5.3.0-BETA-1 Release Notes

New Features

  • Connector for Kafka Connect source: Import data from an external system directly into a Hazelcast data pipeline without the need of a Kafka cluster.

  • Partition-Aware SQL Client: Send the SQL commands only to the members having the relevant data, which reduces the network hops and improves the query performances.

  • Connector for MongoDB: Set up read-through/write-through caching in MongoDB via this connector without writing any code.

Breaking Changes

  • Renamed the DataLinkFactory interface as DataLink. #23748

  • Changed the default cloud coordinator URL from coordinator.hazelcast.cloud to api.viridian.hazelcast.com. The default configuration now connects to Hazelcast Viridian instead of Hazelcast Cloud. If you want to continue accessing your Hazelcast Cloud clusters, you need to set the hazelcast.client.cloud.url property to https://coordinator.hazelcast.cloud in your configuration. #23290



  • When advanced networking is enabled, the Kubernetes discovery plugin might have been discovering several endpoints (per each port) for each member’s pod. The discovery plugin now matches only the private IP per endpoint, ignoring the port values. #23766

  • Added support of IMDSv2 for Hazelcast’s AWS Discovery plugin. #23545

  • Enhanced the handling of errors coming from the Kubernetes API:

    • Not detecting the intent correctly whenever the cluster is shutdown, so the cluster was not restarted successfully.

    • Extensive amount of Kubernetes API calls were being logged when Automatic State Management feature is enabled.

  • Added the support of discovering ECS and EC2 members on AWS. See Deploying a Cluster on Amazon AWS. #22411

  • Enterprise Enhanced the experience of using Hazelcast with persistence under Kubernetes by removing requirements for manual interventions. Previously, when there is a restart on such a cluster, the cluster states were needed to be changed manually. For some cases, this might have led to data loss. Now, the shutdown process is automated. See Configuring Kubernetes for details. #21844


  • Added the Kafka connector to the distribution. #23312

  • Shaded dependencies for Hazelcast Platform are now combined in a dedicated package (com.hazelcast.shaded). #23124


  • Added socket options for per-socket keep-alive configuration - keep-count, keep-idle-seconds, and keep-interval-seconds. #23816

SQL Engine

  • Improved the nested loop joins algorithm by making the CreateDagVisitor interface generic. #23368

  • Implemented the TO_CHAR function which is capable of formatting dates, timestamps, and numbers. #22529

  • Implemented JSON=VARCHAR comparison. #22746

Jet Engine

  • Each Jet engine job was creating a client for connecting to the remote cluster, resulting in an excessive number of clients. This has been fixed by introducing a single data store which can be reused for all job instances. #23472, #23849

  • Added MongoDB read (source) and write (sink) processors. #23348

  • Added new messages (uploadJobMetaData and uploadJobMultipart) to the binary client protocol to be able to execute non-Java codes for streaming data pipelines, i.e., to submit Jet jobs. #23212

  • Introduced JobListener as an alternative to retrieve a job status via the Job.getStatus() method. #23193

  • Updated the job API to add the ability to distinguish the user-cancelled jobs from the failed ones. #22924

  • Added flock to guard all the concurrent pip executions (upgrading pip and protobuf versions) in the Jet-to-Python script. #22829


  • Added the TLS configuration property keyMaterialDuration which allows using a configurable timeout for caching keyStores and trustStores. See TLS Configuration for the details on how to configure this property. #23307


  • Hazelcast now allows serializing unsupported types via zero-configuration provided that there is an explicit serializer implemented for the unsupported type. #23449

  • Setting a generic record now fails if the value is not of the same type of generic record. #23522

  • Introduced clearer exception messages for Generic MapStore, instead of UndefinedErrorCodeException. #22765

WAN Replication

  • Added the REST endpoint /hazelcast/rest/wan/sync/progress which allows tracking the WAN synchronization progress. #22958


  • Added DataLink for Kafka consumers and producers, and for MongoDB #23886, #23837

  • Jet engine’s connector for MongoDB is now a part of the Hazelcast Platform distribution. #22565


  • Added name of the job to the job specific metrics. #23744

  • Added Tiered Storage compactor metrics to be exposed by the tools and interfaces such as Management Center, JMX, and diagnostics. #23643

  • Added a new tag to the CP Subsystem metrics to display the client names for CP sessions on Hazelcast Management Center. #23387

  • Added entry eviction and expiration counts to local map statistics and metrics. #23318

Other Enhancements

  • Introduced the hazelcast.discovery.public.address.fallback property to enable smart client connectivity to Hazelcast clusters deployed on Kubernetes, and having advanced network configuration enabled. #23707

  • External Data Store references have been renamed as Data Links. This includes the renaming of the external-data-store configuration element as data-link. #23597

  • Added the ability to test whether the connection with a JDBC data link is valid (testConnection method for JdbcDataLink). #23405

  • Upgraded Apache Calcite to 1.32.0. #22849


  • Fixed an issue where multiple Hazelcast members were able to use the same base directory for Tiered Storage, causing crashes. #5770

  • Fixed an issue where the members were crashing while they are loading data into the NATIVE memory and using JSON or Kryo serialization. #5384

  • Fixed an issue where the Hazelcast configuration was modified by the Jet engine, if the wildcard configuration is used and Jet is enabled; meaning the configurations made by the user was not being reflected. #23848

  • Fixed an issue where Hazelcast was failing to connect to a database when using Debezium, and the key of a record is null. #23768

  • Fixed an issue where Debezium connector for MongoDB could not keep the connection after some time. #23689

  • Fixed an issue where the unordered mapUsingServiceAsync was losing items when there are no watermarks. #23271

  • Fixed an issue where disabling metrics and JMX on a Hazelcast client was failing the members to start. #23360

  • Fixed an issue where a JDBC related exception message was mentioning data source reference instead of the table name. #23122

  • Fixed an issue where the beginning time of a job was being recorded after the job is submitted, which was causing a race condition. #23065

  • Fixed an issue where the hz-cli shell script was failing to run in bash on Windows systems since it uses a colon-separated CLASSPATH where each path is a Linux-style path. However, Java on Windows expects a semicolon-separated CLASSPATH where each path is a Windows-style path. As a consequence, the hz-cli could not be used from bash on Windows. #22884

  • Fixed a memory leak due to incomplete clean-up of backup replica sync operations. #22769

  • Fixed an issue where a map was loading old values with map.set() when map store offloading is enabled. #22602

  • Fixed an issue where a CP Subsystem object could be destroyed without a permission even the object has security and permission configurations enabled. #22475

  • Fixed an issue where external data store configurations could not be added dynamically. #22450

  • Fixed an issue where replication over WAN was failing on the source cluster members, when there are multiple batch publishers configured in a single WAN replication. #22437

  • Fixed an issue where the map entries recovered from persistence were not expiring after their time-to-live durations. #22279

  • Fixed an issue where Hazelcast 5.1.x members in a Spring Boot application could not establish a connection to each other when deployed in Kubernetes cluster with Istio Envoy Proxy enabled. #22256