Configuring error handling controls for Kafka event sources

You can configure how Lambda handles errors and retries for your Kafka event source mappings. These configurations help you control how Lambda processes failed records and manages retry behavior.

Available retry configurations

The following retry configurations are available for both Amazon MSK and self-managed Kafka event sources:

Configuring error handling controls (console)

You can configure retry behavior when creating or updating a Kafka event source mapping in the Lambda console.

Configuring retry behavior (AWS CLI)

Use the following AWS CLI commands to configure retry behavior for your Kafka event source mappings.

Creating an event source mapping with retry configurations

The following example creates a self-managed Kafka event source mapping with error handling controls:

aws lambda create-event-source-mapping \
  --function-name my-kafka-function \
  --topics my-kafka-topic \
  --source-access-configuration Type=SASL_SCRAM_512_AUTH,URI=arn:aws:secretsmanager:us-east-1:111122223333:secret:MyBrokerSecretName \
  --self-managed-event-source '{"Endpoints":{"KAFKA_BOOTSTRAP_SERVERS":["abc.xyz.com:9092"]}}' \
  --starting-position LATEST \
  --provisioned-poller-config MinimumPollers=1,MaximumPollers=1 \
  --maximum-retry-attempts 3 \
  --maximum-record-age-in-seconds 3600 \
  --bisect-batch-on-function-error \
  --function-response-types "ReportBatchItemFailures"

For Amazon MSK event sources:

aws lambda create-event-source-mapping \
  --event-source-arn arn:aws:kafka:us-east-1:111122223333:cluster/my-cluster/fc2f5bdf-fd1b-45ad-85dd-15b4a5a6247e-2 \
  --topics AWSMSKKafkaTopic \
  --starting-position LATEST \
  --function-name my-kafka-function \
  --source-access-configurations '[{"Type": "SASL_SCRAM_512_AUTH","URI": "arn:aws:secretsmanager:us-east-1:111122223333:secret:my-secret"}]' \
  --provisioned-poller-config MinimumPollers=1,MaximumPollers=1 \
  --maximum-retry-attempts 3 \
  --maximum-record-age-in-seconds 3600 \
  --bisect-batch-on-function-error \
  --function-response-types "ReportBatchItemFailures"

Updating retry configurations

Use the update-event-source-mapping command to modify retry configurations for an existing event source mapping:

aws lambda update-event-source-mapping \
  --uuid 12345678-1234-1234-1234-123456789012 \
  --maximum-retry-attempts 5 \
  --maximum-record-age-in-seconds 7200 \
  --bisect-batch-on-function-error \
  --function-response-types "ReportBatchItemFailures"

PartialBatchResponse

Partial batch response, also known as ReportBatchItemFailures, is a key feature for error handling in Lambda's integration with Kafka sources. Without this feature, when an error occurs in one of the items in a batch, it results in reprocessing all messages in that batch. With partial batch response enabled and implemented, the handler returns identifiers only for the failed messages, allowing Lambda to retry just those specific items. This provides greater control over how batches containing failed messages are processed.

To report batch errors, you will use this JSON schema:

{
  "batchItemFailures": [
    {
      "itemIdentifier": {
        "topic-partition": "topic-partition_number",
        "offset": 100
      }
    },
    ...
  ]
}

The following code examples show how to implement partial batch response for Lambda functions that receive events from Kafka sources. The function reports the batch item failures in the response, signaling to Lambda to retry those messages later.

Here is a Python Lambda handler implementation that shows this approach:

import base64
from typing import Any, Dict, List

def lambda_handler(event: Dict[str, Any], context: Any) -> Dict[str, List[Dict[str, Dict[str, Any]]]]:
    failures: List[Dict[str, Dict[str, Any]]] = []
    records_dict = event.get("records", {})
    
    for topic_partition, records_list in records_dict.items():
        for record in records_list:
            topic = record.get("topic")
            partition = record.get("partition")
            offset = record.get("offset")
            value_b64 = record.get("value")
            
            try:
                data = base64.b64decode(value_b64).decode("utf-8")
                process_message(data)
            except Exception as exc:
                print(f"Failed to process record topic={topic} partition={partition} offset={offset}: {exc}")
                item_identifier: Dict[str, Any] = {
                    "topic-partition": f"{topic}-{partition}",
                    "offset": int(offset) if offset is not None else None,
                }
                failures.append({"itemIdentifier": item_identifier})
    
    return {"batchItemFailures": failures}

def process_message(data: str) -> None:
    # Your business logic for a single message
    pass

Here is a Node.js version:

const { Buffer } = require("buffer");

const handler = async (event) => {
  const failures = [];
  
  for (let topicPartition in event.records) {
    const records = event.records[topicPartition];
    
    for (const record of records) {
      const topic = record.topic;
      const partition = record.partition;
      const offset = record.offset;
      const valueBase64 = record.value;
      const data = Buffer.from(valueBase64, "base64").toString("utf8");
      
      try {
        await processMessage(data);
      } catch (error) {
        console.error("Failed to process record", { topic, partition, offset, error });
        const itemIdentifier = {
          "topic-partition": `${topic}-${partition}`,
          offset: Number(offset),
        };
        failures.push({ itemIdentifier });
      }
    }
  }
  
  return { batchItemFailures: failures };
};

async function processMessage(payload) {
  // Your business logic for a single message
}

module.exports = { handler };