Amazon.com, Inc.
Idempotent processing of data streams

Abstract:

Idempotent processing of data may be implemented for data records retrieved from a data stream. A data stream may receive data records as input and distribute the ingestion, storage, and processing of the data records amongst one or more partitions of the data stream. Partition metadata may be maintained which includes checkpoint metadata for retrieving, processing, and sending data records in the data stream to a specified destination. When assigned a partition for processing, checkpoint metadata for partition may be accessed to determine whether a pending checkpoint for the partition exists. If not pending checkpoint exists, new data records may be retrieved, processed, and sent from the partition of the data stream to a specified destination. If a checkpoint is pending, then the data records identified by the checkpoint metadata as pending may be retrieved, processed, and sent to the specified destination.