Release Notes

Hazelcast Platform is a major release which unifies the former Hazelcast IMDG and Jet products in a single solution. The changes are summarized below.

For information about upgrading from previous IMDG and Jet releases, see the Upgrading to Hazelcast Platform section.

To learn about the changes in previous IMDG and Jet releases, see IMDG release notes and Jet release notes.

New Features

  • Jet Streaming Engine: Hazelcast now incorporates the streaming engine, which was formerly and separately known as Hazelcast Jet.

  • SQL Engine: We have added the support of following operators to our SQL engine:

  • Persistence: This allows individual members and whole clusters to recover faster by persisting map entries and JCache data on disk. Members can use persisted data to recover during restarts. See the Persistence section.

  • Compact Serialization Format (Preview Feature): We have introduced a new serialization format (CompactSerializer), which is in its beta stage currently. This format requires much less storage space, provides better latency and throughput, and supports more extensive type sets:

    • It supports schema evolution of a class and does not require any version to be specified explicitly.

    • On common use cases where the field types of an object are natively supported via this compact format, Java DTO classes are automatically serialized without requiring a serializer (CompactSerializer) to be implemented.

    • Your classes do not need to implement an interface like Portable. Once you enable the compact serialization, the classes not matching to any other serialization methods are serialized in this new format.

  • Security for Jet Engine Jobs: The Jet engine allows you to upload custom code as well as access data outside of Hazelcast such as files on the device that’s running a member. You can secure your Hazelcast member against malicious uses of the Jet API. See the Job Security section.

Breaking Changes

API Changes

  • For Google Cloud Storage (GCS) file sources, the SecuredFunction interface now returns a list of permissions instead of a single permission. Previously, Hazelcast was returning only the permission for the bucket to be read, and this was causing a source to be possibly exploited for gaining information whether some file exists in the system. Now, Hazelcast also returns a permission for the keyfile that is required to read a bucket on GCS.

  • The "Hot Restart Persistence" feature name has been changed to "Persistence"; with this, we’ve performed the following minor changes in its API:

    • Renamed com.hazelcast.hotrestart.BackupTask as com.hazelcast.persistence.BackupTask.

    • Renamed com.hazelcast.hotrestart.BackupTaskState as com.hazelcast.persistence.BackupTaskState.

    • Added the isBackupEnabled() to PersistenceService/HotRestartService that does not have a default implementation.

  • After the unification of cloud discovery plugins (see here), the exception class for the failure in generating SSLSocketFactory is changed from KubernetesClientException to RestClientException for the Kubernetes plugin. #19132

  • Changed the type of YEAR field in the SQL client protocol from short to int. #18984

  • Since the HOST and RACK metadata were deprecated in the previous Hazelcast release, these information are removed from the ZONE_AWARE partition grouping configuration. #18780

  • Since DAG printing was not reflecting the real queue size, the DAG.toDotString(int defaultParallelism) method signature is changed as DAG.toDotString(int defaultLocalParallelism, int defaultQueueSize). You, now, need to supply the queue size that will be shown if it is not overriden on the edge. #18475

Configuration Changes

  • Removed the following properties that were used in the former Jet product, since it is now part of Hazelcast Platform:

    • jet.home

    • jet.imdg.version.mismatch.check.disabled



  • Hazelcast distribution is now available in two versions: full and slim. While the full one includes Hazelcast and all of its modules along with SQL, Jet engine extensions and Management Center, the slim one only includes the Hazelcast runtime, default configurations and scripts. See here for details. #18990, #18989

  • Consolidated the scripts delivered with the former IMDG and Jet distributions. See the Changes in Distribution Packaging section. #18999

  • The former hazelcast-sql, hazelcast-sql-core and hazelcast-jet-sql Maven modules in the distributions have been merged into a single hazelcast-sql module as a part of the Hazelcast Platform distribution. #18584

  • Removed the jackson-core dependency from pom.xml since it was breaking the extensions that depend on jackson. #18665

  • Moved the former Hazelcast Jet code samples into the Hazelcast code samples repository. #475

The Jet and SQL Engine

  • Added support of rolling upgrades for SQL. #19026

  • Added support of reading data from High-Density Memory Store backed maps via the SQL engine. #19328

  • Added support of regular and index (migration tolerant) scan of High-Density Memory Store via SQL. #19227

