Hazelcast

What is Hazelcast

The world’s leading companies trust Hazelcast to modernize applications and take instant action on data in motion to create new revenue streams, mitigate risk, and operate more efficiently. Businesses use Hazelcast’s unified real-time data platform to process streaming data, enrich it with historical context and take instant action with standard or ML/AI-driven automation - before it is stored in a database or data lake.

Hazelcast is named in the Gartner Market Guide to Event Stream Processing and a leader in the GigaOm Radar Report for Streaming Data Platforms. To join our community of CXOs, architects and developers at brands such as Lowe’s, HSBC, JPMorgan Chase, Volvo, New York Life, and others, visit hazelcast.com.

When to use Hazelcast

Hazelcast provides a platform that can handle multiple types of workloads for building real-time applications.

Stateful data processing over streaming data or data at rest
Querying streaming and batch data sources directly using SQL
Ingesting data through a library of connectors and serving it using low-latency SQL queries
Pushing updates to applications on events
Low-latency queue-based or pub-sub messaging
Fast access to contextual and transactional data via caching patterns such as read/write-through and write-behind
Distributed coordination for microservices
Replicating data from one region to another or between data centers in the same region

Key Features

Stateful and fault-tolerant data processing and querying over data streams and data at rest using SQL or dataflow API
A comprehensive library of connectors such as Kafka, Hadoop, S3, RDBMS, JMS and many more
Distributed messaging using pub-sub and queues
Distributed, partitioned, queryable key-value store with event listeners, which can also be used to store contextual data for enriching event streams with low latency
A production-ready Raft-implementation which allows lineralizable (CP) concurrency primitives such as distributed locks.
Tight integration for deploying machine learning models with Python to a data processing pipeline
Cloud-native, run everywhere architecture
Zero-downtime operations with rolling upgrades
At-least-once and exactly-once processing guarantees for stream processing pipelines
Data replication between data centers and geographic regions using WAN
Microsecond performance for key-value point lookups and pub-sub
Unique data processing architecture results in 99.99% latency of under 10ms for streaming queries with millions of events per second.
Client libraries in Java, Python, Node.js, .NET, C++ and Go

Operational Data Store

Hazelcast provides distributed in-memory data structures which are partitioned, replicated and queryable. One of the main use cases for Hazelcast is for storing a working set of data for fast querying and access.

The main data structure underlying Hazelcast, called IMap, is a key-value store which has a rich set of features, including:

Integration with data sources for one time or continuous ingestion
Read-through and write-through caching patterns
Indexing and querying through SQL
Processing entries in place for atomic updates
Expiring items automatically based on certain criteria like TTL or last access time
Near cache for caching entries on the client
Listeners for pushing changes to clients
Data Replication between datacenters (Enterprise version only)
Persistence of data on disk (Enterprise version only)

Hazelcast stores data in partitions, which are distributed to all the nodes. You can increase the storage capacity by adding additional nodes, and if one of the nodes go down, the data is restored automatically from the backup replicas.

You can interact with maps using SQL or a programming language client of your choice. You can create and interact with a map as follows:

CREATE MAPPING myMap (name varchar EXTERNAL NAME "__key", age INT EXTERNAL NAME "this") 
TYPE IMap
OPTIONS ('keyFormat'='varchar','valueFormat'='int');
INSERT INTO myMap VALUES('Jake', 29);
SELECT * FROM myMap;

The same can be done programmatically as follows using one of the supported programming languages. Here are some exmaples in Java and Python:

var hz = HazelcastClient.newHazelcastClient();
IMap<String, Integer> map = hz.getMap("myMap");
map.set(Alice, 25);

import hazelcast

client = hazelcast.HazelcastClient()
my_map = client.get_map("myMap")
age = my_map.get("Alice").result()

Other programming languages supported are C#, C++, Node.js and Go.

Alternatively, you can ingest data directly from the many sources supported using SQL:

CREATE MAPPING csv_ages (name VARCHAR, age INT)
TYPE File
OPTIONS ('format'='csv',
    'path'='/data', 'glob'='data.csv');
SINK INTO myMap
SELECT name, age FROM csv_ages;

Hazelcast also provides additional data structures such as ReplicatedMap, Set, MultiMap and List. For a full list, refer to the distributed data structures section of the docs.

Stateful Data Processing

Hazelcast has a built-in data processing engine called Jet. Jet can be used to build both streaming and batch data pipelines that are elastic. You can use it to process large volumes of real-time events or huge batches of static datasets. To give a sense of scale, a single node of Hazelcast has been proven to aggregate 10 million events per second with latency under 10 milliseconds. A cluster of Hazelcast nodes can process billion events per second.

An application which aggregates millions of sensor readings per second with 10-millisecond resolution from Kafka looks like the following:

var hz = Hazelcast.bootstrappedInstance();

var p = Pipeline.create();

p.readFrom(KafkaSources.<String, Reading>kafka(kafkaProperties, "sensors"))
 .withTimestamps(event -> event.getValue().timestamp(), 10) // use event timestamp, allowed lag in ms
 .groupingKey(reading -> reading.sensorId())
 .window(sliding(1_000, 10)) // sliding window of 1s by 10ms
 .aggregate(averagingDouble(reading -> reading.temperature()))
 .writeTo(Sinks.logger());

hz.getJet().newJob(p).join();

Use the following command to deploy the application to the server:

bin/hazelcast submit analyze-sensors.jar

Jet supports advanced streaming features such as exactly-once processing and watermarks.

Data Processing using SQL

Jet also powers the SQL engine in Hazelcast which can execute both streaming and batch queries. Internally, all SQL queries are converted to Jet jobs.

