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

Cloudberry​

Apache Cloudberry is an advanced and mature open-source Massively Parallel Processing (MPP) database, derived from the last open-source version of the Greenplum Database® but built on a more modern PostgreSQL kernel and with more advanced enterprise capabilities. Cloudberry can serve as a data warehouse and can also be used for large-scale analytics and AI/ML workloads.

Cloudberry has been incubating since 2024-10-11.

Three most important unfinished issues to address before graduating:​

1. Grow the contributor and community to ensure long-term sustainability.
2. Publish a few more Apache releases following the ASF release processes.

How has the community developed since the last report?​

• The community continued its public bi-weekly meetings, holding six meetings from May through July 2026. Discussions covered the PostgreSQL 16 kernel upgrade, release coordination, CI and packaging, ecosystem projects, documentation, community governance, and open technical topics.
• New Committers: Anton Kurochkin (@woblerr) and Liu Shengsong (@lss602726449) were announced as new committers on June 26 and July 6, respectively.
• Contributors and community participation continued to grow through work on the main repository and related ecosystem projects. The community also discussed new Kubernetes operators and the creation of an Apache Cloudberry Command Center repository.
• Community governance and responsible AI use continued to receive attention. The community discussed the AI guideline and the use of AI-assisted code review tools, with AI-assisted code review tools, such as GitHub Copilot, were discussed as optional review suggestions subject to normal community review.
• Events and outreach:
  • Community Over Code Asia 2026: Seven Cloudberry-related talks, a Cloudberry booth: https://s.apache.org/knmjs
  • Saint Highload 2026: our PPMC member Leonid Borchuk presented a session on PAX: https://s.apache.org/re82o

How has the project developed since the last report?​

• PostgreSQL kernel upgrade and core development:
  • Merged the PostgreSQL 16 kernel upgrade into the main branch with 5730 commits. The remaining FIXME items are planned to be addressed as part of the future 3.0 release.
  • Started discussion about a future kernel upgrade from PostgreSQL 16 to PostgreSQL 18.
  • Continued PostgreSQL 14.x maintenance on the 2.x stable branch, including work to backport 14.8 and archive-related fixes.
  • Added security fixes for several CVEs and continued dependency and security maintenance across the repositories.
  • New extensions:
    • Added the gp_relsizes_stats extension.
    • Introduced the yezzey extension for the 2.x stable branch as a submodule.
    • Added the TRY_CONVERT extension.
    • Added the reject_partition_fullscan extension.
  • Continued ORCA and planner improvements, including fixes for CTE handling, intra-segment parallel table scans, HAVING and grouping-set handling, non-ASCII column aliases, empty-partition statistics, outer-join predicate pushdown, and other correctness and stability issues.
• datalake_fdw: Started implementing Iceberg support in the datalake_fdw.
• Ecosystem and sub-repositories:
  • cloudberry-pxf:
    • Continued the 2.2 release plan.
    • Added Java 21 support, upgraded the Gradle wrapper, updated HBase and Log4j dependencies, improved S3-related Testcontainers coverage, and added session and command-count information to logging context.
    • Applied security and dependency updates and restored green CI testing.
  • cloudberry-backup:
    • Completed the Go 1.25 upgrade and continued PostgreSQL 16 support work.
    • Pinned the CI branch to Cloudberry 2.x stable branch and applied dependency updates to address security alerts.
    • Added the gpbackup_exporter Prometheus exporter for the backup history database.
  • cloudberry-go-libs:
    • Added a new CI/CD workflow and completed the ASF compliance work. The project is ready for the 2.2 release.
  • Other ecosystem work:
    • WAL-G added support for PAX incremental backups.
    • Odyssey 1.5.1 was released, and Odyssey support was documented on the Cloudberry website.
    • Community members developed Kubernetes operators for Cloudberry.
    • Yandex Cloud plans to donate the Command Center project to Apache Cloudberry - https://s.apache.org/urvsc
• Release, CI, and packaging:
  • Release:
    • Provided the convenience packages for 2.1 release.
    • Started preparing Apache Cloudberry (Incubating) 2.2.0. By the end of July, the release was considered ready to enter the release process, with remaining work including cherry-picking changes to REL_2_STABLE, and finalizing PXF Rocky Linux 10 support.
    • Created the release runbook wiki page for the release manager.
  • CI workflows:
    • Added Rocky Linux 10 workflow support.
    • Completed the CI consolidation work from OS-specific CI workflows into matrix-based workflows for easier maintenance and future platform expansion.
    • Bumped the Go version to 1.25 in the development Docker images and CI workflows.
    • Planned to add a macOS CI workflow.
  • Packaging:
    • Improved RPM packaging for the core repository, including custom installation prefixes and support for installing multiple major versions side by side.
    • Merged the source-release-based convenience packages workflow for the main repository, cloudberry-backup and cloudberry-pxf for the coming 2.2+ releases.
• Website and documentation:
  • Upgraded the website to Docusaurus 3.10.2 and continued UI improvements.
  • Added the “Powered by” page and integrated cloudberry-pxf documentation into the main website.
  • Added a project history page and ecosystem documentation for Odyssey and WAL-G.
  • Added a new page on backward-incompatible changes from Greenplum 6.x.
  • Refreshed the deployment guide.
  • Published five new blog posts to share the latest community news.

How would you assess the podling's maturity?​

• Initial setup
• Working towards first release
• Community building
• Nearing graduation
• Other:

Date of last release:​

• April 14, 2026 - Apache Cloudberry (Incubating) 2.1.0

When were the last committers or PPMC members elected?​

• June 22, 2026 - Anton Kurochkin (woblerr) was announced as a new committer.
• July 6, 2026 - Liu Shengsong (@lss602726449) was announced as a new committer.

Join Us​

Apache Cloudberry follows the principle of open and transparent governance. You can follow our quarterly incubation reports to stay updated on key community events and project progress.

Get involved with the community:

• GitHub: github.com/apache/cloudberry
• Slack: Join our Slack workspace
• Discord: Join our Discord server
• Mailing Lists: lists.apache.org/list.html?dev@cloudberry.apache.org
• Website: cloudberry.apache.org

We're excited to welcome you to Community Over Code Asia 2026 — the premier open-source event in Asia, hosted by the Apache Software Foundation. Taking place in Beijing from August 7–9, 2026, this in-person gathering is your chance to dive into the latest innovations, learn from domain experts, and connect with the global open-source community.

Apache Cloudberry will be featured across seven sessions, spanning topics from new releases and query optimization to community building and network performance. In addition, we are hosting a booth and an in-person community gathering during the conference — there are plenty of opportunities to meet the team and fellow community members!

Featured Sessions​

Here is the list of Apache Cloudberry related sessions accepted for Community Over Code Asia 2026:

Congratulations to all speakers — we can't wait to see your talks!

Visit the Cloudberry Booth​

The Apache Cloudberry community will be onsite with a dedicated booth. Drop by to meet the team, ask questions, check out live demos, and grab some awesome Cloudberry swag!

Community Gathering​

We are also organizing an in-person Apache Cloudberry community gathering during the conference. The discussion will focus on Cloudberry's incubation, graduation, and community development. If you have topics you'd like to discuss, we'd love to hear from you — come join the conversation!

How to Participate​

Ready to join? Register now to attend Community Over Code Asia 2026 in person. We look forward to seeing you in Beijing!

Liu Shengsong has consistently been one of the long-term contributors to the Cloudberry community, delivering critical code enhancements for key capabilities including pg_upgrade, external table, gp_expand, and unique index. In the recently concluded PostgreSQL 15 and 16 merge initiative, he served as a core contributor, implementing the MERGE command, contributing critical fixes for ORCA, and resolving numerous parallel test cases — his exceptional code quality and efficiency played a pivotal role in the project's success. His consistent engagement with the community makes him a natural fit for the committer role.

Becoming a committer recognizes both Liu Shengsong's past contributions and the community's trust in his continued involvement. Committers play an important role in shaping the project by reviewing and merging contributions, participating in design decisions, and helping guide the project's growth.

Please join us in congratulating Liu Shengsong on this well-deserved recognition. We look forward to his continued contributions and collaboration within the Apache Cloudberry community!

You can find the official announcement on the Apache Cloudberry mailing list.

Over the past several months, Anton has been an active and impactful contributor to the Cloudberry community, with a particular focus on the cloudberry-backup project and related Go-based tooling. His contributions span a wide range of areas — from roadmap discussions and code improvements to project modernization and community collaboration.

Some highlights of Anton's contributions include:

• Engaging deeply in the cloudberry-backup roadmap discussions, sharing practical operational experience and helping shape the project's direction.
• Contributing gpbackman, a CLI tool for managing gpbackup backups more conveniently.
• Contributing gpbackup_exporter, a Prometheus exporter that collects metrics from gpbackup_history.db.
• Modernizing the Go development infrastructure, including upgrading the Go toolchain from 1.21 to 1.24, updating golangci-lint, and refactoring deprecated syntax and usages across the project.

Anton's work has meaningfully strengthened Cloudberry's backup ecosystem and Go tooling capabilities. His combination of technical depth, practical experience, and community spirit makes him a natural fit for the committer role.

Becoming a committer recognizes both Anton's past contributions and the community's trust in his continued involvement. Committers play an important role in shaping the project by reviewing and merging contributions, participating in design decisions, and helping guide the project's growth.

Please join us in congratulating Anton Kurochkin on this well-deserved recognition. We look forward to his continued contributions and collaboration within the Apache Cloudberry community!

You can find the official announcement on the Apache Cloudberry mailing list.

Apache Cloudberry users can now leverage PostgreSQL 16 features—including the MERGE command, SQL/JSON constructors, parallel VACUUM, and a rewritten statistics subsystem—directly in their MPP workloads. This upgrade spans two major PostgreSQL upstream releases (PostgreSQL 15 and 16) and incorporates over 5,730 upstream commits. It marks Cloudberry’s second cross-version kernel upgrade since the project was derived from Greenplum Database and began its independent evolution (the first being the leap from PostgreSQL 12.x — the kernel carried by Greenplum Database 7.0 — to 14.4). It is also the first major kernel upgrade driven by a broad group of community contributors since Cloudberry joined the Apache Incubator.

Why this upgrade matters​

The PostgreSQL database kernel is the foundation Cloudberry is built on. Tracking the PostgreSQL upstream release cadence is both an engineering necessity and a responsibility to our ecosystem.

