Hazelcast: data infrastructure, deployment choices and edition boundaries

The financial-services material covers retail banking, payment processing and trade monitoring for risk and compliance. For a brokerage, the relevant role is an operational data layer behind customer applications or internal monitoring. The institution still designs the trading workflow, event interpretation and downstream response; storing and processing events does not make the platform a broker or execution venue.[2]

Strong consistency is a separate design choice within the platform. CPMap provides a minimal key-value interface with atomic operations and requires Enterprise Edition, an ADVANCED_CP entitlement and an Enterprise client. It is not simply a switch applied to every existing map. CPMap also needs explicit lifecycle and memory planning: unused instances are not automatically removed, and snapshots consume additional heap capacity.[14]

Hazelcast lists supported clients for Java, .NET, C++, Python, Node.js and Go. The language matrix separates ordinary data access from functions such as pipeline submission, transactions and administration. A client connection therefore establishes only part of compatibility; the team should match the exact application operation to the client version and feature support it needs.[9]

The Connector Hub includes Kafka streams, Kafka Connect sources, PostgreSQL and MySQL change capture, MongoDB, Redis, JDBC, HTTP, cloud storage and Pulsar. These are distinct integration paths: the JDBC driver lets a Java application query Hazelcast, whereas the JDBC connector moves data between a pipeline and an external relational system. Source, sink and application interfaces should be selected according to the direction of the data flow.[7]

Database change capture is based on Debezium and feeds changes into processing pipelines. This allows a financial application to maintain an operational view alongside an existing system of record. It also leaves the source database, change-log configuration and pipeline transformation as separate parts of the implementation, each of which needs ownership and monitoring.[8]

Cloud Standard measures more than the business records inserted into a map. Charged storage includes metadata, backups and custom code; indexes, stateful jobs and serialization can add memory usage. A capacity estimate based only on the raw source dataset can consequently understate billable consumption. The specific rate and any committed-credit arrangement need to be established for the selected service.[4]

The current edition documentation references both Apache 2.0 and the Hazelcast Community License. The latter permits specified use and distribution but excludes offering a competing online service using the covered software. This is a component and licence review, not an assumption that every file in a free distribution has identical commercial permissions.[13]

Advanced CP is a premium addition introduced with Platform 5.7 for stretched, multiple-datacentre deployments. The release announcement distinguishes it from baseline CP features that remain in the base Platform subscription. Procurement should therefore name the advanced features required, rather than assuming that every capability described under CP is included in any Enterprise order.[15]

The public Enterprise route offers a 30-day trial requested through a form. Moving to a paid installation still requires the appropriate software distribution and commercial agreement. Trial access is useful for a scoped technical evaluation, but it does not establish production capacity, renewal pricing or a migration commitment.[20]

Self-managed Platform can run in public or private cloud infrastructure. Hazelcast lists AWS, Azure, Google Cloud and Red Hat marketplace routes; the ability to apply existing cloud credits depends on the purchaser’s cloud-provider contract. Buying through a marketplace and buying a managed database service are separate choices, so the order should make operational responsibility explicit.[17]

Replica placement needs to match real failure boundaries. Default per-member grouping can leave several replicas on the same physical host if multiple members share it. Host-aware and zone-aware grouping place backups in different groups; zone-aware deployments should use equal member counts in each availability zone. The configured backup count alone is therefore insufficient evidence of resilience to a host or zone failure.[18]

Cloud Standard development clusters are limited to 0.5 GiB, one member and no persistent data or backups; restart or pause loses their contents, and they cannot be converted into production clusters. Production clusters provide daily disk backups and manual backups, and preserve data when paused. Both types can pause after 24 hours without a client connection. These differences should be reflected in test-data handling and production availability planning.[4]

Cloud Dedicated uses private endpoints and infrastructure reserved for the customer. Its documented controls include region selection, member memory capacity and automatic scaling. This provides a different isolation and connectivity arrangement from a public development endpoint, while still requiring an application network path and a topology that fits the institution’s residency requirements.[5]

For the inspected Platform 5.7.2 compatibility matrix, production JDK options include supported distributions of Java 17, 21 and 25. Windows 10/11 and macOS are described as development-only. Management Center, Operator and language clients have their own version requirements, so a working developer laptop is not a complete production compatibility assessment.[10]

