.svg)
.png)
Enterprise data platforms have to do two things at once. They have to keep adding modern capabilities like new compute engines, decoupled architectures, and expanded high availability. They also have to keep the existing stack secure, current, and maintainable. Neither can wait.
That is the focus of this trio of releases. ODP 3.3.6.4-1 is a major capability release. It introduces six new platform integrations: Apache Spark 4.1.1, Apache Celeborn, Spark Connect, Apache Superset, and the next-generation NiFi 2.x and Ozone 2.x lines. It also adds new high availability for Trino Gateway, Schema Registry, NiFi Registry, and Multi-Standby NameNode operations. ODP 3.2.3.6-2 and ODP 3.2.3.6-3 are maintenance releases on the 3.2.3 line, delivering one of our largest security-hardening updates to date and modernizing core dependencies across more than 30 components.
Across all three releases, 1,443 CVEs were resolved.
Acceldata Open Data Platform (ODP) is an enterprise-ready open data platform built on a secure, flexible foundation. It brings together core open-source technologies so teams can standardize and modernize big data environments across on-premises and hybrid deployments.
At a glance
ODP 3.3.6.4-1, major capability release
- Apache Spark 4.1.1 as the latest engine, alongside Spark 3.3.3, 3.5.1, and 3.5.5
- Spark Connect, a decoupled client-server architecture for remote Spark development (3.5.1, 3.5.5, 4.1.1)
- Apache Celeborn 0.6.2, a high-performance Remote Shuffle Service for Spark, Flink, Tez, and MapReduce
- Apache Superset 6.0.0, modern open-source BI and visualization
- Apache NiFi 2.7.2, the next-generation dataflow engine (1.28.1 retained)
- Apache Ozone 2.1.0, the next-generation object storage line (1.4.1 retained)
- New Spark 4 Hive Warehouse Connector and Spark 4 HBase Connector
- New HBase Thrift Server, Multi-Standby NameNode support, and extended Oozie/Spark3 integration
- High availability for Trino Gateway, Schema Registry, and NiFi Registry
- Kafka Tiered Storage with S3, plus a new Kafka AWS Sink Connector
- Expanded JupyterHub with multi-kernel and multi-Spark support
- 550 CVEs resolved (71 Critical, 327 High, 135 Medium, 17 Low)
ODP 3.2.3.6-2 and 3.2.3.6-3, maintenance on the 3.2.3 line
Both releases ship together and deliver the same platform-wide improvements. ODP 3.2.3.6-2 is the Python 2 variant. ODP 3.2.3.6-3 is the Python 3 variant.
- 893 CVEs resolved across 30+ components (180 Critical, 492 High, 215 Medium, 6 Low)
- Oozie Spark3 support with Spark 3.5.5 as the default runtime
- Netty unified to 4.1.132.Final platform-wide
- Modernized core dependencies: log4j2 (2.25.3), Jackson, Avro, Protobuf, Guava, and commons-lang3
- Targeted bug fixes across Hive, Ambari, Oozie, NiFi, Airflow, and more
Highlights
A modernized Spark stack: Spark 4.1.1, Spark Connect, and Apache Celeborn
Apache Spark sits at the center of much of what runs on ODP, which is why upgrading the Spark layer is always a high-stakes change. Big-bang migrations break things. Falling behind has a real cost. ODP 3.3.6.4-1 avoids both problems.
Apache Spark 4.1.1 joins ODP with significant improvements across Streaming, SQL, PySpark, connectivity, data sources, custom functions, and developer usability. It ships alongside Spark 3.3.3, 3.5.1, and 3.5.5, all fully supported. Existing jobs keep running. New workloads adopt 4.x on your timeline.
Spark Connect introduces a decoupled client-server architecture for Spark applications. PySpark and Scala clients connect remotely to a Spark server, debug interactively, and upgrade independently of the cluster. This removes the tight coupling that has slowed Spark application development for years. It is supported across Spark 3.5.1, 3.5.5, and 4.1.1.
Apache Celeborn 0.6.2 lands as a first-class component. It is a high-performance Remote Shuffle Service with native integration for Spark, Flink, Tez, and MapReduce, providing faster and more reliable shuffle for large jobs with substantially lower storage overhead.
For Spark 4 workloads, the release ships first-class connectors out of the box. The Spark 4 Hive Warehouse Connector provides unified access to Hive-managed tables, and the Spark 4 HBase Connector provides direct read/write integration with HBase. A new HBase Thrift Server enables cross-language client access (Python, Ruby, C++, PHP, and others) to HBase data.
What you gain
- Latest Spark engine with significant Streaming, SQL, and PySpark improvements
- A non-disruptive path to Spark 4, with old and new versions deployable side by side
- Faster, more reliable shuffle for large jobs, with lower storage overhead
- Decoupled, more productive Spark application development through Spark Connect
- First-class connectivity for Spark 4 workloads to existing Hive and HBase data
New platform integrations: Apache Superset 6.0.0, NiFi 2.7.2, and Ozone 2.1.0
Modern data services rarely ship together. Teams piece BI together from one vendor, dataflow from another, and object storage from a third. ODP 3.3.6.4-1 closes those gaps with three new capability lines.