  • Tables were called mapping in information_schema.mappings and table in information_schema.columns. This inconsistency has been fixed by renaming mapping as table. #19210

  • Added support of the CONCAT_WS function, which takes variable sized VARCHAR (minimum 3) and returns a VARCHAR that consists of the concatenation of the arguments except the first one using the first argument as a separator. #19094

  • The 'SELECT' statement now also supports queries without the FROM clause so that you can submit queries like SELECT rand() without this clause. #19030

  • The Jet engine jobs submitted in a Hazelcast cluster are now cancelled when you upgrade your Hazelcast version since the Jet engine doesn’t provide backwards compatibility. #19012

  • Implemented the partition-tolerant index scan processor for Hazelcast maps: during a partition migration, this processor searches all the migrated partitions on all available cluster members. #18968

  • Added support of the putIfAbsentAsync() method for maps on the member side; which is required for the usage of INSERT INTO statements in SQL queries. #18946

  • Added support of returning nested fields without having to deserialize them, which enables you to use Portable in client/server deployments without touching the server side; for example, SQL queries can now return columns without having the class on the server-side classpath. #18922

  • Standardized the TIME and TIMESTAMP temporal formats for the SQL engine: You can now use TIME without leading zeroes and TIMESTAMP with space instead of the T symbol. Also added support of leading non-zero characters for the DATE formats. #18881, #18842

  • Added support of OFFSET for SQL queries. #18866

  • Implemented IdentifiedDataSerializable for SQL schema objects. #18851

  • Changed the since tags in Jet engine API and its extension modules from @since x.y to @since Jet x.y. #18832

  • Implemented the OnHeapMapScanP class to read the Hazelcast maps directly by the SQL engine. #18685

  • Implemented a basic memory management for the SQL engine so that number of records accumulated by it can be limited to avoid out of memory failures. You can use the max-processor-accumulated-records configuration element for this purpose. #18671

  • Added support of dynamic parameters for the SQL engine and file table functions. #18613, #18522

  • Introduced QueryDataType.MAP and QueryDataTypeFamily.MAP to support map operand checks for file table functions. #18602

  • Added support of EXTRACT(field FROM source) for the SQL engine. The function computes date parts from the source field. The supported types for source argument are as follow:

    • Date

    • Time

    • Timestamp

    • Timestamp With Time Zone

  • Added support of the LIMIT <n> and ORDER BY clauses for the streaming engine. #18479

  • Implemented the following functions for the SQL engine:

  • Added support of plan caching for Jet engine based queries. #18446

  • Added support of plus and minus operations for interval types (date, time, etc.) for the SQL engine. #18390

  • Added support of various new Portable types for the SQL engine. #18115

  • Added support of IN and BETWEEN operators for the SQL queries. #18483, #18422, #18067

Data Structures

  • The previous Replicated Map implementation was iterating all the values while calculating the size of map; this was causing latencies and performance issues as the entries in a Replicated Map grows. The related size() method has been refactored to eliminate the aforementioned situation. #19005

Cloud Discovery Plugins

  • In Kubernetes, Hazelcast resolves its public addresses by finding an individual service that points to the given Hazelcast pod. If there are multiple services pointing to one pod, then the discovery could not work or might have chosen the wrong service. The following changes have been made to address this:

    • Added label-based filtering for the Kubernetes Service per pod.

    • Added matching service and pod by name (if there are multiple services per pod is configured, the priority takes a service with the same name as the pod, before it was a random service.

    • Added resolving load balancer service if "hostname" is defined.

  • The code of the AWS, Azure, Kubernetes and GCP discovery plugins' in their own Github repos have been moved into the hazelcast/hazelcast repo. Their documentation also has been merged and unified into Hazelcast documentation. #19132

  • Added Kubernetes plugin’s configuration file for role based access control into the hazelcast/hazelcast Github repository as kubernetes-rbac.yaml. #19093


  • Added support of default serializers for the following classes which has been necessary for non-Java clients to use these:

    • LocalDate

    • LocalTime

    • LocalDateTime

    • OffsetDatetime


  • Added an example Hazelcast configuration file (hazelcast-security-hardened.yaml) focused on hardened security to the distribution packages; it lists configuration options with their descriptions which may help securing your Hazelcast deployment. #18843

  • Introduced the simple authentication configuration; it allows to have users and their assigned roles stored together with other Hazelcast configurations. See the example:

        enabled: true
          - name: simpleRealm
                  - username: test
                    password: 'a1234'
                      - monitor
                      - hazelcast
                  - username: root
                    password: 'secret'
                      - admin


  • Unless you explicitly disable them, the Merkle Trees are now enabled automatically when your cluster has a map or cache whose persistence is enabled; this is to improve a single member recovery from a crash, and it does not have a high memory overhead. #19502

  • Added the expose-externally configuration parameter for objects that expose an external IP address In. Kubernetes. See Configuring Kubernetes for its description.

