note

Apache Cloudberry 2.2.0 has not been officially released yet. This post highlights several changes included in the current release candidates. Final contents may still change before the release is approved.

Apache Cloudberry (Incubating) 2.2.0 is the widest release the project has cut so far. Alongside the core database, this cycle publishes coordinated releases of Apache Cloudberry PXF, Apache Cloudberry Backup, and, for the first time, Apache Cloudberry Go Libs.

This article is a preview rather than an official release announcement, so it focuses on the changes users and developers are most likely to notice instead of a complete changelog.

Apache Cloudberry Core (apache/cloudberry)

Tracking upstream PostgreSQL: 14.4 to 14.9

The single most consequential change in 2.2.0 is not a feature. It is a change in posture.

Cloudberry now actively tracks upstream PostgreSQL 14 minor releases, absorbing the optimizations, fixes, and hardening that come with each one. The kernel has moved from PostgreSQL 14.4 to 14.9 in this release, and the work continues one minor version at a time until PostgreSQL 14 reaches the end of its upstream life.

This is an ongoing, reviewable effort rather than a one-time jump. If you want to help, the tracking board is open: https://github.com/orgs/apache/projects/572

Special thanks to our Cloudberry PPMC member @reshke for the sustained effort that made this possible.

Six new extensions

2.2.0 significantly expands what ships in the tree. Several extensions have been introduced by community members, adapted for Apache Cloudberry.

• diskquota (gpcontrib/diskquota) enforces disk usage limits on database objects. It supports quota limits per schema and per role within a database, implemented as a soft limit: a query targeting an over-quota schema or role is rejected before it starts, and a query that crosses the limit while running is cancelled. This is a legacy extension from the official Greenplum Database project, distributed under the PostgreSQL license, now brought forward to PostgreSQL 14 and Cloudberry. Enable it at build time with --with-diskquota.

• gp_stats_collector (gpcontrib/gp_stats_collector) collects query execution metrics and reports them to an external agent over a Unix domain socket. It captures the query lifecycle, EXPLAIN and EXPLAIN ANALYZE output, and instrument, system, network, interconnect, and spill metrics, all governed by gpsc.* GUCs. This is the extension the forthcoming Cloudberry Command Center consumes. Enable it with --with-gp-stats-collector.

• gp_relsizes_stats (gpcontrib/gp_relsizes_stats) calculates and stores statistics on file and table sizes and on space occupied on coordinator and segment host disks. A background worker collects on a configurable schedule, with per-database and per-file nap times so the load can be spread over time. It is built and installed by default.

• reject_partition_fullscan (gpcontrib/reject_partition_fullscan) rejects queries against partitioned tables that prune no partitions, which is a common way for a missing WHERE clause on the partition key to turn into a very expensive query. Two user-settable GUCs control it: reject_partition_fullscan and partition_fullscan_threshold. Both the PostgreSQL planner and ORCA dynamic scan paths are handled. It is built and installed by default; load it through shared_preload_libraries or LOAD to activate the planner hook.

• try_convert (contrib/try_convert) adds error-safe type casts, in the spirit of SQL Server's TRY_CAST. A failed cast returns a default value instead of aborting the statement. Casting from hstore and citext is supported via add_type_for_try_convert(). It is built and installed by default.

• yezzey (gpcontrib/yezzey) is included as a submodule at version 1.8.11, and offloads AO/AOCO table data to S3-compatible object storage for tiered storage workloads. Enable it with --with-yezzey.

Query processing and observability

• Interconnect statistics views. A new interconnect extension exposes cumulative counters from the UDPIFC interconnect protocol through three views: gp_interconnect_stats for the whole cluster, gp_interconnect_stats_per_segment, and gp_interconnect_stats_per_host. This makes interconnect behaviour observable with plain SQL rather than log scraping.
• Intra-segment parallel table scan in ORCA. ORCA can now generate worker-level parallel scans within a segment, with parallel-safety checks, cost model integration, and DXL serialization. Worker count follows max_parallel_workers_per_gather. This requires enable_parallel, which is off by default.
• Parallel Hash Full Join and Right Join. Two join shapes that previously forced serial execution across all segments can now run in parallel, with the distributed planner taught to carry worker counts through the HashedOJ locus a full outer join produces. Also gated on enable_parallel.
• AQUMV for multi-table joins. Answer-Query-Using-Materialized-Views now supports exact matches for multi-table join queries, including matview status maintenance. Enable it with enable_answer_query_using_materialized_views, which is off by default.