• PostgreSQL 14.x is approaching end of life. Following the PostgreSQL community’s release lifecycle, PostgreSQL 14 is scheduled to reach end of support in November 2026. Completing the major-version upgrade in advance avoids a future where Cloudberry users have to choose between security patches and stability.
• Faster access to PostgreSQL innovation. Features like pg_stat_io, Memoize, and LZ4 WAL compression—originally developed by the upstream community—now flow into Cloudberry without years of backporting effort. Staying aligned with PostgreSQL upstream lets new features, planner improvements, and ecosystem tooling reach Cloudberry users naturally.

By staying aligned with a modern PostgreSQL releases, Cloudberry ensures users run their MPP workloads on an actively maintained PostgreSQL kernel — one that delivers a decade of upstream improvements without the delays and risks of backporting.

What this upgrade brings​

PostgreSQL 15 and 16 together accumulated a large body of improvements. To make this extensive list of updates easy to navigate, we have grouped the key changes into four primary categories for users and administrators, followed by developer-specific internal changes:

1. Modern SQL Capabilities & Query Standards​

Upgrading to PostgreSQL 16 brings full support for the SQL-standard MERGE command, robust SQL/JSON constructors, and advanced database collations to Cloudberry:

• MERGE Statement: Full support for the SQL-standard MERGE command, allowing INSERT / UPDATE / DELETE to be expressed in a single statement. In Cloudberry’s distributed environment, this significantly simplifies ETL/ELT pipelines, replacing complex client-side workarounds and reducing coordinator-segment round-trips. Accompanying fixes cover partitioned tables, cross-partition triggers, RLS, EvalPlanQual, self-modified tuples, and pg_stat_statements integration.
• SQL/JSON Enhancements: New IS JSON predicate and standard JSON constructor functions (json_array, json_object, json_scalar, json_serialize), plus auto-generated query jumbling via gen_node_support.pl and stability fixes such as the json_array() subquery double-transformation issue.
• Built-in ICU Support: ICU support is now built by default, bringing superior multilingual sorting and localization out-of-the-box. initdb uses uloc_getDefault(), and CREATE DATABASE LOCALE applies across all collation providers.
• SQL Types & Functions:
  • Numeric scale can now be negative or greater than precision.
  • New ANY_VALUE aggregate.
  • unnest(multirange) and range_agg with multirange inputs.
  • pg_size_pretty / pg_size_bytes extended to petabyte scale.
  • ALTER TABLE SET ACCESS METHOD for switching table storage backends.
  • New SYSTEM_USER function and a numeric_mul() overflow fix.

2. Performance & Resource Optimization​

Tracking upstream PostgreSQL releases introduces critical performance, query planning, and storage enhancements designed to scale with your MPP workloads:

• Parallel VACUUM & Autovacuum:
  • Parallel VACUUM (PostgreSQL 16, with the new vacuumparallel.c) helps reclaim space faster, allowing segment nodes to execute vacuuming operations in parallel to reduce disk I/O bottlenecks and locking overhead.
  • ON COMMIT DELETE ROWS no longer errors out after relhassubclass=f.
  • VACUUM end-of-truncation phase uses WaitLatch() instead of pg_usleep().
  • Numeric overflow fix in parallel numeric aggregates.
  • ANALYZE preserves relhasindex for partitioned tables.
• Optimizer & Executor Improvements:
  • Hash-table acceleration for NOT IN (values).
  • Datum sorts for single-column sorts.
  • Result Cache renamed to Memoize, with continued performance tuning.
  • Parallel Hash Full Join to speed up complex distributed queries.
  • Lower planning cost for deeply nested views, and fewer pallocs when building partition bounds.
  • Outer join optimization enhancements and symmetric hash join improvements (PostgreSQL 16).
• WAL & Storage Subsystem:
  • LZ4 compression for full-page writes (FPW) to reduce WAL volume and speed up replication under write-heavy workloads.
  • Custom WAL resource managers so extensions can register their own WAL record types.
  • Direct I/O on macOS, and skipping WAL recycling/preallocation during archive recovery.
  • Buffer manager refactored for faster relation extension, with new helpers like smgrzeroextend(), FileZero(), and FileFallocate().
  • WAL/Recovery code split into xlogrecovery.c and xlogprefetcher.c, plus a new backup sink architecture (PostgreSQL 15).
  • To guarantee data integrity, the HeapBitmapScan skip_fetch optimization, which had correctness concerns, has been removed.
• Critical Correctness Planner/Executor Fixes:
  • Recognition of multiple hashable ScalarArrayOpExprs in the planner.
  • Consistent whole-row Var construction between parsing and planning.
  • WindowAgg evaluation crash.
  • WITH RECURSIVE UNION assertion failure.
  • ORDER BY / DISTINCT aggregates with FILTER.
  • Missing expression processing on prune steps in setrefs.c.
  • Extreme skew detection in Parallel Hash Join.

3. Data Sync & Ecosystem Connectivity​

This upgrade enhances the reliability and performance of logical replication, backup/restore workflows, and client-side protocols:

• Major Logical Replication Improvements:
  • Built-in support for two-phase commit (2PC) on logical replication, which can be enabled directly through the replication protocol. This is vital for maintaining transactional consistency across distributed segments in an MPP layout.
  • Parallel apply workers (PostgreSQL 16), row/column-level filtering and schema-level publications (PostgreSQL 15) significantly improve replication throughput and flexibility.
  • New predefined role pg_create_subscription for finer-grained permission management.
  • A series of correctness and robustness fixes, including a critical data-loss issue, slot invalidation, snapshot handling, and memory leaks.
• Backup, Restore, and Upgrade Tooling:
  • pg_dump: generic compression API, with zstd and LZ4 frame-only formats.
  • pg_basebackup: extended compression options, server-side compression (including -Fp), client-side LZ4 decompression, and parallel zstd compression.
  • pg_upgrade: fixes for an unintentional 'NULL' string literal and cross-version XMLSERIALIZE(NO INDENT) handling; pg_dumpall correctly handles dangling OIDs in pg_auth_members.
• Client Protocol & libpq:
  • New PQsendFlushRequest and a fix for the pipeline-mode state machine.
  • SASL code refactored into a generic interface.
  • Escape function fixes for invalid encoding data, including security-relevant changes.
  • Length handling fixes in PQescapeLiteral() / PQescapeIdentifier().
  • Build-time check that libpq does not call exit() or abort().
• Operator & psql Experience:
  • \watch honors PSQL_WATCH_PAGER.
  • New \drg command for inspecting role grants.
  • \copy from now sends data in larger chunks.
  • pg_waldump --rmgr accepts multiple specifications.

4. Enterprise Observability, Security & Stability​

Upgrading the core kernel to PostgreSQL 16 brings a modern, shared-memory statistics subsystem, security guardrails, and key robustness fixes:

• Observability & the pgstat Rewrite:
  • Shared-Memory pgstat Rewrite (PostgreSQL 15): The dedicated stats collector process is replaced with shared memory stats. This eliminates a potential performance bottleneck under high concurrent workloads. The monolithic pgstat.c was split into 14 files under utils/activity/, making the database kernel cleaner and easier to maintain.
  • New pg_stat_io View (PostgreSQL 16): Detailed I/O statistics including shared buffer hits, I/O times, and writebacks.
  • pg_stat_statements Upgrades: JIT counters, temp file block I/O timing, normalized utility query strings, MERGE support, and a 32-bit integer overflow fix.
  • Other improvements: SP-GiST index scans counted in pgstat, enhanced pg_stat_reset_single_table_counters, and new test helpers pg_stat_force_next_flush() and pg_stat_have_stats().
• Security and Privilege Model:
  • CREATEROLE privilege restrictions: A substantial improvement to the role/privilege model that closes long-standing privilege escalation paths.
  • New GUCs: createrole_self_grant, reserved_connections.
  • New predefined roles: pg_maintain, pg_create_subscription.
  • Non-superuser predefined roles for vacuum / analyze workflows.
  • PUBLIC CREATE is revoked from the public schema, which is now owned by pg_database_owner.
  • Security Invoker Views (SECURITY INVOKER).
  • Fixes around session_authorization interactions and a scram_SaltedPassword() integer overflow.
• Partitioned Table Improvements:
  • Self-referencing foreign keys on partitioned tables.
  • Correct handling of detach with top-level FKs.
  • relhassubclass reset on ATTACH; consistent trigger renames; firing state preserved on clone.
  • Partitionwise join / grouping is disabled when collations do not match, preventing incorrect results.
• Critical Stability and Data Correctness Fixes:
  • Fixed data corruption on relation truncation failure.
  • Page header is no longer corrupted when an update lands on a non-LP_NORMAL TID; the operation now fails cleanly.
  • Unsafe BufferDescriptors access.
  • GB18030 SIGSEGV from out-of-bounds read.
  • Snowball stemmer null pointer dereference after OOM.
  • Rare standby assertion failure on primary restart.
  • Catcache invalidation during list entry construction.

5. Under the Hood: Updates for Kernel Contributors​

If you are a contributor working on the Cloudberry database engine, this upgrade introduces several codebase updates, cleaner abstraction layers, and automated testing tools:

• Pervasive Storage Refactor: RelFileNode ➔ RelFileLocator rename. This touches hundreds of files, aligning Cloudberry's internal naming conventions with modern PostgreSQL.
• Configuration Engine Decoupling: GUC system split: guc.c ➔ guc.c + guc_tables.c + guc_funcs.c.
• Header Optimization: varatt.h separated from postgres.h.
• Modernized Testing Cluster: The test suite transitioned from PostgresNode.pm to the new PostgreSQL::Test::Cluster framework, improving testing speed and reliability.
• Parser Automation: gen_node_support.pl auto-generates copyfuncs / equalfuncs / outfuncs / readfuncs and query jumbling code, with a NodeTag ABI stability check.

What this means for Cloudberry users​

• A modern PostgreSQL foundation. Cloudberry users can now benefit directly from features accumulated across PostgreSQL 15 and 16, including MERGE, SQL/JSON, two-phase logical replication, Memoize, Parallel Hash Full Join, and pg_stat_io.
• Better observability and operability. The shared-memory pgstat rewrite, the new I/O view, and pg_stat_statements enhancements give operators and DBAs a richer toolbox for performance analysis.
• A stronger security posture. CREATEROLE tightening, new predefined roles, and security invoker views align Cloudberry’s privilege model more closely with modern production practices.
• Smoother ecosystem integration. Staying close to PostgreSQL upstream means Cloudberry can interoperate more naturally with PostgreSQL extensions, tools, and clients.

What comes next​

Landing this work on the main branch is a milestone, not the finish line.