CREATE MAPPING trades (
    id BIGINT,
    ticker VARCHAR,
    price DECIMAL,
    amount BIGINT)
TYPE Kafka
OPTIONS (
    'valueFormat' = 'json',
    'bootstrap.servers' = 'kafka:9092'
);
SELECT ticker, ROUND(price * 100) AS price_cents, amount
  FROM trades
  WHERE price * amount > 100;
+------------+----------------------+-------------------+
|ticker      |           price_cents|             amount|
+------------+----------------------+-------------------+
|EFGH        |                  1400|                 20|

Messaging

Hazelcast provides lightweight options for adding messaging to your application. The two main constructs for messaging are topics and queues.

Topics

Topics provide a publish-subscribe pattern where each message is fanned out to multiple subscribers. See the examples below in Java and Python:

var hz = Hazelcast.bootstrappedInstance();
ITopic<String> topic = hz.getTopic("my_topic");
topic.addMessageListener(msg -> System.out.println(msg));
topic.publish("message");

topic = client.get_topic("my_topic")

def handle_message(msg):
    print("Received message %s"  % msg.message)
topic.add_listener(on_message=handle_message)
topic.publish("my-message")

For examples in other languages, please refer to the docs.

Queues

Queues provide FIFO-semantics and you can add items from one client and remove from another. See the examples below in Java and Python:

var client = Hazelcast.newHazelcastClient();
IQueue<String> queue = client.getQueue("my_queue");
queue.put("new-item")

import hazelcast

client = hazelcast.HazelcastClient()
q = client.get_queue("my_queue")
my_item = q.take().result()
print("Received item %s" % my_item)

For examples in other languages, please refer to the docs.

Get Started

Follow the Getting Started Guide to install and start using Hazelcast.

Documentation

Read the documentation for in-depth details about how to install Hazelcast and an overview of the features.

Get Help

You can use the following channels for getting help with Hazelcast:

Hazelcast mailing list
Slack for chatting with the development team and other Hazelcast users.
Stack Overflow

How to Contribute

Thanks for your interest in contributing! The easiest way is to just send a pull request. Have a look at the issues marked as good first issue for some guidance.

Building From Source

Building Hazelcast requires at minimum JDK 11. Pull the latest source from the repository and use Maven install (or package) to build:

$ git pull origin master
$ ./mvnw clean package -DskipTests

It is recommended to use the included Maven wrapper script. It is also possible to use local Maven distribution with the same version that is used in the Maven wrapper script.

Additionally, there is a quick build activated by setting the -Dquick system property that skips tests, checkstyle validation, javadoc and source plugins and does not build extensions and distribution modules.

Testing

Take into account that the default build executes thousands of tests which may take a considerable amount of time. Hazelcast has 3 testing profiles:

Default: Type ./mvnw test to run quick/integration tests (those can be run in parallel without using network by using -P parallelTest profile).
Slow Tests: Type ./mvnw test -P nightly-build to run tests that are either slow or cannot be run in parallel.
All Tests: Type ./mvnw test -P all-tests to run all tests serially using network.

Some tests require Docker to run. Set -Dhazelcast.disable.docker.tests system property to ignore them.

When developing a PR it is sufficient to run your new tests and some related subset of tests locally. Our PR builder will take care of running the full test suite.

Trigger Phrases in the Pull Request Conversation

When you create a pull request (PR), it must pass a build-and-test procedure. Maintainers will be notified about your PR, and they can trigger the build using special comments. These are the phrases you may see used in the comments on your PR:

run-lab-run - run the default PR builder
run-lts-compilers - compiles the sources with JDK 11 and JDK 17 (without running tests)
run-ee-compile - compile hazelcast-enterprise with this PR
run-ee-tests - run tests from hazelcast-enterprise with this PR
run-windows - run the tests on a Windows machine (HighFive is not supported here)
run-with-jdk17 - run the tests with JDK 17
run-with-ibm-jdk-8 - run the tests with IBM JDK 8
run-cdc-debezium-tests - run all tests in the extensions/cdc-debezium module
run-cdc-mysql-tests - run all tests in the extensions/cdc-mysql module
run-cdc-postgres-tests - run all tests in the extensions/cdc-postgres module
run-mongodb-tests - run all tests in the extensions/mongodb module
run-s3-tests - run all tests in the extensions/s3 module
run-nightly-tests - run nightly (slow) tests. WARNING: Use with care as this is a resource consuming task.
run-ee-nightly-tests - run nightly (slow) tests from hazelcast-enterprise. WARNING: Use with care as this is a resource consuming task.
run-sql-only - run default tests in hazelcast-sql, hazelcast-distribution, and extensions/mapstore modules
run-docs-only - do not run any tests, check that only files with .md, .adoc or .txt suffix are added in the PR
run-sonar - run SonarCloud analysis
run-arm64 - run the tests on arm64 machine

Where not indicated, the builds run on a Linux machine with Oracle JDK 11.

Creating PRs for Hazelcast SQL

When creating a PR with changes located in the hazelcast-sql module and nowhere else, you can label your PR with SQL-only. This will change the standard PR builder to one that will only run tests related to SQL (see run-sql-only above), which will significantly shorten the build time vs. the default PR builder. NOTE: this job will fail if you've made changes anywhere other than hazelcast-sql.

Creating PRs which contain only documentation

When creating a PR which changes only documentation (files with suffix .md or .adoc) it makes no sense to run tests. For that case the label docs-only can be used. The job will fail in case you've made other changes than in .md, .adoc or .txt files.

License

Source code in this repository is covered by one of two licenses:

The default license throughout the repository is Apache License 2.0 unless the header specifies another license.

Acknowledgments

Thanks to YourKit for supporting open source software by providing us a free license for their Java profiler.

We owe (the good parts of) our CLI tool's user experience to picocli.

Copyright

Visit www.hazelcast.com for more info.

zhmaeff/hazelcast