  • The properties provided in former JetProperties are now merged into ClusterProperty. Also added the hazelcast prefix to the former Jet property names, e.g., jet.job.scan.period has become hazelcast.jet.job.scan.period and the former one is deprecated. #19146

  • Added a configuration option to enable/disable resource uploading for Jet engine jobs. See here for details.

  • Even when the factory configuration is missing on the member but the map is configured to have the the in-memory format as OBJECT, Hazelcast now can store portables as PortableGenericRecord and still query them without needing to convert them to Object/Data. #18891

  • Introduced the following properties:

    • hazelcast.partition.rebalance.mode: It determines whether cluster membership change triggers partition rebalancing automatically (auto) or explicit action is required for rebalancing to occur (manual). Its default is auto.

    • hazelcast.partition.rebalance.delay.seconds: it specifies the time in seconds to wait before triggering automatic partition rebalancing after a member leaves the cluster unexpectedly. Unexpectedly in this context means that a member leaves the cluster by programmatic termination, a process crash or network partition. Its default is 0, which means rebalancing is triggered immediately.

Other Enhancements

  • Introduced a warning when the users create a job JAR to submit via hz-cli and Hazelcast detects hazelcast or hazelcast-enterprise is packaged in it. #19512

  • Added queueFillPercent metric to show how full the WAN replication queue is, in percentage. #19431

  • The README of hazelcast/hazelcast GitHub repository has been completely rewritten to reflect the unification of former Hazelcast IMDG and Jet products. #19061

  • The hazelcast-sql module is now covered by the Hazelcast Community License; before, it was Apache License, Version 2. #18957

  • Added the merkle tree support for caches to speed up the migration process during a cluster rebalancing. #18898

  • Added the client console entry point to the Hazelcast command line interface; you can now use the hazelcast console command to start the client console application. #18857

  • Enhanced the getPartitionGroupStrategy() method to have cluster members as arguments so that useful partitioning strategies can be implemented by accessing the members using this method. #18794

  • The log message for infinite cluster connection timeout is clearer now. Previously, it was represented as the value of Long.MAX_VALUE. #18642

  • Introduced a new mechanism in the background expiration tasks; now a thread local array controls the allocations for these tasks otherwise which may cause increased garbage collection pressure and CPU usage spikes when you use aggressive expiration configurations, e.g., low time-to-live values. #18633

  • The license key is, now, not shown while starting a member on Docker with overriding configurations. #18568

  • Limited the number of parallel partition reads (to a fixed value of five) for maps and caches to prevent out of memory failures. #18499

  • Added a comprehensive documentation for metrics produced by Hazelcast. See List of Metrics for the full list of metrics with their descriptions. #17880

  • Improved the speed of connection by a member when it joins the cluster, by removing the unnecessary sleep statements in the code. #17428


  • Fixed a possible serialization error while submitting a Jet engine job when the member and client sides have different Hazelcast versions with compatible APIs. #19534

  • Fixed an issue where running the map.clear/cache.clear methods was evicting all entries in all Near Caches of all maps, not only in the requested map/cache. #19523

  • Fixed an issue where the wildcard configuration mechanism was not working correctly if a matching pattern has the same prefix and suffix. #19357

  • If the connector permission for file includes wildcard (*) then any file in the system could be read by using .. in the path in connector. See the below example:

    <connector-permission name="file:/home/user/workspace/*" principal="role1">

    Then one can read a file like readFrom(FileSources.files("/home/user/workspace/../some_secure_file"). This has been fixed by converting the file path to canonical path for file permissions.