• Try it out. Some issues uncovered during the upgrade (tracked on the community project board) are not yet fully resolved; the community is actively working on them and aims to address them as quickly as possible. We warmly invite users and developers to exercise the new kernel in test environments and share feedback.
• Continued optimization. Expect ongoing work on MPP-side integration with the new kernel, performance regression fixes, and further tuning of existing capabilities on top of PostgreSQL 16.
• Community collaboration. Issues, pull requests, and real-world usage reports all help Cloudberry move forward. If you would like to contribute, the door is open.

Acknowledgments​

This kernel upgrade was a collaborative effort across the community. Special thanks to the contributors who drove the merge:

@chenjinbao1989, @lss602726449, @reshke, @gfphoenix78, @yjhjstz, @tuhaihe

And to everyone who participated in code reviews, testing, and issue reporting throughout the year-long effort.

Join us​

Apache Cloudberry is a community-driven open-source project, and we welcome contributors of all backgrounds—whether you are a database veteran or just getting started with MPP systems.

• Website: https://cloudberry.apache.org
• Repository: https://github.com/apache/cloudberry
• Discussions: https://github.com/apache/cloudberry/discussions
• Issue tracker: https://github.com/apache/cloudberry/issues
• Mailing list: https://lists.apache.org/list.html?dev@cloudberry.apache.org
• Slack workspace: https://join.slack.com/t/asf-cloudberry/shared_invite/zt-3um34r7hf-Sh~6jG6hVxlQJo1tbhK2sw
• Discord server: Join our Discord server

Cloudberry​

Cloudberry is an advanced and mature open-source Massively Parallel Processing (MPP) database, derived from the open-source version of Pivotal Greenplum Database®️ but built on a more modern PostgreSQL 14 kernel, whereas Greenplum is based on PostgreSQL 12. This upgrade brings enhanced enterprise capabilities, making Cloudberry well-suited for data warehousing, large-scale analytics, and AI/ML workloads.

Cloudberry has been incubating since 2024-10-11.

Three most important unfinished issues to address before graduating:​

1. Grow the contributor and community to ensure long-term sustainability.
2. Publish a few more Apache releases following the ASF release processes.

How has the community developed since the last report?​

• The community continued the Apache Cloudberry bi-weekly community meeting, holding five public meetings from February to April 2026. These meetings covered release coordination, kernel upgrades, ecosystem projects, testing strategy, events, and open community topics.
• New Committers: Xiaoyu Liu (@misterRaindrop), Nikolay Antonov (@ostinru), Robert Mu (@robertmu)
• New PPMC Member: Leonid Borchuk
• Community channels:
  • Created a new Slack workspace and a new Discord server after the previous Slack workspace became unavailable.
  • Updated the website and planned updates across related repositories to point users to the current communication channels.
• Governance and community process:
  • Started discussion on an AI conduct policy to guide responsible use of LLM tools while following ASF copyright and attribution expectations.
  • Added a website page documenting how to invite new PPMC members.
• Events and outreach:
  • Community Over Code Asia 2026: around 11 Cloudberry-related proposals, now waiting for the review results. Plan to organize one in-person community gathering in Beijing during the conference.
  • A Cloudberry meetup was organized in Moscow, with public recordings and a Russian Habr article bringing more user feedback to the community.
  • Two community members were confirmed as PGConf.dev 2026 speakers.
• New contributors appeared in community discussions and PRs, including contributions around Cloudberry Backup, PXF, and core repository features.

How has the project developed since the last report?​

• Released Apache Cloudberry (Incubating) 2.1.0 in April 2026, the project's second Apache release after entering the ASF Incubator. Convenience RPM/DEB packages and binary packages were provided for the first time. Manual packages were made for 2.1.0, and automation is being considered for 2.2.0.
• PostgreSQL kernel upgrade:
  • Started the PostgreSQL 14.x minor-version upgrade work for the 2.x release branch. The branch progressed from PostgreSQL 14.4 through 14.8, with later 14.x updates being tracked in the project Kanban.
  • Continued the PostgreSQL 16 kernel upgrade on the work development branch. This work is expected to be completed by the end of Q2 2026.
  • Fixed PostgreSQL upstream CVEs and security-related patches, including libpq SSL/GSS negotiation error handling.
• Core project improvements:
  • Merged the diskquota extension into the main repository.
  • Added the gp_stats_collector extension, improving observability and laying groundwork for dashboard related work.
  • Added AQUMV exact-match support for multi-table JOIN queries.
  • Started discussion on an Iceberg subsystem for datalake_fdw.
• CI, testing, and platform support:
  • Added Ubuntu 24.04 Docker image and test matrix support.
  • Continued Rocky Linux 10 support work, with initial build docker images. The related test matrix support is to be added.
  • Started discussion on rethinking the testing strategy for better test efficiency and coverage.
  • Upgraded Go to 1.24 in Cloudberry development images and related ecosystem repositories. Newer Go version support is in the discussion.
  • Started adding Behave tests for gpMgmt tools (WIP).
• Ecosystem and sub-repositories:
  • cloudberry-pxf:
    • Started a roadmap discussion for the PXF project.
    • Released 2.1.0, the first Apache release for PXF.
    • Renamed Java packages from org.greenplum to org.apache.cloudberry and updated ASF compliance files.
    • Added Java 17 support and Rocky Linux 9 CI support.
    • Removed gradle-wrapper.jar from source release artifacts and download/verify it during build for ASF license compliance.
    • Added or improved JDBC tests using Testcontainers for ClickHouse, MS SQL, and Oracle.
    • Upgraded dependencies including Apache AVRO, Apache ORC, Parquet, and HBase related components, and discussed Java 21 support.
  • cloudberry-backup:
    • Released 2.1.0, the first Apache release for gpbackup.
    • Completed migration work from legacy Concourse CI to GitHub Actions.
    • Added ASF release script support and Cloudberry 2.1 compatibility.
    • Started a roadmap discussion for the backup project.
    • Merged gpbackman-related backup management work. The Prometheus exporter for the gpbackup history database work is in the review process.
  • cloudberry-go-libs:
    • Tracked a security issue and dependency update needed by related backup tooling.
  • Apache MADlib:
    • Continued collaboration with Apache MADlib to add Apache Cloudberry support upstream, with community discussion ongoing. See: https://s.apache.org/bsu9m
• Next release: Started early discussion for Apache Cloudberry 2.2.0, including release manager, release process documentation, packaging automation, and testing strategy.

How would you assess the podling's maturity?​

• Initial setup
• Working towards first release
• Community building
• Nearing graduation
• Other:

Date of last release:​

• April 14, 2026 - Apache Cloudberry (Incubating) 2.1.0

When were the last committers or PPMC members elected?​

• March 24, 2026 - Xiaoyu Liu (@misterRaindrop), Nikolay Antonov (@ostinru), Robert Mu (@robertmu) were announced as new committers.
• April 07, 2026 - Leonid Borchuk was announced as a new PPMC member.

Join Us​

Apache Cloudberry follows the principle of open and transparent governance. You can follow our quarterly incubation reports to stay updated on key community events and project progress.

Get involved with the community:

• GitHub: github.com/apache/cloudberry
• Slack: Join our Slack workspace
• Discord: Join our Discord server
• Mailing Lists: lists.apache.org/list.html?dev@cloudberry.apache.org
• Website: cloudberry.apache.org

The Apache Cloudberry (Incubating) community is pleased to announce the release of Apache Cloudberry (Incubating) version 2.1.0.

Apache Cloudberry (Incubating) is a Massively Parallel Processing (MPP) database for large-scale data analytics, derived from PostgreSQL and the last open-source version of Greenplum Database. It is designed to support both on-premise and cloud deployments, providing a scalable foundation for data warehousing and advanced analytics.

This release covers the main Cloudberry project and the two components cloudberry-backup and cloudberry-pxf.

We'd like to express our gratitude to all contributors to this release, as well as to the mentors and the Apache Incubator community for their invaluable support. This release continues the project's work on database kernel improvements, ecosystem modernization, and ASF-oriented release engineering.

Release highlights​

Apache Cloudberry 2.1.0 includes significant updates across the core database, Apache Cloudberry PXF, and Apache Cloudberry Backup:

• UDP2 interconnect: New UDP2 interconnect protocol implementation for improved distributed execution
• MCP server: Added MCP server support for easier integration with LLM-based tools and workflows
• PAX enhancements: LZ4 compression support for table columns with I/O and memory-management refinements
• Fast ANALYZE for AO tables: Introduced fast ANALYZE for append-optimized tables, addressing a common operational pain point
• ORCA optimizer improvements: CTE pruning, partial aggregate pushdown below joins, and multiple correctness and memory-leak fixes
• Environment naming transition: Completed migration from greenplum_path.sh to cloudberry-env.sh
• PXF ecosystem modernization: Realigned codebase with greenplum/pxf-archive baseline, updated Java packages to org.apache.cloudberry, and refreshed dependencies
• Backup tooling updates: Renamed to cloudberry-backup and merged S3 plugin into main repository

For more highlights, see the Apache Cloudberry 2.1.0 preview blog post and the changelog.

Download​

• Download the release: https://cloudberry.apache.org/releases
• Changelog for 2.1.0: https://cloudberry.apache.org/releases/2.1.0-incubating
• For documentation, visit: https://cloudberry.apache.org/docs/

Contributors team​

We would like to thank all contributors who made this release possible:

cloudberry: @tuhaihe, @gfphoenix78, @hanwei, @yjhjstz, @andr-sokolov, @reshke, @leborchuk, @zhangyue1818, @water32, @NJrslv, @lss602726449, @zhan7236, @oracleloyall, @edespino, @antoniopetrole, @liang8283, @RyanWei, @ginobiliwang, @kongfanshen, @zhangwenchao-123, @avamingli, @weinan003, @王平10304955, @gongxun0928, @yangs16, @Smyatkin-Maxim, @smartyhero, @HuSen8891, @jiaqizho, @付与龙10336563, @fanfuxiaoran, @yihong0618, @roseduan, @Xue-Bin Cong

cloudberry-pxf: @ostinru, @MisterRaindrop, @Mulily0513, @edespino, @tuhaihe, @my-ship-it, @weinan003, @Terry1504, @gfphoenix78

cloudberry-backup: @robertmu, @tuhaihe, @my-ship-it, @jiaqizho

Join us​

We are eager to expand our community and extend an invitation to new contributors. We genuinely welcome the opportunity to collaborate with individuals who share our passion and expertise.

