A company has a data ingestion workflow consisting of an Amazon SNS topic that receives notifications about new data deliveries and an AWS Lambda function that processes and stores the data. Occasionally the ingestion fails due to network connectivity issues, and those data are not ingested unless a job is manually rerun. What should a solutions architect do to ensure all notifications are eventually processed?

The correct solution is to configure an Amazon SQS queue as the on-failure destination for the Lambda function. When the Lambda function fails, the message will be sent to the SQS queue, ensuring no data is lost. A separate process or the same Lambda function can then process messages from the SQS queue, providing an eventual consistency model for data ingestion. Configuring the Lambda function across multiple Availability Zones increases availability but doesn't address processing failures. Increasing CPU and memory might prevent some failures but doesn't guarantee all data is processed if other issues occur. Configuring the SNS topic's retry strategy is for SNS delivery to subscribers, not for Lambda processing failures after invocation.