This post is written by Mithun Mallick, Senior Specialist Solutions Architect.
AWS Lambda now supports configuring a maximum batch window for instance-based message broker services to fine tune when Lambda invocations occur. This feature gives you an additional control on batching behavior when processing data. It applies to Amazon Managed Streaming for Apache Kafka (Amazon MSK), self-hosted Apache Kafka, and Amazon MQ for Apache ActiveMQ and RabbitMQ.
Apache Kafka is an open source event streaming platform used to support workloads such as data pipelines and streaming analytics. It is conceptually similar to Amazon Kinesis. Amazon MSK is a fully managed, highly available service that simplifies the setup, scaling, and management of clusters running Kafka.
Amazon MQ is a managed, highly available message broker service for Apache ActiveMQ and RabbitMQ that makes it easier to set up and operate message brokers on AWS. Amazon MQ reduces your operational responsibilities by managing the provisioning, setup, and maintenance of message brokers for you.
Amazon MSK, self-hosted Apache Kafka and Amazon MQ for ActiveMQ and RabbitMQ are all available as event sources for AWS Lambda. You configure an event source mapping to use Lambda to process items from a stream or queue. This allows you to use these message broker services to store messages and asynchronously integrate them with downstream serverless workflows.
In this blog, I explain how message batching works. I show how to use the new maximum batching window control for the managed message broker services and self-managed Apache Kafka.
For event source mappings, the Lambda service internally polls for new records or messages from the event source, and then synchronously invokes the target Lambda function. Lambda reads the messages in batches and provides these to your function as an event payload. Batching allows higher throughput message processing, up to 10,000 messages in a batch. The payload limit of a single invocation is 6 MB.
Previously, you could only use batch size to configure the maximum number of messages Lambda would poll for. Once a defined batch size is reached, the poller invokes the function with the entire set of messages. This feature is ideal when handling a low volume of messages or batches of data that take time to build up.
The new Batch Window control allows you to set the maximum amount of time, in seconds, that Lambda spends gathering records before invoking the function. This brings similar batching functionality that AWS supports with Amazon SQS to Amazon MQ, Amazon MSK and self-managed Apache Kafka. The Lambda event source mapping batching functionality can be described as follows.
MaximumBatchingWindowInSeconds, you can set your function to wait up to 300 seconds for a batch to build before processing it. This allows you to create bigger batches if there are enough messages. You can manage the average number of records processed by the function with each invocation. This increases the efficiency of each invocation, and reduces the frequency.
MaximumBatchingWindowInSeconds to 0 invokes the target Lambda function as soon as the Lambda event source receives a message from the broker.
Message broker batching behavior
For ActiveMQ, the Lambda event source mapping uses the Java Message Service (JMS) API to receive messages. For RabbitMQ, Lambda uses a RabbitMQ client library to get messages from the queue.
The Lambda event source mappings act as a consumer when polling the queue. The batching pattern for all instance-based message broker services is the same. As soon as a message is received, the batching window timer starts. If there are more messages, the consumer makes additional calls to the broker and adds them to a buffer. It keeps a count of the number of messages and the total size of the payload.
The batch is considered complete if the addition of a new message makes the batch size equal to or greater than 6 MB, or the batch window timeout is reached. If the batch size is greater than 6 MB, the last message is returned back to the broker.
Lambda then invokes the target Lambda function synchronously and passes on the batch of messages to the function. The Lambda event source continues to poll for more messages and as soon as it retrieves the next message, the batching window starts again. Polling and invocation of the target Lambda function occur in separate processes.
Kafka uses a distributed append log architecture to store messages. This works differently from ActiveMQ and RabbitMQ as messages are not removed from the broker once they have been consumed. Instead, consumers must maintain an offset to the last record or message that was consumed from the broker. Kafka provides several options in the consumer API to simplify the tracking of offsets.
Amazon MSK and Apache Kafka store data in multiple partitions to provide higher scalability. Lambda reads the messages sequentially for each partition and a batch may contain messages from different partitions. Lambda then commits the offsets once the target Lambda function is invoked successfully.
Configuring the maximum batching window
To reduce Lambda function invocations for existing or new functions, set the
MaximumBatchingWindowInSeconds value close to 300 seconds. A longer batching window can introduce additional latency. For latency-sensitive workloads set the
MaximumBatchingWindowInSeconds value to an appropriate setting.
To configure Maximum Batching on a function in the AWS Management Console, navigate to the function in the Lambda console. Create a new Trigger, or edit an existing once. Along with the Batch size you can configure a Batch window. The Trigger Configuration page is similar across the broker services.
You can also use the AWS CLI to configure the
For example, with Amazon MQ:
aws lambda create-event-source-mapping --function-name my-function \ --maximum-batching-window-in-seconds 300 --batch-size 100 --starting-position AT_TIMESTAMP \ --event-source-arn arn:aws:mq:us-east-1:123456789012:broker:ExampleMQBroker:b-24cacbb4-b295-49b7-8543-7ce7ce9dfb98
You can use AWS CloudFormation to configure the parameter. The following example configures the
MaximumBatchingWindowInSeconds as part of the
AWS::Lambda::EventSourceMapping resource for Amazon MQ:
LambdaFunctionEventSourceMapping: Type: AWS::Lambda::EventSourceMapping Properties: BatchSize: 10 MaximumBatchingWindowInSeconds: 300 Enabled: true Queues: - "MyQueue" EventSourceArn: !GetAtt MyBroker.Arn FunctionName: !GetAtt LambdaFunction.Arn SourceAccessConfigurations: - Type: BASIC_AUTH URI: !Ref secretARNParameter
You can also use AWS Serverless Application Model (AWS SAM) to configure the parameter as part of the Lambda function event source.
MQReceiverFunction: Type: AWS::Serverless::Function Properties: FunctionName: MQReceiverFunction CodeUri: src/ Handler: app.lambda_handler Runtime: python3.9 Events: MQEvent: Type: MQ Properties: Broker: !Ref brokerARNParameter BatchSize: 10 MaximumBatchingWindowInSeconds: 300 Queues: - "workshop.queueC" SourceAccessConfigurations: - Type: BASIC_AUTH URI: !Ref secretARNParameter
If your function times out or returns an error for any of the messages in a batch, Lambda retries the whole batch until processing succeeds or the messages expire.
When a function encounters an unrecoverable error, the event source mapping is paused and the consumer stops processing records. Any other consumers can continue processing, provided that they do not encounter the same error. If your Lambda event records exceed the allowed size limit of 6 MB, they can go unprocessed.
For Amazon MQ, you can redeliver messages when there’s a function error. You can configure dead-letter queues (DLQs) for both Apache ActiveMQ, and RabbitMQ. For RabbitMQ, you can set a per-message TTL to move failed messages to a DLQ.
Since the same event may be received more than once, functions should be designed to be idempotent. This means that receiving the same event multiple times does not change the result beyond the first time the event was received.
Lambda supports a number of event sources including message broker services like Amazon MQ and Amazon MSK. This post explains how batching works with the event sources and how messages are sent to the Lambda function.
Previously, you could only control the batch size. The new Batch Window control allows you to set the maximum amount of time, in seconds, that Lambda spends gathering records before invoking the function. This can increase the overall throughput of message processing and reduces Lambda invocations, which may improve cost.
For more serverless learning resources, visit Serverless Land.