• Website: https://cloudberry.apache.org
• Repository: https://github.com/apache/cloudberry
• Discussions: https://github.com/apache/cloudberry/discussions
• Issue tracker: https://github.com/apache/cloudberry/issues
• Mailing list: https://lists.apache.org/list.html?dev@cloudberry.apache.org
• Slack channel: https://join.slack.com/t/asf-cloudberry/shared_invite/zt-3um34r7hf-Sh~6jG6hVxlQJo1tbhK2sw
• Discord server: https://discord.com/invite/GJrz3Fxf6y

What Happened​

On April 1, 2025, we were surprised to find our workspace suddenly inaccessible. After contacting Slack support, we received confirmation that our workspace was affected by a policy change regarding workspaces created from certain regions. While Slack had sent a notification email to the workspace's primary owner role about this policy change, it was unfortunately overlooked. We immediately reached out to explore options for data recovery and export, but have not received any updates on this matter.

Our Response​

In response to this situation, we have established new communication channels to ensure our community remains connected:

• New Slack Workspace: We have created a new Slack workspace at asf-cloudberry.slack.com. You can join using the following invitation link: https://join.slack.com/t/asf-cloudberry/shared_invite/zt-3um34r7hf-Sh~6jG6hVxlQJo1tbhK2sw.

• Discord Server: We have also set up a Discord server to provide an additional platform for real-time communication. Discord is widely accessible globally and popular among developer communities. You can join our Discord server here: https://discord.gg/GJrz3Fxf6y.

Historical Channel Archives​

For those who wish to access historical conversations from the previous workspace, we have made archives available through Linen:

• Previous workspace (apache-cloudberry.slack.com): https://www.linen.dev/s/cloudberrydb/
• Current workspace (asf-cloudberry.slack.com): https://www.linen.dev/s/apache-cloudberry

Important Reminder: Mailing Lists Remain Our Primary Channel​

This incident reinforces the importance of relying on mailing lists as our primary and official communication channel. As an Apache project, we follow the Apache Way, which emphasizes:

• Mailing lists are the official record: All important discussions and decisions must take place on our mailing lists
• Transparency and inclusivity: Mailing lists ensure that all community members can participate, regardless of their access to third-party services
• Permanence: Mailing list archives are maintained by the Apache Software Foundation and will remain accessible

Real-time platforms like Slack and Discord serve as supplementary channels for informal discussions and day-to-day collaboration. However, any significant outcomes or decisions from these platforms should always be summarized and documented on the mailing lists.

Our Mailing Lists​

• Dev mailing list: For development discussions, proposals, and technical decisions

  • Subscribe: dev-subscribe@cloudberry.apache.org
  • Archives: https://lists.apache.org/list.html?dev@cloudberry.apache.org

• Commits mailing list: For tracking code commits and pull requests

  • Subscribe: commits-subscribe@cloudberry.apache.org
  • Archives: https://lists.apache.org/list.html?commits@cloudberry.apache.org

For complete information about our mailing lists, please visit our Mailing Lists page.

Looking Forward​

While we have established new real-time communication channels, we acknowledge that as third-party services, they may face similar challenges in the future. This is why we remain committed to the Apache Way and continue to prioritize mailing lists as our primary communication channel.

We appreciate your understanding and continued participation in our community. If you have any questions or concerns, please feel free to reach out through any of our communication channels or contact us at dev@cloudberry.apache.org.

Thank you for being part of the Apache Cloudberry community.

For the discussion thread on this topic, please see the dev mailing list: https://lists.apache.org/thread/fzdo4y45tv1ro4mlwf0w26ozwbnyj2bf

The PPMC has voted to invite Leonid Borchuk to join as a PPMC member, and we are delighted that he has accepted this important role.

Since joining the Apache Cloudberry project, Leonid has been an active and highly engaged contributor. Beyond his solid technical contributions—including helping cherry-pick changes from Greenplum, reviewing PRs, contributing to CI/CD improvements, and more—he has demonstrated a strong commitment to the community and overall project development.

Community Leadership and Contributions​

What sets Leonid apart is his proactive involvement in community-facing responsibilities that go beyond code contributions:

• Incubation Reporting: Proactively volunteered to draft the Cloudberry Incubation Report, helping the project maintain transparency and accountability with the Apache Incubator
• Community Meetings: Initiated and co-organized the Cloudberry bi-weekly community meetings, and has served as a rotating host, fostering regular communication and collaboration
• Code Review and Mentorship: Actively reviews pull requests and engages with contributors through constructive feedback, helping maintain code quality and supporting other contributors
• Strategic Planning: Contributed ideas and discussions around the project roadmap and future directions, demonstrating a vision for the project's growth

These contributions demonstrate not only technical capability, but also a clear sense of responsibility and growing involvement in the project's community and governance.

What is a PPMC Member?​

The Podling Project Management Committee (PPMC) is responsible for the oversight and direction of a podling project during its incubation at the Apache Software Foundation. PPMC members help build a healthy and sustainable community, guide the project's development under ASF principles, and work toward graduation as a top-level Apache project.

Looking Forward​

Leonid's consistent contributions, active participation, and demonstrated leadership make him a strong addition to the PPMC. We are confident that he will continue to help us build a thriving, inclusive, and sustainable community for Apache Cloudberry.

Please join us in congratulating Leonid Borchuk on this well-deserved recognition and welcoming him to the PPMC!

You can explore his contributions here:

• Pull Requests
• Mailing List Discussions

You can find the official announcement on the Apache Cloudberry Dev mailing list.

Welcome to the PPMC, Leonid!

Over the past few months, Nikolay has been an active and impactful contributor, particularly in the cloudberry-pxf project. He has worked on aligning the Cloudberry PXF codebase with the archived Greenplum PXF project, contributed multiple code improvements and optimizations, and actively participated in community discussions with constructive ideas and feedback.

What makes Nikolay's contributions especially noteworthy is his work beyond Cloudberry itself. He has contributed to the broader Cloudberry/PostgreSQL ecosystem by adding Cloudberry support to WAL-G and continuing to improve Cloudberry integration in WAL-G. This work significantly enhances backup and recovery capabilities for Cloudberry users.

Becoming a committer is a recognition of both Nikolay's past contributions and the community's trust in his continued involvement. Committers play an important role in shaping the project by reviewing and merging contributions, participating in design decisions, and helping guide the project's growth.

Please join us in congratulating Nikolay on this well-deserved recognition. We look forward to his continued contributions and collaboration within the Apache Cloudberry community!

You can find the official announcement on the Apache Cloudberry mailing list.

Xiaoyu has been involved in the Cloudberry project from an early stage, reporting issues and contributing improvements to the main repository. In recent months, he has taken a leadership role in aligning the cloudberry-pxf codebase with the archived Greenplum PXF project, added CI/CD workflows for PXF, and actively fixed issues to improve the project.

What makes Xiaoyu's contributions particularly valuable is his collaborative approach. He has worked closely with other developers in the community, participating in discussions and sharing technical ideas and suggestions that benefit the entire project.

Becoming a committer is a recognition of both Xiaoyu's past contributions and the community's trust in his continued involvement. Committers play an important role in shaping the project by reviewing and merging contributions, participating in design decisions, and helping guide the project's growth.

Please join us in congratulating Xiaoyu on this well-deserved recognition. We look forward to his continued contributions and collaboration within the Apache Cloudberry community!

You can find the official announcement on the Apache Cloudberry mailing list.

Apache Cloudberry (Incubating) 2.1.0 continues the project's work on database kernel improvements, ecosystem modernization, and ASF-oriented release engineering. Based on the current release candidate, this upcoming version includes changes across the core database, Apache Cloudberry PXF, and Apache Cloudberry Backup.

This article is a preview rather than an official release announcement, so it focuses on a few representative updates instead of a complete changelog.

Apache Cloudberry Core (apache/cloudberry)​

The core repository in 2.1.0 includes both user-facing improvements and behind-the-scenes engineering work.

It is important to emphasize that 2.1.0 is built on a stable branch. The community is currently undertaking an effort to upgrade the underlying PostgreSQL core from version 14.4 to the latest 14.22. This approach ensures that the database receives essential kernel updates while maintaining strict stability. While 2.1.0 does not include the entirety of this massive upgrade, these updates will be included in the future 2.x release.

Selected kernel and performance updates​

• UDP2 interconnect: 2.1.0 adds a new UDP2 interconnect protocol implementation, continuing work on the interconnect layer and distributed execution path.
• MCP server: The release candidate adds an MCP server for Cloudberry, making it easier to connect Cloudberry with LLM-based tools and workflows.
• PAX improvements: PAX gains support for LZ4 compression for table columns, along with I/O and memory-management refinements.
• Fast ANALYZE for AO tables: The release candidate introduces fast ANALYZE for append-optimized tables, targeting a common operational pain point for large analytical workloads.
• ORCA optimizer work: 2.1.0 includes a series of ORCA improvements such as CTE pruning, partial aggregate pushdown below joins, and multiple correctness and memory-leak fixes.
• Runtime filter pushdown: Runtime filter pushdown work continues in this cycle, including support for pushing filters down to Table AM and related fixes.

Developer workflow and release engineering​

• Release-branch testing and ASF checks: The project added binary swap testing for the REL_2_STABLE branch and expanded Apache RAT-related audit checks in CI.
• Sandbox and developer tooling: The earlier apache/cloudberry-devops-release and apache/cloudberry-bootcamp work has been integrated into the main repository under devops, and the sandbox gains a local mode for development and validation.
• Environment naming transition: In 2.1, the environment setup has completed the move from greenplum_path.sh to cloudberry-env.sh. For background on that transition, see Goodbye greenplum_path.sh, Hello cloudberry-env.sh: A Phased Transition Plan.
• Compression and dependency cleanup: The codebase also removes support for QuickLZ, encouraging the use of maintained alternatives such as zlib or zstd.

There are many other fixes in areas such as hot standby, disaster recovery, utilities, CI portability, and testing. For a broader view of the current release candidate, see the core branch comparison.

Apache Cloudberry PXF (apache/cloudberry-pxf)​

Apache Cloudberry PXF is one of the clearest examples of ecosystem modernization in the 2.1.0 cycle.

First, the codebase has been realigned with the archived greenplum/pxf-archive baseline, and Cloudberry-specific support has been ported onto that newer base for Cloudberry 2.1. The repository also received ASF-oriented cleanup, including updates to DISCLAIMER, LICENSE, and NOTICE, as well as removal of the pre-bundled gradle-wrapper.jar.

Second, the project has refreshed its engineering workflow. Legacy Concourse CI files have been removed in favor of a refactored and robust CI layout oriented around current Cloudberry development and testing.

Third, 2.1.0 continues the broader Cloudberry branding refresh. Java package declarations move from org.greenplum to org.apache.cloudberry, and the Go library dependency references are updated to apache/cloudberry-go-libs.

