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
  • Confluent 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": "Dot Matrix Printers",
    "description": "Impactful industrial implements for impressing ink iteratively",
    "stock": 27
}

 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": "Dot Matrix Printers",
        "description": "Impactful industrial implements for impressing ink iteratively",
        "stock": 27
    }
]

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.

GAdmin

The Import feature of Kinetica's GAdmin imports data from a variety of sources using three different interfaces:

• A drag & drop tool, for loading local files
• An Advanced Import (KIO) tool, for complex file types or remote data sources
• An Advanced CSV Import tool, for loading local files with complex configurations

To access the Import page and begin loading data, click the Data menu option in the main navigation bar and then the Import menu item. Then, choose the tool that best suits your needs.

Workbench

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.

JDBC Client

Any JDBC client can use the Kinetica JDBC driver to initiate data loads. The two primary commands for data loading are:

• CREATE EXTERNAL TABLE: Create a data container that is bound to a given source of data.
• LOAD INTO: Initiate a data load or subscription into a table that will not be bound to a given source of data.

• CREATE TABLE EXTERNAL
• LOAD INTO

1
2
3

CREATE LOGICAL EXTERNAL TABLE example.ext_employee_dept2
REMOTE QUERY 'SELECT * FROM example.employee WHERE dept_id = 2'
WITH OPTIONS (DATA SOURCE = 'example.jdbc_ds')

1
2
3

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

See Advanced Ingestion for more comprehensive ingestion scheme coverage, Examples for more ingestion examples in SQL, or Loading Data - Snippets for a wide variety of copy/paste SQL loading examples.

KiSQL

KiSQL is a java-based application that is capable of performing all the commands that a JDBC client can as noted above, as well as 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'

Type Inferencing

If your table does not already exist, the data types will be automatically inferred upon ingestion based on the data values processed. There are two type inferencing modes available through the TYPE_INFERENCE_MODE option:

• SPEED: (default) Only use a portion of the incoming data to determine column types, potentially over-estimating the column width needs.
• ACCURACY: Use the entire data set to determine the exact column types needed to hold the data, taking much longer than the speed-based estimate.

• Speed (Default) Type Inference
• Accuracy Type Inference

1
2
3

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

1
2
3

LOAD INTO example.employee_inf_accuracy
FROM REMOTE QUERY 'SELECT * FROM example.employee'
WITH OPTIONS (DATA SOURCE = 'example.jdbc_ds', TYPE_INFERENCE_MODE = 'ACCURACY')

Change Data Capture

Kinetica supports a change data capture subscription scheme for both files and queries, loading only data that has changed since the last check:

• For files, subscriptions will save the timestamp of the previous check, and at each interval, check for files modified after that time.
• For JDBC queries, the REMOTE_QUERY_INCREASING_COLUMN option is used to name a column in the source table that can be used to identify new data. The value in the column needs to be ever-increasing and will often be a sequence-based ID or a timestamp.

• CDC via Remote File Store
• CDC via Query

 1
 2
 3
 4
 5
 6
 7
 8
 9
10

-- Load files in the orders directory with timestamps newer than the time of the
--   previous poll interval; poll every 5 minutes
LOAD DATA INTO example.orders
FROM FILE PATHS 'orders/'
WITH OPTIONS
(
      DATA SOURCE = 'example.order_ds',
      SUBSCRIBE = TRUE,
      POLL_INTERVAL = 300
)

 1
 2
 3
 4
 5
 6
 7
 8
 9
10

-- Load new orders for product 42 continuously into a table
--   order_id is an ever-increasing sequence allotted to each new order
LOAD INTO example.order_product42
FROM REMOTE QUERY 'SELECT * FROM example.orders WHERE product_id = 42'
WITH OPTIONS
(
      DATA SOURCE = 'example.jdbc_ds',
      SUBSCRIBE = TRUE,
      REMOTE_QUERY_INCREASING_COLUMN = 'order_id'
)

Tuning

1
2
3

LOAD DATA INTO example.product_load_options
FROM FILE PATHS 'kifs://data/products.csv'
WITH OPTIONS (BATCH SIZE = 20000)

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.

File/Queue-Based

• One-Time Load
• Change Data Capture
• Streaming
• External Table

1
2
3

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

 1
 2
 3
 4
 5
 6
 7
 8
 9
10

-- Load files in the orders directory with timestamps newer than the time of the
--   previous poll interval; poll every 5 minutes
LOAD DATA INTO example.orders
FROM FILE PATHS 'orders/'
WITH OPTIONS
(
      DATA SOURCE = 'example.order_ds',
      SUBSCRIBE = TRUE,
      POLL_INTERVAL = 300
)

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
)

Query-Based

• One-Time Load
• Change Data Capture
• External Table

1
2
3

LOAD INTO example.employee_dept2
FROM REMOTE QUERY 'SELECT * FROM example.employee WHERE dept_id = 2'
WITH OPTIONS (DATA SOURCE = 'example.jdbc_ds')

 1
 2
 3
 4
 5
 6
 7
 8
 9
10

-- Load new orders for product 42 continuously into a table
--   order_id is an ever-increasing sequence allotted to each new order
LOAD INTO example.order_product42
FROM REMOTE QUERY 'SELECT * FROM example.orders WHERE product_id = 42'
WITH OPTIONS
(
      DATA SOURCE = 'example.jdbc_ds',
      SUBSCRIBE = TRUE,
      REMOTE_QUERY_INCREASING_COLUMN = 'order_id'
)

1
2
3
4
5
6
7

CREATE EXTERNAL TABLE example.ext_employee_dept2
REMOTE QUERY 'SELECT * FROM example.ext_employee WHERE dept_id = 2'
WITH OPTIONS
(
      DATA SOURCE = 'example.jdbc_ds',
      REFRESH ON START = TRUE
)

More SQL Examples

See this page for a larger list of SQL examples using the LOAD INTO syntax with various options and authentication methods.