Beyond these, the cycle carries a long list of correctness fixes in ORCA, PAX storage, append-optimized vacuum, the UDP interconnect, pg_dump and pg_dumpall upgrade paths from Greenplum 5, 6, and 7, and cluster utilities such as gpexpand.

Platform coverage, packaging, and developer experience

Release engineering received as much attention as the kernel this cycle.

• Rocky Linux 10 and Ubuntu 24.04 are now supported; Ubuntu 20.04 is dropped. The supported platform set for 2.2.0 is Rocky Linux 8, 9, and 10 and Ubuntu 22.04 and 24.04. If you are still on Ubuntu 20.04, plan an upgrade before moving to 2.2.0.
• Wider CI coverage behind that. 2.1.0 gated on Rocky Linux 9 and Ubuntu 22.04; 2.2.0 builds and tests on all five supported platforms. New development images are published as apache/incubator-cloudberry:cbdb-build-rocky8-latest, apache/incubator-cloudberry:cbdb-build-rocky10-latest, and apache/incubator-cloudberry:cbdb-build-ubuntu24.04-latest.
• Two workflows instead of many. The per-OS test workflows have been consolidated into one matrix-driven Rocky Linux workflow and one matrix-driven Ubuntu workflow. Adding the next OS version is now a matrix entry rather than a new file, and every day-to-day change is validated across all supported versions.
• Automated DEB and RPM builds. A convenience package workflow lets the release manager produce packages for x86_64 and ARM64 across the supported operating systems, starting from a verified ASF source release tarball with signature and checksum validation built in.
• Relocatable RPMs with major-version coexistence. RPM packages can now be installed to a custom location with rpm --prefix, and the package name carries the major version, so multiple Cloudberry major versions can live side by side under one prefix.
• Python build dependencies on modern distributions. Building with --with-pythonsrc-ext now works where Python is externally managed, such as Ubuntu 24.04, and the packages can be pre-staged with make -C gpMgmt/bin download-python-deps. Build prerequisites changed for 2.2.0, so check the updated build documentation before you configure.
• macOS build portability. A series of fixes to PAX, the UDP interconnect, and shared-library linking make the source tree considerably friendlier to build on macOS for local development.

Security fixes

For anyone running Cloudberry in production, this is the most concrete reason to plan an upgrade. 2.2.0 carries 41 CVE fixes that 2.1.0 does not, covering privilege-escalation checks, memory-safety hardening across the parser, regex engine, text search and pgcrypto, injection paths in pg_dump and psql, and a path traversal in pg_basebackup and pg_rewind.

• 2026: CVE-2026-2003, CVE-2026-2004, CVE-2026-2005, CVE-2026-2006, CVE-2026-6464, CVE-2026-6470, CVE-2026-6472, CVE-2026-6473, CVE-2026-6474, CVE-2026-6475, CVE-2026-6477, CVE-2026-6478, CVE-2026-6479, CVE-2026-6637, CVE-2026-14662, CVE-2026-14663, CVE-2026-14664, CVE-2026-14666, CVE-2026-14668, CVE-2026-14669, CVE-2026-14677, CVE-2026-14678, CVE-2026-14679, CVE-2026-15741, CVE-2026-16239, CVE-2026-16241, CVE-2026-18408
• 2025: CVE-2025-1094, CVE-2025-4207, CVE-2025-8713, CVE-2025-8714, CVE-2025-8715, CVE-2025-12817, CVE-2025-12818
• 2024: CVE-2024-4317, CVE-2024-7348, CVE-2024-10976, CVE-2024-10977, CVE-2024-10979
• 2023 and earlier: CVE-2023-5870, CVE-2022-2625

Note the dates: most of these advisories postdate PostgreSQL 14.9 and arrived as targeted cherry-picks rather than through the version merge. That is the practical point of the kernel work described above. The project no longer has to wait for a full version upgrade to ship a security fix, and a few of the fixes from the PostgreSQL 14.5 to 14.9 releases had already landed in 2.1.0. The list above is the delta against 2.1.0, taken from the release candidate's commit history.

For the full picture, see the core branch comparison.

Apache Cloudberry PXF (apache/cloudberry-pxf)

See and control what PXF is doing

The headline addition is observability that DBAs have wanted for years. pxf_stat_activity shows what is running inside the PXF server, one row per active operation, with segment id, session id, command count, transaction id, operation, user, server, profile, schema, table, data source, and start time. Two companion functions, pxf_cancel_backend and pxf_interrupt_backend, let you stop a runaway external-table query without restarting PXF.