PXF also includes a substantial set of dependency and connector updates in this cycle:

• Apache Avro is updated to newer Java 8 compatible versions, with added zstd support in tests and related automation.
• Apache ORC is updated to 1.7.11.
• Parquet is updated first to 1.12.3 and later to 1.15.2, adding support such as ZSTD, LZ4_RAW, and vectored I/O improvements.
• HBase client support is updated from 1.3.2 to 2.3.7.

For more detail, see the PXF branch comparison.

Apache Cloudberry Backup (apache/cloudberry-backup)​

Apache Cloudberry Backup also sees significant repository and tooling work in the current 2.1.0 release candidate.

The project has been renamed from cloudberry-gpbackup to cloudberry-backup, and its codebase has been aligned with the archived greenplum/gpbackup-archive baseline while preserving Cloudberry-specific adaptation for Cloudberry 2.1. Dependency references have also been updated from gp-common-go-libs to apache/cloudberry-go-libs.

One practical usability change is that the archived gpbackup-s3-plugin has been merged into the main repository under plugins/s3plugin, reducing the amount of separate setup required for S3-based backup workflows.

The repository also continues the project's ASF-oriented cleanup work, including updates to LICENSE and NOTICE files, removal of legacy files, and modernization of CI coverage for unit, integration, end-to-end, S3 plugin, and scale-related testing.

For the complete set of changes in the current release candidate, see the Backup branch comparison.

Looking Ahead​

Even before the 2.1.0 vote is complete, the current release candidate already shows the project's direction clearly: continued kernel work in the main database, modernization of surrounding components, and a cleaner foundation for future community development.

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

We welcome everyone to continue following and participating in the Apache Cloudberry community to witness the 2.1.0 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

Cloudberry​

Cloudberry is an advanced and mature open-source Massively Parallel Processing (MPP) database, derived from the open-source version of Pivotal Greenplum Database®️ but built on a more modern PostgreSQL 14 kernel, whereas Greenplum is based on PostgreSQL 12. This upgrade brings enhanced enterprise capabilities, making Cloudberry well-suited for data warehousing, large-scale analytics, and AI/ML workloads.

Cloudberry has been incubating since 2024-10-11.

Three most important unfinished issues to address before graduating:​

1. Grow the contributor and community to ensure long-term sustainability.
2. Publish a few more Apache releases following the ASF release processes.

How has the community developed since the last report?​

• Mailing list Activity: 82 new emails on the Dev mailing list since the last report, covering community, repositories, support, technical and Apache-related discussions.
• GitHub Discussions: 5 new threads since last report.
• New Committers: Rose Duan
• Events:
  • Hosted a Cloudberry community booth at the China Open Source Conference (COSCon'25) in Beijing, Dec 6-7, 2025.
  • Apache Cloudberry Meetup organized in Shenzhen: 30+ attendees on Jan 17th
  • Commenced Apache Cloudberry Bi-weekly Community Meeting, see details in https://s.apache.org/9ba4d, the 1st meeting was held on January 30, 2026.

How has the project developed since the last report?​

• PostgreSQL Kernel upgrade (14~>16): at this stage, most of the core enablement work for PostgreSQL 16.9 has been completed. Currently addressing Cloudberry schedule tests.
• Performance
  • PAX: optimize I/O read for multiple discrete columns in a group.
  • Add UDP2 interconnect protocol implementation (right now it is still an experimental feature).
  • Set join_collapse_limit default value to 13.
• Stability
  1. Merged 72 commits, most of them - bugfixes and CI/CD improvements.
  2. Fixed 2 CVEs.
• CI and infra
  1. Added support for Ubuntu 22.04 and Debian packages to CI/CD
  2. Added support for Rocky Linux 8 to CI/CD
• Sub-repositories:
  • cloudberry-pxf. After several months of continuous effort, the PXF project has recently achieved a significant milestone:
    1. The code baseline has been 100% aligned with the archived pxf-archive from GP.
    2. Completed full adaptation for Cloudberry.
    3. Added a relatively comprehensive and extensive CI pipeline for PXF.
    4. Source code cleanup is in progress.
  • Squash cloudberry-bootcamp repository into devops/sandbox in the main repository catalog.
• New release: 2.1
  • Merged 244 commits from main to REL_2_STABLE. We're ready to release Cloudberry 2.1.
• Ecosystem
  1. Madlib integration is still in progress. We successfully performed manual tests, but PR is still open to fix the CI bugs: https://github.com/apache/madlib/pull/627.
  2. Working on diskquota extension integration: still in progress - see https://github.com/apache/cloudberry/pull/1490.
• 2025 Summary and 2026 Overview discussions
  1. Published Cloudberry 2025 Review blog post: https://s.apache.org/fxde8
  2. Initiated Cloudberry 2026 roadmap discussions: https://s.apache.org/jxd7s

How would you assess the podling's maturity?​

• Initial setup
• Working towards first release
• Community building
• Nearing graduation
• Other:

Date of last release:​

• August 25, 2025

When were the last committers or PPMC members elected?​

• Rose Duan, 19th Jan 2026

Join Us​

Apache Cloudberry follows the principle of open and transparent governance. You can follow our quarterly incubation reports to stay updated on key community events and project progress.

Get involved with the community:

• GitHub: github.com/apache/cloudberry
• Slack: Join our Slack workspace
• Mailing Lists: lists.apache.org/list.html?dev@cloudberry.apache.org
• Website: cloudberry.apache.org

Rose has been an active and valuable contributor to Apache Cloudberry over the past few years. Through his consistent involvement, he has demonstrated strong technical expertise, a collaborative mindset, and a clear commitment to the long-term success of the project.

Rose has contributed to a wide range of Cloudberry features and improvements, including enhancements to warehouse grammar and numerous bug fixes. Beyond the core repository, Rose has also played an important role in maintaining and improving ecosystem projects closely related to Cloudberry, such as kafka_fdw for Cloudberry and pgvector for Cloudberry.

In addition to his technical work, Rose has been actively involved in community outreach and advocacy. Notably, he shared Cloudberry-related content at Community Over Code Asia 2025 (formerly ApacheCon), helping raise awareness of the project and connect Cloudberry with a wider audience.

His willingness to engage in discussions, respond to feedback, and support other contributors reflects the collaborative spirit we value deeply in the Apache community.

After discussion on the private mailing list and a successful vote by the PPMC, Rose was formally invited to become a Cloudberry Committer. This recognition reflects not only his past contributions, but also the community’s trust in his continued involvement and leadership going forward.

Please join us in welcoming Rose Duan as the newest Apache Cloudberry Committer! We look forward to his ongoing contributions and to working together to grow the Cloudberry project and community.

Welcome aboard, Rose!

You can find the official announcement on the Apache mailing list.

This post looks back at the key milestones of Apache Cloudberry in 2025 and reflects on how the project has evolved since entering the Apache Incubator.

From Proposal to Incubation​

Apache Cloudberry officially entered the Apache Incubator in October 2024. Following the successful vote, the project completed the migration of its source code repositories and infrastructure to the Apache Software Foundation in November 2024.

This initial phase focused on establishing a solid foundation under the ASF:

• Migrating repositories and CI infrastructure
• Completing project renaming and branding updates
• Aligning documentation and workflows with ASF policies
• Starting systematic code cleanup and compliance work
• Gathering broad community feedback and discussing the roadmap in depth

These efforts set the stage for sustained development and growth throughout 2025.

First Apache Release: Cloudberry 2.0.0 (Incubating)​

One of the most important milestones in 2025 was the release of Apache Cloudberry (Incubating) 2.0.0 in August 2025 — the project's first official Apache release.

Leading up to this release, the community completed several major efforts:

• Cherry-picking and consolidating large portions of the archived Greenplum open-source codebase
• Completing ASF license and compliance requirements (LICENSE, NOTICE, RAT audits)
• Cleaning up legacy files and restructuring the repository
• Building a new, Apache-compliant CI/CD workflow
• Running multiple release candidate (RC) rounds and addressing review feedback from the Incubator community

A major highlight of this release cycle was the introduction of the PAX (Partition Attributes Across) storage engine, a row-column hybrid format that optimizes performance for mixed workloads—a significant innovation contributed back to the core.

The 2.0.0 release marked a major transition from incubation setup to regular, disciplined Apache-style releases.

For more details, see the release announcement and What's New in Apache Cloudberry 2.0.0.

Community Growth​

Community is the heartbeat of any Apache project, and 2025 saw the Cloudberry family grow stronger and more diverse.

During the year, the Cloudberry Podling Project Management Committee (PPMC) was proud to welcome four new committers:

• Xiong Tong
• Wenchao Zhang
• Xun Gong
• Leonid Borchuk

Mailing list discussions, GitHub issues, and design conversations continued to be kept active, covering topics such as kernel upgrades, performance improvements, extensions, and ecosystem integration.

The project has also emphasized transparent governance, clear contribution guidelines, and mentoring new contributors as part of its incubation journey.

Community Activities and Events​

Apache Cloudberry community members were active across a wide range of global and regional events in 2025, including:

• PostgresConf Global 2025
• Flink Forward Asia 2025
• Community Over Code Asia 2025
• PostgreSQL and data community meetups in Asia, Europe, and North America

In addition to conference talks, the community members organized and participated in Apache Cloudberry Meetups in multiple cities, providing opportunities to share technical updates, discuss roadmap items, and introduce the Apache way to new users and contributors.

These activities helped increase the project's visibility and attracted new participants from the broader PostgreSQL and data ecosystem.

PostgreSQL Kernel Upgrade: PG 14 → PG 16​

In 2025, the project also initiated the PostgreSQL kernel upgrade, moving Apache Cloudberry from PostgreSQL 14.x toward PostgreSQL 16.x.

This effort represents a significant technical investment and includes:

• Merging the new PostgreSQL kernel into Cloudberry work branches
• Resolving conflicts introduced by Cloudberry's MPP architecture
• Incrementally restoring build, initialization, and test coverage
• Validating parallel execution and regression test suites

The kernel upgrade work remains ongoing, but important milestones — including successful cluster startup — were achieved during the year. This challenging effort is expected to be completed in early 2026.

If you're interested in contributing to this effort, check out the kernel upgrade discussion on the mailing list and the GitHub Project board.

Ecosystem and Integrations​

The Apache Cloudberry ecosystem expanded significantly in 2025, driven by a "Upstream first" philosophy.

Native Integrations​

Several widely used tools now provide native support for Cloudberry, including:

• DBeaver: For seamless database management and SQL development.
• WAL-G: Enabling robust disaster recovery and continuous archiving.
• Apache SeaTunnel: Facilitating high-performance data integration.
• Flink JDBC Connector (v3.3.0+): Supporting real-time streaming data ingestion.

Community-Maintained Extensions​

The community also maintains a set of adapted extensions for Cloudberry:

• PostGIS for Cloudberry: Upgraded to PostGIS 3.3.2, unlocking advanced geospatial analytics capabilities.
• kafka_fdw for Cloudberry: Enabling efficient real-time data processing from Kafka.
• cbcopy: Supporting migration from Greenplum to Cloudberry.
• ParadeDB for Cloudberry: Enabling search and analytics workloads on Cloudberry.
• pgvector for Cloudberry: Upgraded to version 0.8, enabling similarity search capabilities.

Cross-Project Collaboration​

Work is actively underway to add native Apache Cloudberry support to Apache MADlib, strengthening Cloudberry's position in advanced analytics and machine learning workloads. This collaboration highlights the strength of the Apache ecosystem, where projects work together to deliver integrated solutions.

Repository Consolidation​

To streamline the developer experience, the community consolidated several sub-repositories: cloudberry-gpbackup was renamed to cloudberry-backup, and the S3 plugin was merged into the main backup repository.

Furthermore, apache/cloudberry-devops-release and apache/cloudberry-bootcamp were integrated into the main Cloudberry repository under the devops/ and devops/sandbox directories, respectively.

Looking Ahead​

As 2025 comes to a close, Apache Cloudberry stands on a much stronger foundation than it did a year ago:

• A completed first Apache release
• A growing and more diverse contributor community
• Active participation in global open-source events
• Ongoing kernel modernization
• An expanding ecosystem of integrations

The project will continue focusing on community growth, regular Apache releases, technical stability, and progress toward graduation from the Apache Incubator.

We would like to thank all contributors, mentors, and community members who helped make 2025 a productive and rewarding year for Apache Cloudberry.

Stay Connected​

Apache Cloudberry follows the principle of open and transparent governance. You can follow our quarterly incubation reports to stay updated on key community events and project progress.

Get involved with the community:

• GitHub: github.com/apache/cloudberry
• Slack: Join our Slack workspace
• Mailing Lists: lists.apache.org/list.html?dev@cloudberry.apache.org
• Website: cloudberry.apache.org

Cloudberry​

Cloudberry is an advanced and mature open-source Massively Parallel Processing (MPP) database, derived from the open-source version of Pivotal Greenplum Database®️ but built on a more modern PostgreSQL 14 kernel, whereas Greenplum is based on PostgreSQL 12. This upgrade brings enhanced enterprise capabilities, making Cloudberry well-suited for data warehousing, large-scale analytics, and AI/ML workloads.

Cloudberry has been incubating since 2024-10-11.

Three most important unfinished issues to address before graduating:​

1. Grow the contributor and community to ensure long-term sustainability.
2. Publish a few more Apache releases following the ASF release processes.

How has the community developed since the last report?​

• Mailing list Activity: 210 new emails on the Dev mailing list since the last report, covering community, technical and Apache-related discussions.
• GitHub Discussions: 14 new threads since last report.
• New Committers:
  • Sep 25, 2025 - Leonid Borchuk (leborchuk)
• Events:
  • Apache Cloudberry Meetup organized in Beijing on August 16, 2025, which attracted 30~ attendees. This meetup focused on the new features and highlights of the 2.0.0 release and the Apache way. See recap post at https://s.apache.org/xqyi9.
  • Our committer Leonid Borchuk presented a talk, "PAX — column store for Apache Cloudberry/Postgres 14", at PGConf.SPb 2025 in Saint Petersburg on September 29, 2025. See details at https://pgconf.ru/talk/2484244.

How has the project developed since the last report?​

• Publish the first Apache Cloudberry (Incubating) 2.0.0 release on August 25, 2025. See announcement email at https://s.apache.org/fn13p
• PostgreSQL Kernel upgrade (14~>16): the work has been in progress. See the monthly report at https://s.apache.org/kp3lj
• 113 new commits to the main branch since the last report, focusing on CI, performance, bug fixes, and new features.
• Sub-repositories:
  • Archived the apache/cloudberry-devops-release repo and migrated these files to the devops/ directory of the main repository.
  • For the backup tool repos:
    • The cloudberry-gpbackup repo has been renamed to cloudberry-backup, reflecting the project’s independence and alignment with the Cloudberry brand. Its codebase has synced with the GP’s archived version.
    • The core files from apache/cloudberry-gpbackup-s3-plugin have been merged into the cloudberry-backup repos under plugins/s3plugin. Now, the S3 plugin will be installed together with Cloudberry Backup.
  • The cloudberry-pxf is still in progress on the archived Greenplum commits sync to Cloudberry.
  • cloudberry-go-libs has synced with the GP’s archived version.
• Started the discussion on the next release 2.1.0. See discussion at https://s.apache.org/rx7s8
• It has been one year since Cloudberry joined the ASF Incubator. Here is our Roadmap recap at https://s.apache.org/5fxn9
• Ecosystem:
  • The community developers are working with the Apache MADlib team to add support for Cloudberry to MADlib upstream. See the PR at https://github.com/apache/madlib/pull/627
  • The PostGIS for Cloudberry has been upgraded from 2.5 to 3.3.2

How would you assess the podling's maturity?​

• Initial setup
• Working towards first release
• Community building
• Nearing graduation
• Other:

Date of last release:​

• August 25, 2025

When were the last committers or PPMC members elected?​

• Sep 25, 2025 - Leonid Borchuk (leborchuk)

The Podling Project Management Committee (PPMC) for Apache Cloudberry has invited Leonid Borchuk to become a committer, and we are pleased to announce that he has accepted.

Leonid has been an active contributor to the Apache Cloudberry project since October 2024. He has demonstrated consistent engagement and technical expertise across multiple areas, including cherry-picking commits from Greenplum to Cloudberry, reviewing PRs and providing valuable feedback to other contributors, participating in roadmap discussions, and contributing innovative ideas. He has been instrumental in renaming greenplum_path.sh to cloudberry-env.sh.

He also helped migrate the cloudberry-devops-release repo to the main repo and implemented Ubuntu support workflow for CI/CD. These efforts have brought significant improvements to our infrastructure.

Becoming a committer recognizes both Leonid's past contributions and the community's trust in his continued involvement. Committers play a vital role in shaping the project by reviewing and merging contributions, participating in design decisions, and helping guide the project's growth.

Please join us in congratulating Leonid Borchuk on this well-deserved recognition. We look forward to his continued contributions and collaboration within the Apache Cloudberry community!

You can find the official announcement on the Apache mailing list.

Xun has been a valuable contributor to the project, focusing on performance improvements and system reliability. His work spans query optimization, bug resolution, and enhancing core database features. Check out his contributions: Xun's GitHub Activity.

What sets Xun apart is his thoughtful approach to code reviews and his ability to identify optimization opportunities that benefit the entire community. He brings both technical depth and a collaborative spirit that strengthens our development process.

Becoming a committer is a recognition of both Xun's past contributions and the community's trust in his continued involvement. Committers play an important role in shaping the project by reviewing and merging contributions, participating in design decisions, and helping guide the project's growth.

Please join us in congratulating Xun on this well-deserved recognition. We look forward to his continued contributions and collaboration within the Apache Cloudberry community!

You can find the official announcement on the Apache mailing list.

In this article, we highlight only the most important features, enhancements, and fixes to help you quickly understand the key improvements in this release. For more details, please refer to the Apache Cloudberry 2.0.0 Changelog.

New features​

Query processing and optimization​

Index and scan​

Enhanced index-only scan capabilities​

• Supports index-only scans on a broader range of index types when using the GPORCA optimizer, including those with covering indexes using INCLUDE columns. This helps improve query performance.

• Supports dynamic index-only scan when using the GPORCA optimizer to accelerate queries on partitioned tables. This feature combines partition pruning with index-only access to avoid heap lookups, significantly reducing I/O and improving performance. It is ideal for wide tables with narrow covering indexes and can be enabled using SET optimizer_enable_dynamicindexonlyscan = on.

• Supports index-only scans when using the GPORCA optimizer on append-only (AO) tables and PAX tables, enabling faster query execution by avoiding block access when possible. This improves performance in scenarios where traditional index scans on AO and PAX tables were previously inefficient.

Improved index scan performance and flexibility​

• Supports backward index scans when using the GPORCA optimizer for queries with ORDER BY ... DESC, eliminating the need for explicit sorting when a B-tree index exists in the opposite order. This optimization reduces resource usage and improves performance, especially for top-N and pagination queries.

• The GPORCA optimizer supports triggering Bitmap Index Scans using array comparison predicates like col IN (...) or col = ANY(array), including for hash indexes. This improves query performance on large datasets by enabling more efficient multi-value matching. The optimizer automatically chooses the bitmap scan path based on cost estimation.

• The GPORCA optimizer now considers the width of INCLUDE columns when costing index-only scans, favoring narrower indexes that return fewer unused columns. This improves plan selection for queries where multiple covering indexes are available. The cost model also more accurately estimates I/O by refining how relallvisible is used in index-only scan costing.

BRIN index enhancements​

• Redesigns BRIN index internals for AO/CO tables to replace the UPPER page structure with a more efficient chaining model. This significantly reduces disk space usage for empty indexes and improves performance by avoiding unnecessary page access. The new design better handles the unique layout of AO/CO tables while maintaining correctness and compatibility.

• BRIN indexes on AO/CO tables now support summarizing specific logical heap block ranges using brin_summarize_range(), enabling more precise control during index maintenance and testing. This enhancement also adds improved coverage for scenarios involving aborted rows, increasing robustness and correctness in edge cases.

• Supports generating IndexScan plans when using the GPORCA optimizer with ScalarArrayOp qualifiers (for example, col = ANY(array)) for B-tree indexes. This enhancement aligns ORCA with the planner's behavior and allows more efficient execution of array comparison queries, as long as the predicate column is the first key in a multicolumn index.

View and materialized view​

• Improves performance of REFRESH MATERIALIZED VIEW WITH NO DATA by avoiding full query execution. The command now behaves like a TRUNCATE,significantly reducing execution time while preserving proper dispatch to segments.

Join​

• Supports left join pruning when using the GPORCA optimizer, allowing unnecessary left joins to be eliminated during query optimization. This applies when the query only uses columns from the outer table and the join condition fully covers the inner table's unique or primary keys. This can lead to more efficient query plans.

• Supports FULL JOIN using the Hash Full Join strategy when using the GPORCA optimizer. This approach avoids sorting join keys and reduces data redistribution, making it suitable for large datasets or joins on non-aligned distribution keys. All FULL JOIN queries now use Hash Full Join.

