Hazelcast Jet

Rich Set of Transforms and Connectors

Jet supports a rich set of transforms such as joins, aggregations, sliding and tumbling windows and complex event processing. Use it to transform data streams and write the output via connectors such as Apache Kafka, relational databases, message brokers, Debezium, Hadoop, Amazon S3 among others.

Massively Parallel with Predictable Latency

Jet's core execution engine was designed for high throughput with low system overhead and latency. It uses a fixed-size thread pool to run any number of parallel tasks. The engine is based on coroutines that implement suspendable computation, allowing many of them to run concurrently on a single thread.

Elastic and Fault-tolerant

You can scale your cluster up and down without losing the computational state without touching disk. Support for at-least-once and exactly-once processing means you never miss an event when processing incoming data.

Runs Everywhere

Jet is a single 12 MB Java archive (JAR). It’s lightweight enough to run on small devices, you can embed it into your application as just another dependency or deploy it as a standalone cluster. First-class support for Kubernetes is included.

Co-locate Data and Compute

You can store data in memory inside a Jet cluster, allowing your pipeline to access data locally and make use of partition affinity, dramatically speeding up your computation and eliminating latency.