SELECT * FROM pxf_stat_activity;

Server-side logging is easier to correlate too: gp_session_id and gp_command_count are now added to the logging MDC, so PXF log lines can be tied back to the Cloudberry session that produced them.

Java, toolchain, and dependencies

• Java 17 support, with Gradle upgraded to 8.14.4. PXF now builds and passes its test suite on Java 8, 11, 17, and 21; CI's default build JDK is Java 11. Artifacts are still compiled to Java 8 bytecode, so existing Java 8 deployments keep running unchanged. PXF 3.0 is expected to raise the minimum to Java 11 as a step toward the Hadoop 3.x migration, with Java 17 discussed as a later target. That work is in review and does not affect 2.2.0, but 2.2.0 is the right release to plan your JVM upgrade on.
• Go toolchain moved to 1.25 for the PXF CLI.
• Log4j is at 2.25.4 and HBase at 2.5.15. Whether to drop HBase support in PXF 3.0 is currently under community discussion.
• Other dependencies including golang.org/x/crypto and tomcat-embed-core were refreshed to clear security advisories.

The PXF 3.0 direction, including the Java 11 minimum, Spring Boot and Hadoop library upgrades, and which legacy connectors to retire, is being planned in the open in cloudberry-pxf#131.

Testing and connectors

Connector testing has expanded substantially, with Testcontainers-based suites now covering ClickHouse, Oracle, and Microsoft SQL Server over JDBC, plus S3. CI gained Rocky Linux 9 and Rocky Linux 10 environments, and a convenience package build workflow now produces PXF packages the same way the core project does.

There are functional gains as well: Parquet UUID types can be read and written, and filter pushdown now works for numeric columns compared against integer constants.

Finally, PXF documentation has moved to the Cloudberry website and is now browsable at https://cloudberry.apache.org/pxf/

For more detail, see the PXF branch comparison.

Apache Cloudberry Backup (apache/cloudberry-backup)

Two new commands ship in 2.2.0.

gpbackman manages the backups that gpbackup creates, working directly against the gpbackup_history.db SQLite history database. It can display backup information and reports, delete a specific backup or every backup older than a time condition, from local storage or through storage plugins, clean deleted backups out of the history database, and synchronize the cluster history database to the standby coordinator either manually or automatically after a deletion. This release also adds database filters to its commands.

gpbackup_exporter is a Prometheus exporter for the same history database. It exposes backup status, deletion status, backup metadata, backup duration, and seconds since the last completed backup, listening on port 19854 by default with optional TLS and authentication.

Other changes worth knowing about:

• The gpbackup history database is now synchronized with the standby coordinator.
• gprestore --resize-cluster no longer fails when --jobs is greater than 1.
• The column permissions query during backup is faster.
• The Go toolchain moved to 1.25.
• A portable binary package workflow lets the release manager produce x86_64 and ARM64 packages quickly.

For the complete set of changes, see the Backup branch comparison.

Apache Cloudberry Go Libs (apache/cloudberry-go-libs)

This is the first release of cloudberry-go-libs as its own artifact. The motivation is downstream: components that depend on these libraries, WAL-G among them, need a released, versioned, ASF-compliant dependency rather than a moving branch.

The repository received the compliance work that an ASF release requires, its CI/CD was modernized onto GitHub Actions, the Go module path was renamed to github.com/apache/cloudberry-go-libs, and the PostgreSQL driver was migrated from pgx v4 to v5. Dependencies including golang.org/x/crypto and golang.org/x/net were upgraded to clear security advisories.

Looking Ahead

If there is a theme to 2.2.0, it is Apache Cloudberry continuing to advance as a PostgreSQL-based MPP database. The kernel now moves with upstream PostgreSQL, parallel execution reaches further into the segments, the extension surface grew by six, and the ecosystem components release together on one cadence with shared packaging machinery.

If the releases are approved, the project will share the official announcement and related materials through the website and community channels. Until then, please read this post as a preview of the current release candidates rather than final release notes.

We welcome everyone to continue following and participating in the Apache Cloudberry community to witness the 2.2.0 release:

• Visit our website: https://cloudberry.apache.org
• Follow us on GitHub: https://github.com/apache/cloudberry
• Join our Slack workspace: https://join.slack.com/t/asf-cloudberry/shared_invite/zt-3um34r7hf-Sh~6jG6hVxlQJo1tbhK2sw
• Subscribe to the mailing lists: https://cloudberry.apache.org/community/mailing-lists