After careful consideration, we have decided to discontinue Amazon Kinesis Data Analytics for SQL applications in two steps:

1. From October 15, 2025, you will not be able to create new Kinesis Data Analytics for SQL applications.

2. We will delete your applications starting January 27, 2026. You will not be able to start or operate your Amazon Kinesis Data Analytics for SQL applications. Support will no longer be available for Amazon Kinesis Data Analytics for SQL from that time. For more information, see Amazon Kinesis Data Analytics for SQL Applications discontinuation.

Parallelizing Input Streams for Increased Throughput

Amazon Kinesis Data Analytics applications can support multiple in-application input streams, to scale an application beyond the throughput of a single in-application input stream. For more information on in-application input streams, see Amazon Kinesis Data Analytics for SQL Applications: How It Works.

In almost all cases, Amazon Kinesis Data Analytics scales your application to handle the capacity of the Kinesis streams or Firehose source streams that feed into your application. However, if your source stream's throughput exceeds the throughput of a single in-application input stream, you can explicitly increase the number of in-application input streams that your application uses. You do so with the InputParallelism parameter.

When the InputParallelism parameter is greater than one, Amazon Kinesis Data Analytics evenly splits the partitions of your source stream among the in-application streams. For instance, if your source stream has 50 shards, and you set InputParallelism to 2, each in-application input stream receives the input from 25 source stream shards.

When you increase the number of in-application streams, your application must access the data in each stream explicitly. For information about accessing multiple in-application streams in your code, see Accessing Separate In-Application Streams in Your Amazon Kinesis Data Analytics Application.

Although Kinesis Data Streams and Firehose stream shards are both divided among in-application streams in the same way, they differ in the way they appear to your application:

Evaluating Whether to Increase Your Number of In-Application Input Streams

In most cases, a single in-application input stream can handle the throughput of a single source stream, depending on the complexity and data size of the input streams. To determine if you need to increase the number of in-application input streams, you can monitor the InputBytes and MillisBehindLatest metrics in Amazon CloudWatch.

If the InputBytes metric is greater that 100 MB/sec (or you anticipate that it will be greater than this rate), this can cause an increase in MillisBehindLatest and increase the impact of application issues. To address this, we recommend making the following language choices for your application:

If the MillisBehindLatest metric has either of the following characteristics, you should increase your application's InputParallelism setting:

You don't need to increase your application's InputParallelism setting if the following are true:

For more information on using CloudWatch, see the CloudWatch User Guide.

Implementing Multiple In-Application Input Streams

You can set the number of in-application input streams when an application is created using CreateApplication. You set this number after an application is created using UpdateApplication.

Setting a New Application's Input Stream Count

The following example demonstrates how to use the CreateApplication API action to set a new application's input stream count to 2.

For more information about CreateApplication, see CreateApplication.

{
   "ApplicationCode": "<The SQL code the new application will run on the input stream>",
   "ApplicationDescription": "<A friendly description for the new application>",
   "ApplicationName": "<The name for the new application>",
   "Inputs": [ 
    { 
      "InputId": "ID for the new input stream",
      "InputParallelism": { 
        "Count": 2
    }],
   "Outputs": [ ... ],
	}]
}

Setting an Existing Application's Input Stream Count

The following example demonstrates how to use the UpdateApplication API action to set an existing application's input stream count to 2.

For more information about Update_Application, see UpdateApplication.

{
   "InputUpdates": [ 
      { 
         "InputId": "yourInputId",
         "InputParallelismUpdate": { 
            "CountUpdate": 2
         }
      }
   ],
}

Accessing Separate In-Application Streams in Your Amazon Kinesis Data Analytics Application

To use multiple in-application input streams in your application, you must explicitly select from the different streams. The following code example demonstrates how to query multiple input streams in the application created in the Getting Started tutorial.

In the following example, each source stream is first aggregated using COUNT before being combined into a single in-application stream called in_application_stream001. Aggregating the source streams beforehand helps make sure that the combined in-application stream can handle the traffic from multiple streams without being overloaded.

CREATE OR REPLACE STREAM in_application_stream_001 (
    ticker VARCHAR(64),
    ticker_count INTEGER
);

CREATE OR REPLACE PUMP pump001 AS 
INSERT INTO in_application_stream_001
SELECT STREAM ticker_symbol, COUNT(ticker_symbol)
FROM source_sql_stream_001
GROUP BY STEP(source_sql_stream_001.rowtime BY INTERVAL '60' SECOND),
    ticker_symbol; 
        
CREATE OR REPLACE PUMP pump002 AS 
INSERT INTO in_application_stream_001
SELECT STREAM ticker_symbol, COUNT(ticker_symbol)
FROM source_sql_stream_002
GROUP BY STEP(source_sql_stream_002.rowtime BY INTERVAL '60' SECOND),
    ticker_symbol;
        

The preceding code example produces output in in_application_stream001 similar to the following:

Additional Considerations

When using multiple input streams, be aware of the following: