Data Loading Concepts

The Basics of Loading Data

At a high level, Kinetica ingests data that is stored in a supported data provider like S3 or Azure BLOB, or Kinetica's native file system (KiFS). Your data will need to be stored in a supported provider, or moved to one for Kinetica to load it. If your data cannot be stored in a supported provider you can upload files to Kinetica's file system using the UI-driven workflow, Kinetica's command line SQL interface (KiSQL), or using one of Kinetica's native APIs. The data itself will need to be stored in a supported format, like CSV or Parquet (see the full list below).

Once Kinetica has access to your data and it is in a supported format, you can load data either from a UI-driven workflow, or by using SQL directly.

Supported Data Sources

A data source is an object that references a location on a remote system, and acts as a proxy for accessing external data. Once a data source is created you can access different files stored within.

Data sources may be created for the following providers:

• Azure: Microsoft BLOB Storage
• GCS: Google Cloud Storage
• HDFS: Apache Hadoop Distributed File System
• JDBC: Java Database Connectivity, using a user-supplied driver
• Kafka: Apache Kafka streaming feed
• S3: Amazon S3 Bucket

If the file is not stored in a compatible provider, then you will need to move it to a supported provider or location in order to ingest it.

Creating a Data Source

1
2
3
4

CREATE CREDENTIAL s3_cred
TYPE = 'aws_access_key',
IDENTITY = 'AKIAIOSFODNN7EXAMPLE',
SECRET = 'wJalrXUtnFEMI/K7MDENG/bPxRfiCYEXAMPLEKEY'

1
2
3
4
5
6
7

CREATE OR REPLACE DATA SOURCE s3_ds
LOCATION = 'S3'
WITH OPTIONS (
  CREDENTIAL = 's3_cred',
  BUCKET NAME = 'examplebucket',
  REGION = 'us-east-1'
)

Parquet Limitations

Kinetica supports most primitive Arrow types when ingesting data from Parquet files. Non-primitive types will be either transformed or skipped.

JSON/GeoJSON Limitations

For formats like JSON and GeoJSON, both a single JSON object and multiple JSON objects (via array) can be ingested:

• Single Record (Object)
• Multiple Records (Array of Objects)

1
2
3
4
5
6
7

{
    "id": "14",
    "category": "Technology",
    "name": "DVDs",
    "description": "Lightweight storage for low-res screenplays and licensed software",
    "stock": "5000"
}

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16

[
    {
        "id": "1",
        "category": "Furniture",
        "name": "Clamp-on Lamps",
        "description": "Strong, steadfast fasteners; localized lighting for laboring",
        "stock": "10"
    },
    {
        "id": "14",
        "category": "Technology",
        "name": "DVDs",
        "description": "Lightweight storage for low-res screenplays and licensed software",
        "stock": "5000"
    }
]

The following handling will be applied to individual value types:

• Primitives - ingested as is

• Arrays - stringified and stored in unrestricted string columns, rendering values within them inaccessible as column data

• Objects - flattened; e.g.:

  Source structure	Aa {Bb, Cc {Dd, Ee}}
  Resulting (3) columns	Aa_Bb Aa_Cc_Dd Aa_Cc_Ee

If you require arrays to be flattened, so that they can be directly accessed as column data, you will need to pre-flatten your data before ingesting it into Kinetica. This must be done using a third-party tool or script.

Loading via Workbench UI

The Import interface of Kinetica's Workbench imports data from a variety of sources using a simple wizard workflow. In addition to importing data, it generates the SQL equivalent that can be copied and pasted into a workbook.

To access the Import page and begin loading data, click the Import link in the main navigation bar. Then, follow the workflow to import data from one of Kinetica's supported data providers.

Loading via SQL

1
2
3

LOAD DATA INTO example.product
FROM FILE PATHS 'products.csv'
WITH OPTIONS (DATA SOURCE = 'example.product_ds')

1
2
3

LOAD DATA INTO example.product
FROM FILE PATHS 'products.csv'
WITH OPTIONS (DATA SOURCE = 'example.product_ds', SUBSCRIBE = TRUE)

1
2
3

LOAD DATA INTO example.orders
FORMAT JSON
WITH OPTIONS (DATA SOURCE = 'kafka_ds', SUBSCRIBE = TRUE)

1
2
3
4
5
6
7

CREATE EXTERNAL TABLE example.ext_product
FILE PATHS 'products.csv'
WITH OPTIONS
(
      DATA SOURCE = 'example.product_ds',
      REFRESH ON START = TRUE
)

Loading with KiSQL

KiSQL is a java-based application that is capable of uploading files from a client machine into Kinetica's KiFS. This is a great option if you don't have direct access to Kinetica's web interface, or if you want to script files to be loaded on a set schedule from a client.

First, download KiSQL from this GitHub Repository. If you're on a Windows machine, you do not need Java pre-installed. For MacOS or Linux machines, you will need to install Java 8 to run KiSQL.

Next, start KiSQL with the URL of your Kinetica instance.

1

./kisql --url <kinetica_url> --user <username>

Create a directory, in which to load your file(s).

1

CREATE DIRECTORY 'my_directory';

Upload your file(s) to KiFS.

1

UPLOAD FILES 'my_data.csv' INTO 'my_directory';

Finally, load your file into a table.

1
2

LOAD INTO ki_home.my_data
FROM FILE PATHS 'kifs://my_directory/my_data.csv'

Custom Ingest Applications

In some circumstances, you may want to write a custom application to ingest data into Kinetica using Kinetica's native APIs or JDBC Driver. Kinetica offers several endpoints for uploading files, loading data, or inserting records. The usage of such endpoints is outside of the scope of this article. See Kinetica's API section for more information.

Kinetica's native APIs are capable of distributing records to the various nodes of your cluster at the time of ingest. This provides a more direct pathway that avoids the need for records to first be sent to Kinetica's head node for evaluation. When writing a custom application, refer to the Multi-head Ingest page for information on how to achieve the best ingestion performance.

3rd Party Tools

If the data you wish to load is not in a supported format, if it is stored in an incompatible provider, or it requires ETL before you load it into Kinetica, you may use Kinetica's JDBC Driver in conjunction with an ETL tool to load data into Kinetica. In many cases, this process is as easy as dropping in Kinetica's JDBC Driver and defining a few configuration parameters. Note, however, that this depends on the tool you have chosen and the way that the JDBC connection was implemented.

For more information about how to use Kinetica's JDBC Driver, see the JDBC driver documentation.

Batch Size

Users can adjust the batch size used to insert records with the SQL Hint KI_HINT_BATCH_SIZE(n), which defaults to 200,000. For tables with very large columns (e.g., extremely complex WKT records), users should adjust the batch size down to see an increase in performance. For very simple tables with short columns, users can adjust the batch size to be higher to see an increase in throughput. These are just rough guidelines and your mileage may vary depending on your use case.