  • If two clusters with different cluster names run locally and both of them has enabled security, then a Hazelcast client was ignoring the configured cluster names and connecting to any of them; a check has been put to eliminate this issue. #19344

  • Fixed an issue where a high amount of garbage collection pressure was occurring during repartitioning especially when having a high partition count. #19312

  • Fixed an issue where the MultiMap operation statistics were not being updated after these operations are called from client. #19296

  • Fixed an issue where the hz-cli submit script was not working properly with relative path: if the script is called from a different directory (like ./bin/hz-cli), the bin directory was taken as root for the relative path instead of the directory from where the script is called. #19204

  • Fixed an issue where ElasticSearch did not have a client method that allows HTTPS connections; added a new client with HTTP and HTTPS schemes. #19139

  • SQL expressions now does not fail when used with trailing semicolons. 18976

  • Fixed an issue where the health monitor was incorrectly showing the value for free metadata memory. #18951

  • Some merge policies like LatestUpdateMergePolicy for the map and WAN replication configurations require the per-entry statistics to be enabled. Previously, this configuration inconsistency was causing the related member to fail at runtime. Now, the Hazelcast member fails to start, i.e., fast fails, in such a case. #18928

  • Fixed an issue where the maximum size policy for a map was being ignored when the policy is PER_NODE and the cluster is scaled down (due to losing or killing a member). #18927

  • The LRU eviction policy now takes last access time value into account to prevent premature removal of the lately added but not yet accessed map entries. #18909

  • Fixed an issue where the map’s Near Cache was setting its maximum size as 10.000 even if the configured eviction policy is NONE. #18835

  • Fixed a regression issue where a job using map reader/writer could not be completed when the target map has a configured Near Cache. #18696

  • Fixed an issue where the updates made to a persistent map store might be lost when the write coalescing is enabled. #18686

  • Fixed a reconnection flood when members are separated by a proxy: When a member is disconnected from the cluster, the alive cluster members still try to reconnect to it if the dying member connection is not closed explicitly. In the cases where the connection is explicitly closed with a cause (such as Connection reset by peer or Remote socket closed!), a new connection was being established if the member is placed behind a proxy. This scenario was end causing opening and closing connections continuously. This issue has been fixed. #18673

  • Fixed an issue where the multicast discovery was not working between the members when the loopback mode is enabled. #18669

  • The HazelcastInstance.shutdown() method now gracefully terminate Jet engine jobs, too. After the merge of IMDG and Jet, it was failing. #18625

  • Replicated Map does not fail to publish events anymore, from an entry listener with a predicate which has an attribute path. #18623

  • Fixed a possible performance regression by not starting the cooperative threads until a job is submitted; otherwise the Jet engine was consuming system resources. #18574

  • Fixed an issue where running SQL statements was fetching results incorrectly (from an unexpected mapping) when there are different user-provided schemas for data structures and mappings. #18428

  • Fixed an issue where the client state listener was not properly working with failover clients (in blue-green deployments); it was failing with invalid configuration exception. #18351

  • Fixed an issue where there might be continuous reconnection attempts by the cluster members to a failed member, even its connection is explicitly closed and when Hazelcast is placed behind a proxy. #18320

  • Hazelcast now properly works on hosts with multiple NICs. #17834

Removed/Deprecated Features

  • The following properties have been deprecated:

    • hazelcast.client.statistics.enabled

    • hazelcast.client.statistics.period.seconds

  • The HotRestartService class has been deprecated; you can use PersistenceService instead. #19404

  • The following property have been removed:


  • Former Jet, JetInstance and JetCacheManager classes have been deprecated. See here for details. Accordingly JetInstance has been removed from Hazelcast’s command line interface (CLI) and Jet engine tests (also the name of CLI has been changed to HazelcastCommandLine). #18829, #18775, #18667

  • Former Hazelcast Jet’s bootstrappedInstance() has been deprecated. Instead, you can use Hazelcast.bootstrappedInstance(). See here for details.

  • The support of NULLS FIRST and NULLS LAST has been removed from the SQL engine; the indices treat NULL as the smallest value in ordering, therefore we needed to disable temporarily these constructs. #19031

  • The configuration element hot-restart-persistence has been deprecated. You can use persistence instead, which is the successor of hot-restart-persistence. If both are enabled, Hazelcast uses the persistence configuration. The hot-restart-persistence element will be removed in a future release. #19004

  • The hazelcast-all module has been removed from the Hazelcast distribution after the merge of former IMDG and Jet products.


We would like to thank the contributors from our open source community who worked on this release: