from HDFS.

SQL

CREATE PIPELINE locations
  SOURCE hdfs
    ENDPOINT 'hdfs-namenode:9000'
    FILTER '/locations/2026/**/*.json'
  EXTRACT
    FORMAT json
  INTO locations
  SELECT
    $name AS name,
    $zipcode AS zipcode,
    ST_POINT($longitude, $latitude) AS location;

After you successfully create the locations pipeline, execute the START PIPELINE SQL statement to start the data pipeline.

START PIPELINE orders_pipeline;

Step 4: Observe the Load Progress

With your pipeline running, data begins to load immediately from the JSON files. If there are many files in each file group, the load process first sorts the files into batches, partitions them for parallel processing, and assigns them to Loader Nodes. You can check the pipeline status and progress by querying the information_schema.pipeline_status system catalog table or executing the SHOW PIPELINE_STATUS SQL statement.

SHOW PIPELINE_STATUS;

Output

database_name         pipeline_name       table_names             status      status_message                               percent_complete duration_seconds files_processed     files_failed        files_remaining     records_processed   records_loaded      records_failed
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
geo                   locations           ["public.locations"]    RUNNING     Started processing pipeline locations        0.0              2.025824         0                   0                   1                   0                   0                   0

After the status of the pipeline changes to COMPLETED, all data is available in the target table. After a few seconds, the data is available for query in the public.locationstable.

SELECT COUNT(*) FROM public.locations;

Output

count(*)
--------------------
25605

You can drop the pipeline with the DROP PIPELINE locations; SQL statement. Execution of this statement leaves the data in your target table, but removes metadata about the pipeline execution from the system. Data Pipelines Reference Load JSON Data Data Types for Data Pipelines Transform Data in Data Pipelines Manage Errors in Data Pipelines

​HDFS Advanced Source Options

​Kerberos Authentication for HDFS Sources

​HDFS Loading Example

​Step 1: Create a New Database

​Step 2: Create a New Table in the Database

​Step 3: Preview and Create a Data Pipeline

​Step 4: Observe the Load Progress

​Related Links