Jet

Hazelcast Jet is a distributed stream and batch processing engine built on the Hazelcast In-Memory Data Grid (IMDG), designed for applications that require ultra-low latency and high throughput in real-time data processing. Originally released as an independent product in 2010 by Hazelcast, Inc. (headquartered in Palo Alto, United States), Jet targets developers and architects building data-intensive systems such as real-time analytics, fraud detection, IoT data pipelines, and microservices-based event processing. It is particularly suited for use cases where processing speed under 10 milliseconds and handling millions of events per second are critical, and where the operational simplicity of an embedded deployment model is preferred over separate clusters. Jet's programming interface closely mirrors the Java Stream API, making it accessible to Java developers familiar with functional-style data transformations.

Jet achieves its performance through a combination of in-memory computing, cooperative multi-threading (green threads), and a Directed Acyclic Graph (DAG) execution model. In benchmarks published by Hazelcast, Jet processed over 2 GB per second in a word count batch job (640 GB dataset, 1 million distinct keys), outperforming Apache Spark and Apache Flink on the same hardware. For streaming workloads, Jet can sustain 10 million events per second with a latency of 10 milliseconds, as measured in trade monitor benchmarks involving windowed aggregations and event-time processing. The engine runs embedded within the application JVM, eliminating network hops for data access, and uses the Hazelcast IMDG for distributed state management, caching, and storage. Jet also provides a Management Center for monitoring job execution, cluster health, and performance metrics.

In the stream processing market, Jet competes directly with Apache Spark, Apache Flink, and Amazon Kinesis Data Analytics. Compared to Spark, Jet offers lower latency due to its in-memory, micro-batch-free architecture, but Spark has a larger ecosystem and broader adoption for batch ETL. Against Flink, Jet provides comparable streaming performance with a simpler deployment model (embedded vs. standalone), but Flink offers more mature event-time semantics and state management features. Amazon Kinesis Data Analytics is a fully managed service that abstracts cluster management, whereas Jet requires self-hosting or cloud VM provisioning. Jet's integration into Hazelcast Platform 5.0 in 2023 marked the end of Jet as a standalone product, consolidating its capabilities into a unified platform that also includes distributed caching, computation, and storage.

Key trade-offs include the product's end-of-life as an independent release—users now must adopt the broader Hazelcast Platform, which may introduce additional features and costs not needed for pure stream processing. The embedded deployment model, while low-latency, can complicate service testing and resource isolation compared to standalone frameworks. Documentation for feature implementation compliance is reported as poorly documented, and the codebase's heavy use of proxy patterns can reduce readability. The Management Center, while useful for monitoring, has experienced version compatibility issues. Pricing is subscription-based with annual or multi-year terms, varying by deployment environment and data scale, but no public tier pricing is disclosed.

Subscription-based licensing with annual or multi-year terms; pricing varies by deployment environment (on-premises or cloud), data scale, and feature set. No public tier pricing is disclosed; custom quotes are provided through Hazelcast sales.