Apache Superset 6.0.0 brings modern open-source BI and visualization natively into Ambari-managed ODP, with full lifecycle support for installation, configuration, connectors, and administration. Teams running Superset separately, or paying for proprietary BI, can bring it under the same governance as the rest of the platform.
Apache NiFi 2.7.2 introduces the NiFi 2.x line, with a modernized UI, Python-based processors, parameter providers, and significant performance and security improvements over 1.x. NiFi 1.28.1 remains available so existing dataflows can migrate on their own timeline.
Apache Ozone 2.1.0 brings the Ozone 2.x line, offering improved scalability for billions of objects, enhanced S3 gateway compatibility, and stronger multi-tenancy and security controls. Ozone 1.4.1 remains available.
What you gain
- Modern open-source BI natively in your platform, with no separate stack to manage
- Next-generation dataflow with NiFi 2.7.2, on your own migration timeline
- More scalable, more secure object storage with Ozone 2.1.0
- Optionality to choose which versions to deploy based on workload and readiness
JupyterHub: a unified, multi-kernel notebook workspace
JupyterHub in ODP 3.3.6.4-1 becomes a more capable, governed notebook environment. It now supports multi-kernel notebooks (PySpark, Scala, R, SQL), works across all four supported Spark versions in the same workspace (3.3.3, 3.5.1, 3.5.5, and 4.1.1, with PySpark only on 4.1.1), and provides a consistent governance, access-control, and observability model with the rest of the platform.
For teams that have been juggling fragmented notebook environments across data engineering and analytics, this consolidates the workflow into a single managed surface, without sacrificing the flexibility of choosing the right kernel and Spark version per workload.
What you gain
- One workspace covering data engineering and interactive analytics
- Multi-Spark and multi-kernel flexibility for diverse workloads
- Governance and access control consistent with the rest of the platform
Expanded high availability and lower Kafka storage cost
Production teams know the services that can take down a cluster. Trino Gateway, Schema Registry, and NiFi Registry were single points of failure in ODP until this release. They now support high availability natively. Multi-Standby NameNode also strengthens resilience for clusters with federated namespaces. Query execution, Kafka schema management, and versioned dataflow configuration all become more resilient by default.
On cost, Kafka storage tends to grow faster than budgets. Kafka Tiered Storage with S3 offloads Kafka log segments to S3-compatible object storage, dramatically reducing broker storage costs for long-retention use cases, with topic-level configuration and validation native to Ambari. The new Kafka AWS Sink Connector also lets teams stream Kafka data directly into AWS services.
What you gain
- Production-grade HA for query execution, Kafka schemas, and versioned NiFi flows
- Stronger resilience for clusters with federated NameNodes
- Significant Kafka storage cost reduction with tiered S3 storage
- Native streaming integration with AWS sinks
Modernizing the 3.2.3 line: ODP 3.2.3.6-2 and 3.2.3.6-3
Not every team is ready to move to a different release line, and security, dependency hygiene, and modern runtimes should not have to wait. That is what the 3.2.3.6 releases deliver.
Both releases ship together. ODP 3.2.3.6-2 is the Python 2 variant, ODP 3.2.3.6-3 is the Python 3 variant, and they deliver the same platform-wide improvements. Three things stand out:
- Oozie now supports Spark3 with Spark 3.5.5 as the default runtime, so Oozie-orchestrated workflows can move to modern Spark without re-platforming. The integration is extended further in ODP 3.3.6.4-1.
- Netty unified to 4.1.132.Final platform-wide, alongside modernization of log4j2 (2.25.3), Jackson, Avro, Protobuf, Guava, and commons-lang3. This closes CVE exposure and eliminates version-skew across cluster components.
- 893 CVEs resolved across 30+ components (180 Critical, 492 High, 215 Medium, 6 Low), with targeted bug fixes in Hive, Ambari, Oozie, NiFi, Airflow, and more.
What you gain
- Modern Spark performance for Oozie-orchestrated workloads
- A single, current Netty version across every component
- Reduced technical debt and dependency-related upgrade surprises
- Fewer production issues in heavily used components
Component coverage at a glance
A summary of what is included in the 3.2.3.6 maintenance releases, component by component:
Stronger security across the platform: 1,443 CVE fixes
CVE pressure on modern data platforms doesn't let up. Every quarter, the Apache components that make up the stack ship dozens of patches that have to be tracked, validated, and rolled into a coherent release. This cycle, we've gone heavy.
1,443 CVEs resolved across the three releases, including 251 Critical and 819 High severity, with the heaviest concentration of fixes in Hive, Spark 3, Oozie, and ClickHouse.
For teams running ODP in production, that's a lot to land in a single deployment cycle. New capabilities for where you're heading, a hardened existing line for what's already running, and a stronger security baseline across the board.
What you gain
- A materially stronger security posture across the entire platform
- A faster path through compliance reviews and CVE audits
- Reduced exposure across both the stable maintenance line and the new capability line
- Confidence that your open-source stack is being patched in lockstep, not piecemeal
Learn more
Release notes: ODP 3.3.6.4-1, ODP 3.2.3.6-2, ODP 3.2.3.6-3. See the full release notes for the complete list of enhancements, upgrades, bug fixes, and CVE updates.
Learn more about Open Data Platform (ODP).
Book a free consultation to see how ODP fits your environment and helps with your modernization roadmap.
.webp)
.webp)