Hazelcast offers instructor-led onsite training, online resources and implementation assistance through its services team. Scope should cover the data model, client integration, migration approach and operational handover. The public services page does not establish a standard deployment duration or a universal implementation fee.[19]

The support page advertises continuous coverage, a one-hour SLA, up to 30 email, instant-message and telephone contacts, and access to hot-fix patches. It also offers a Technical Account Management subscription. Detailed severity definitions, exclusions and support windows are linked separately; the one-hour headline should not be read as a guaranteed resolution time or a database-uptime percentage. The purchased agreement needs to establish the applicable commitment.[21]

Community and Enterprise maintenance differ. Hazelcast’s change notice places Community security fixes in major and minor releases from 5.4, while Enterprise receives patches when ready. Operator availability and CP support also changed around 5.5. Teams maintaining an older Community installation should review the current feature and patch policy rather than assuming that historical release behavior continues.[22]

Enterprise security includes TLS, mutual authentication, role-based permission policies and pluggable identity mechanisms. Hazelcast’s Security Suite describes JAAS-based integration and interceptors for additional checks. These controls must be configured for the relevant clients and operations; their availability in an edition is not evidence that a particular installation has enabled them.[24]

The security documentation highlights a separate trust boundary for stream processing. Jobs can deploy custom code, the Jet engine bypasses the access-control layer when using data structures in the same cluster, and SQL file connectors can access the cluster filesystem. Streaming is disabled by default in the JAR distribution for users primarily using storage. Enabling it requires explicit client authorization and a review of which code and connectors may run.[23]

Hazelcast’s 5.7 announcement expressly separates infrastructure resilience from governance, controls and people. That distinction matters for regulated financial use: a data platform can support an institution’s operational architecture, while its own regulatory obligations and evidence of effective controls remain with the institution.[15]

Hazelcast, Inc. is identified as the licensor in the Community License. The public contact page lists its US headquarters in Palo Alto, California, and an office in London. The legal entity and order documentation should be matched to the specific software or Cloud service being purchased.[13][25]

Hazelcast’s product history combines its in-memory data grid with a stream-processing engine. Current documentation refers to the former storage-engine name as Hazelcast IMDG and the streaming engine as Jet. These are part of the same platform family, rather than independent providers to evaluate as separate companies.[26][23]

Viridian is also a historical product name. The command-line release notes record its renaming to Hazelcast Cloud and preserve the old command as an alias. Older integration instructions may therefore use Viridian terminology even when current documentation uses Hazelcast Cloud.[27]

Operator

Hazelcast, Inc.

Service scope

Distributed data storage, caching, stream processing and operational application infrastructure.

Customer types

Financial institutions; Brokerage technology teams; Enterprise application developers

Delivery

On premises; SaaS; Hybrid; Managed service

Pricing publication

Public and quoted pricing

Headquarters

Palo Alto, California, United States

Platforms

Hazelcast Platform; Hazelcast Cloud; Management Center; Platform Operator

Integrations

Java; Python; .NET; C++; Node.js; Go; Kafka; PostgreSQL; MySQL; Redis; JDBC; Debezium

Pricing scope

Free Community Edition; Enterprise subscription and managed Cloud terms require the selected commercial configuration.

Implementation

Data and topology design, client and connector integration, recovery planning and version-specific deployment.

Support

Commercial support and optional technical account management; detailed commitments depend on the purchased agreement.

Permissions and regulatory scope

Technology for operational data workflows; financial institutions retain their governance and regulated responsibilities.

Provider metadata checked 21 Oct 2026Source 8Source 9Source 3Source 17Source 22Source 13Source 26Source 25Source 2Source 7Source 6Source 1Source 11Source 27Source 23Source 19Source 21

The Platform 5.7 release announcement places Vector Collections at release-candidate status and advises deferring critical production adoption until general availability. The current vector documentation still links to a Beta design discussion and requires an explicit collection configuration. Treat the feature as requiring a version-specific availability check, rather than applying the production status of the core platform to every newer component.[15][16]

Upgrade planning must distinguish the cluster from its processing jobs. Enterprise rolling upgrades support specified patch and adjacent-minor transitions, while Community-to-Enterprise changes use a rolling restart. The compatibility documentation preserves job state within a minor-version line but requires jobs to be resubmitted after a minor upgrade. An upgrade can therefore preserve the storage cluster while still requiring a planned pipeline transition.[28][10]