• The GPORCA optimizer now avoids unnecessary data redistribution for multi-way self joins using left or right outer joins when the join keys are symmetric. This optimization improves performance by recognizing that such joins preserve data colocation, eliminating redundant motion operations.

• The GPORCA optimizer no longer penalizes broadcast plans for NOT IN queries (Left Anti Semi Join), regardless of the optimizer_penalize_broadcast_threshold setting. This change improves performance and avoids potential OOM issues by enabling parallel execution instead of concentrating large tables on the coordinator node.

Function & aggregate​

• Supports intermediate aggregates when using the GPORCA optimizer, enabling more efficient execution of queries that include both DISTINCT aggregates and regular aggregates. This ensures correct handling of aggregation stages using AGGSPLIT. In addition, ORCA introduces an optimization for MIN() and MAX() functions by using index scans with a limit, instead of full table scans with regular aggregation. This optimization also supports IS NULL and IS NOT NULL conditions on indexed columns, significantly improving performance for applicable queries.

• Enables more HashAggregate plan alternatives for queries that include DISTINCT aggregates when using the GPORCA optimizer. By generating a two-stage aggregation plan that avoids placing DISTINCT functions in hash-based nodes, ORCA ensures compatibility with the executor and expands the range of supported query plans. This improvement enhances optimization choices for group-by queries.

• Supports queries using GROUP BY CUBE, enabling multi-dimensional grouping sets in query plans. This expands analytic query capabilities. Note that optimization time for CUBE queries may be high due to the large number of generated plan alternatives.

Preprocessing​

• Inlines Common Table Expressions (CTEs) that contain outer references, allowing such queries to be planned and explained successfully. Previously, these queries would fall back to the legacy planner due to limitations in handling shared scans with outer references. This change improves compatibility and enables ORCA to optimize a broader range of CTE-based queries.

• No longer rewrites IN queries to EXISTS when the inner subquery contains a set-returning function. This prevents invalid query transformations that could previously result in execution errors. The change ensures correct handling of queries like a IN (SELECTgenerate_series(1, a)).

Optimization and performance enhancements​

Dynamic Table​

• Introduces dynamic tables, a new feature that enables automatic, scheduled refresh of query results. Dynamic tables are similar to materialized views but are designed for scenarios requiring up-to-date data, such as real-time analytics, lakehouse architectures, and automated ETL pipelines.
• Supports creating dynamic tables from base tables, external tables, or materialized views. Users can define refresh intervals using standard cron expressions, ensuring data is kept current without manual intervention.
• For more details, see the official documentation.

Plan hint​

• Supports plan hints for scan types and join row estimates when using the GPORCA optimizer, enabling users to guide query planning using pg_hint_plan-style comments. Supports scan hints include SeqScan, IndexScan, BitmapScan, and their negations, while row hints allow users to specify expected join cardinalities.

• The plan hint field is now required in the ORCA optimizer configuration. This change simplifies internal parsing logic and ensures consistent handling of optimizer configuration files.

• Supports join order hints for left and right outer joins when using the GPORCA optimizer, extending the existing hint framework beyond inner joins. This enhancement allows users to guide the optimizer's join order more precisely in complex queries involving outer joins, improving plan control and potentially execution performance.

Enhancements to ORCA​

• Supports table aliases in query plans when using the GPORCA optimizer, making EXPLAIN outputs more descriptive and aligned with user-defined query syntax. In addition, ORCA adds support for query parameters, including those used in functions and prepared statements, enabling better compatibility with parameterized workloads and dynamic SQL execution.

• When using the GPORCA optimizer, supports generating plans for queries on tables with row-level security (RLS) enabled. Security policies are enforced during plan generation, ensuring only permitted rows are visible to each user. ORCA still falls back to the planner for RLS queries with sublinks, foreign tables, or for INSERT and UPDATE statements.

• The GPORCA optimizer now gracefully falls back to the Postgres planner when a function in the FROM clause uses WITH ORDINALITY, which is not currently supported. The fallback includes a clear error message indicating the unsupported feature.

• When using the GPORCA optimizer, supports pushing down filters with BETWEEN predicates when combined with constant filters, enabling more effective predicate propagation. This enhancement can reduce the number of rows processed during joins, improving query performance in applicable cases.

• When using the GPORCA optimizer, supports hashed subplans when the subquery expression is hashable and contains no outer references. This enhancement can significantly improve query performance by reducing execution time in applicable cases.

• ORCA now supports executing foreign tables with mpp_execute='ANY' on either the coordinator or segments, depending on cost. This allows more flexible and efficient execution plans for foreign data sources. A new "Universal" distribution type is introduced to support this behavior, similar to how generate_series() is handled.

• ORCA now supports direct dispatch for randomly distributed tables when the query includes a filter on gp_segment_id. This enhancement improves query performance by routing execution directly to the relevant segment, reducing unnecessary data processing across the cluster.

• ORCA now supports generating plans with the ProjectSet node, enabling correct execution of queries that include set-returning functions (SRFs) in the target list. This enhancement prevents fallback to the legacy planner and ensures compatibility with PostgreSQL 11+ behavior.

• ORCA now supports the FIELDSELECT node, which allows it to optimize a broader range of queries involving composite data types. Previously, such queries would fall back to the legacy planner. This enhancement improves compatibility and reduces unnecessary planner fallbacks.

• ORCA now derives statistics only for the columns used in UNION ALL queries, instead of all output columns from the input tables. This optimization reduces unnecessary computation and can improve planning performance for large queries.

• Updates naming in logs and EXPLAIN output to refer to the optimizers as "GPORCA" and "Postgres based planner" for improved clarity and consistency.

• Optimizes ORCA's Union All performance by deriving statistics only for columns used in the query output. This reduces unnecessary computation and improves planning efficiency for queries with unused columns.

Transaction management​

Lock management​

• Updates logic to ignore invalidated slots while computing the oldest catalog Xmin, reducing the risk of deadlocks and improving transaction concurrency.

• Performs serializable isolation checks early for AO/CO tables, ensuring stricter consistency guarantees and reducing the likelihood of isolation conflicts.

• Enhances the index creation process to prevent deadlocks by ensuring the coordinator acquires an AccessShareLock on pg_index before dispatching a synchronization query to segments, thus aligning indcheckxmin and avoiding conflicts that GDD cannot resolve.

Transaction performance and reliability​

• Avoids replaying DTX information in checkpoints for newly expanded segments, preventing potential inconsistencies during recovery.

• Adds gp_stat_progress_dtx_recovery for better observability of distributed transaction recovery progress.

• Improves error reporting for DTX protocol command dispatch errors, making it easier to diagnose and resolve issues.

• Allows utility mode on the coordinator to skip upgrading locks for SELECT locking clauses, improving efficiency for maintenance operations.

Storage​

PAX table format​

• Introduces support for the PAX (Partition Attributes Across) storage format, a hybrid approach that combines the advantages of row-based and column-based storage. PAX is designed to deliver high performance for both data writes and analytical queries, making it well-suited for OLAP workloads and large-scale data analysis. For more details, see the official documentation.

AO/CO table enhancements​

• Optimizes CREATE INDEX operations on AO tables with scan progress reporting, enhancing the efficiency of index creation.

• Declares the connected variable as "volatile" to ensure proper handling across PG_TRY and PG_CATCH blocks, mirroring PostgreSQL's best practices for exception-safe variable usage in transaction control.

Partitioning​

• Extends Orca's planning capabilities to include support for foreign partitions, enabling optimized query execution for tables with a mix of foreign and non-foreign partitions. The implementation introduces new logical and physical operators for foreign partitions, supports static and dynamic partition elimination, and integrates with any foreign data wrapper compatible, enhancing performance and flexibility for external data queries.

• Optimizes the analysis of leaf partitions in multi-level partition tables to avoid unnecessary resampling of intermediate partitions.

• Supports dynamic partition elimination (DPE) when using the GPORCA optimizer for plans involving duplicate-sensitive random motions. This allows partition selectors to pass through segment filters, enabling more efficient query plans and reducing the number of scanned partitions.

• Adds Dynamic Partition Elimination for Hash Right Joins, which enhances the efficiency of join operations on partitioned tables.

• Supports boolean static partition pruning in ORCA, enhancing the efficiency of partition pruning during query optimization.

• Enhances ORCA's query planning by incorporating partition key opfamily checks during partition pruning to optimize data distribution and partition scanning, ensuring correct motion triggering and partition scanning by aligning predicate operators with the distribution or partition key's opfamily, addressing issues with missing motion, incorrect direct dispatch, and ineffective partition pruning.

• Caches the last found partition in ExecFindPartition to improve performance for repeated partition lookups.

• Enables ORCA to derive dynamic table scan cardinality from leaf partitions, addressing limitations in handling date and time-related data types by changing their internal representation to doubles.

• Enhances the DPv2 algorithm to include distribution spec information with partition selectors, improving the efficiency of distributed query execution.

• Introduces a new Non-Replicated distribution specification to optimize join operations in database processing. By relaxing the enforcement of singleton distribution for outer tables when the inner table is universally distributed, it aims to reduce unnecessary data gathering and duplicate-sensitive motions, thereby generating more efficient execution plans.

Memory management​

• Implements a custom allocator to enable ORCA to use standard C++ containers, addressing heap allocation management.

• Refactors ORCA's memory pool by making several methods static and adds assertions to ensure pointer safety.

• Optimizes serialization of IMDId objects in ORCA to be lazy, improving performance by deferring serialization until necessary. Improves optimization time when loading objects into the relcache and when involving large and wide partition tables.

• Ensures that strings returned by GetDatabasePath are always freed using pfree, preventing memory leaks.

• Enables MPP (Massively Parallel Processing) support for pg_buffercache and builds it by default, making buffer cache management more scalable and efficient in distributed environments.

• Introduces pg_buffercache_summary() to offer a high-level overview of buffer cache activity.

Metadata and access methods​

• Allows the definition of lock modes for custom reloptions, providing more control over table and index access.

• Supports specification of reloptions when switching storage models, allowing seamless transitions between different storage formats.

• Introduces a new struct member in CreateStmt to indicate the origin of the statement, specifying if it was generated from GP style classic partitioning syntax.

• Adds syscache lookup for pg_attribute_encoding and pg_appendonly, improving performance and efficiency in metadata access.

• Introduces a new catalog entry in pg_aggregate to store replication safety information for aggregates, allowing users to mark specific aggregates as safe for execution on replicated slices via an optional repsafe parameter during the CREATE AGGREGATE command. This helps optimize performance by avoiding unnecessary broadcasts on large replicated datasets.

• Enhances the dispatch of ALTER DATABASE commands by allowing options like ALLOW_CONNECTIONS and IS_TEMPLATE to be dispatched to segments, ensuring catalog changes are reflected everywhere.

Data loading and external tables​

External table enhancements​

• Adds clearer restrictions and warnings when exchanging or attaching external tables. Writable external tables can no longer be used as partitions, and attaching readable external tables without validation now triggers a warning instead of requiring a no-op clause.

• Disables SET DISTRIBUTED REPLICATED for ALTER EXTERNAL TABLE to prevent misuse and ensure consistency.

Foreign data wrapper​

• Improves performance and stability for gpfdist external tables. Adds TCP keepalive support for more reliable reads, and increases the default buffer size to enhance write throughput for writable external tables.

• ORCA now falls back to the planner for queries involving foreign partitions using greenplum_fdw, preventing crashes caused by incompatible execution behavior. Queries on non-partitioned foreign tables using greenplum_fdw remain supported by ORCA.

High availability and high reliability​

Backup and disaster recovery​

• Improves archiver performance when handling many .ready files by reducing redundant directory scans. This change speeds up WAL archiving, especially when archive_command has been failing and many files have accumulated.

• gp_create_restore_point() can only be executed on the Coordinator node. Calling this function on a segment node will result in an error. The function returns a structured record value, including the restore point name and LSN, which you can view directly by running SELECT * FROM gp_create_restore_point().

WAL​

• Improves WAL replication management by restricting a coordinator-specific tracking mechanism to the coordinator only. This change simplifies primary segment behavior and aligns replication practices more closely across segments. No functional change for users, but helps reduce unnecessary complexity in WAL retention logic.

• Enhances WAL retention logic to improve reliability of incremental recovery using pg_rewind. Physical replication slots now retain WAL files up to the last common checkpoint, reducing risk of missing WAL during recovery. This change also simplifies the underlying logic and adds test coverage for WAL recycling.

• Switches WAL replication connections to use the standard libpq protocol instead of a legacy internal one. This improves compatibility and reliability of replication behavior. Also fixed test failures and improved error handling for replication connections.

Security​

DB Operations​

• REFRESH MATERIALIZED VIEW CONCURRENTLY runs all internal operations in the correct security context to prevent potential privilege escalation. This change ensures safer execution by restricting operations to the appropriate permission level.

• Improves internal handling of new aoseg and aocsseg tuples by aligning tuple freezing behavior with other catalog operations. This change enhances consistency with upstream PostgreSQL practices and removes the need for CatalogTupleFrozenInsert.

System processes​

• Orphaned file checks now exclude idle sessions during safety validation. This prevents unnecessary errors when persistent connections from services are active, allowing the detection process to complete successfully.

• Adds a safety check in backend signal handlers to ensure signals are handled by the correct process. This prevents unintended shared memory access by child processes and improves overall process isolation and stability.

• Improves process safety by preventing child processes spawned via system() from calling proc_exit(). This avoids potential corruption of shared memory structures and ensures only the parent process performs cleanup operations.

• Removes the permission check for cpu.pressure when using gp_resource_manager='group-v2'. This prevents startup failures on systems where PSI is disabled, without affecting resource management functionality.

Replication/Mirrorless clusters​

• Improves replication error reporting by setting persistent WalSndError when a replication slot is invalidated. This ensures accurate error visibility in gp_stat_replication.

Permission management​

• Strengthens security by rejecting extension schema or owner substitutions containing potentially unsafe characters like $, ', or \. This prevents SQL injection in extension scripts and protects against privilege escalation in certain non-bundled extensions.

• Creating or assigning roles to the system_group resource group now results in an error, as this group is reserved for internal system processes only.

• Reverts the restriction requiring superuser privileges to set the gp_resource_group_bypass GUC. This allows applications like GPCC to function more easily while still limiting resource impact.

• Altering the mpp_execute option of a foreign server or wrapper is now disallowed to prevent inconsistencies in foreign table distribution policies. Changing these options previously could result in outdated cached plans and incorrect query execution. This update ensures plan correctness by enforcing cache invalidation only when appropriate.

pgcrypto​

• Adds support for FIPS mode in pgcrypto, controlled by a GUC. This allows Cloudberry to operate in FIPS-compliant environments when linked with a supported FIPS-enabled OpenSSL version. Certain ciphers are disabled in this mode to comply with FIPS requirements.

• pgcrypto now allows enabling FIPS mode even on systems where FIPS is not pre-enabled by the OS or environment. This change removes the dependency on FIPS_mode() checks, offering more flexibility in managing FIPS compliance through the database.

Resource management​

Resource group management​

• Renames the memory_limit parameter to memory_quota in CREATE/ALTER RESOURCE GROUP to clarify its meaning and unit.

• Adds a new system view gp_toolkit.gp_resgroup_status_per_segment to monitor memory usage per resource group on each segment. This view helps database administrators track real-time vmem consumption (in MB) when resource group-based resource management is enabled.

• Improves logging behavior when memory usage reaches Vmem or resource group limits. The system now prints log messages directly to stderr to avoid stack overflow errors during allocation failures.

• Removes unnecessary permission check for cpu.pressure when using the group-v2 resource manager. This prevents startup failures on systems where PSI is not enabled, improving compatibility across Linux distributions.

Logging and monitoring​

• Adds additional log messages for GDD backends to help investigate memory-related issues. These logs provide better visibility into backend behavior during high memory usage scenarios.

• Adds a log ignore rule for "terminating connection" messages to reduce noise in test outputs. This helps avoid unnecessary diffs in CI for tests that involve connection termination.

• Adds more verbose logging to ResCheckSelfDeadlock().

• Logs queue IDs and portal IDs in resource queue logs.

• Dumps more information when releasing resource queue locks to aid in troubleshooting and monitoring.

• Uses ERROR for dispatcher liveness checks.

• Enhances logging for dispatch connection liveness checks to improve clarity during connection failures. Logs now include more accurate error messages based on socket state and system errors.

Platform compatibility and build​

• Improves gp_sparse_vector compatibility with ARM platforms by fixing type handling in serialization logic. This ensures consistent behavior across different architectures.

• Adds support for sigaction() on Windows to align signal handling behavior with other platforms. This reduces platform-specific differences and improves code consistency.

• Updates ACL mode type in ORCA to match the parser's definition, ensuring consistent type usage.

System views and statistics​

• Improves join cardinality estimation for projected columns that preserve the number of distinct values (NDVs), such as additions or subtractions with constants. This allows the optimizer to use underlying column histograms for more accurate estimates, improving plan quality for queries with scalar projections in join conditions.

• Increases precision for frequency and NDV values in ORCA when processing metadata population scripts (MDPs). This change ensures consistent behavior between MDPs and live database queries, reducing discrepancies caused by rounding small values.

• ORCA now considers null value skew when costing redistribute motions, improving plan accuracy for queries involving columns with many nulls. This helps avoid performance issues caused by data being unevenly distributed across segments.

• ORCA now supports extended statistics to improve cardinality estimation for queries with correlated columns. This allows the optimizer to use real data-driven correlation factors instead of relying on arbitrary GUC settings, leading to more accurate query plans.

• Introduces gp_log_backend_memory_contexts to log memory contexts across segments, with optional targeting by content ID. This enhances observability and helps diagnose memory issues in distributed queries.

• ORCA now supports statistics derivation for predicates involving different time-related data types, such as date and timestamp. This improves plan accuracy and performance for queries comparing mixed temporal types.

• Autostats now uses SKIP LOCKED for ANALYZE operations to avoid blocking on locks, reducing the risk of deadlocks and improving predictability. This behavior is enabled by default and can be controlled using the gp_autostats_lock_wait GUC.

• ORCA now supports STATS_EXT_NDISTINCT extended statistics for estimating cardinality on correlated columns. This improves accuracy for queries using GROUP BY or DISTINCT on such columns.

Network connections​

• Marks gp_reject_internal_tcp_connection as defunct to improve reliability of internal QD-to-entry DB connections. These connections over TCP/IP are now treated as authenticated by default, preventing authentication errors caused by pg_hba.conf settings.

Tools and utilities​

analyzedb​

• analyzedb now includes materialized views in its list of tables to analyze. This improves the performance immediately after analysis.

gpexpand​

• gpexpand now includes a cluster health check to ensure all segments are up and in their preferred roles before proceeding. This prevents incorrect port assignments and avoids potential issues during expansion when nodes are not in a stable state.

gp_toolkit​

• Added an update path for the gp_toolkit extension to version 1.6. This update renames the column memory_limit to memory_quota in the gp_resgroup_config view for improved clarity. Users can apply the update using ALTER EXTENSION gp_toolkit UPDATE TO '1.6'.

Bug fixes​

• Fixed data loss caused by incorrect shared snapshot handling.
• Fixed memory corruption during AOCO ADD COLUMN abort.
• Fixed checkpoint WAL replay failure.
• Fixed incorrect results when using UNION for RECURSIVE_CTE.
• Fixed incorrect results from hash joins on char columns.
• Fixed incorrect results produced by WITH RECURSIVE queries.
• Fixed incorrect results when a REPLICATED table is unioned with a DISTRIBUTED table.
• Fixed incorrect results when the outer query had ORDER BY after a LATERAL subquery.
• Fixed incorrect behavior of DELETE with split update.
• Fixed incorrect results when using direct dispatch.
• Fixed memory leaks in ORCA and various components.
• Fixed long-running execution with bitmap indexes.
• Fixed redundant SORT enforcement on group aggregates.
• Fixed incorrect index position in target list in ExecTupleSplit.
• Fixed incorrect value in the cpu_usage column returned by pg_resgroup_get_status().
• Fixed incorrect behavior of gp_toolkit.gp_move_orphaned_files.
• Fixed incorrect results in multi-stage aggregate queries.
• Fixed incorrect plan and output in multi-stage aggregate queries.
• Fixed incorrect reltuples value after VACUUM.
• Fixed incorrect index->reltuples value after VACUUM.
• Fixed a vulnerability where LDAP leaked user information.
• Fixed incorrect permissions warning on the pgpass file.
• Fixed incorrect handling of ONLY keyword for multiple tables in GRANT/REVOKE statements.
• Fixed incorrect permissions in resource management DDL.
• Fixed incorrect security context in REFRESH MATERIALIZED VIEW CONCURRENTLY.
• Fixed deadlock between coordinator and segments.
• Fixed race condition in CTE reader-writer communication.
• Fixed race condition when invalidating obsolete replication slots.
• Fixed deadlock by allowing concurrent creation of non-first indexes on AO tables.
• Fixed locking issue when opening range tables inside ExecInitModifyTable().
• Fixed incorrect unlock mode in DefineRelation.
• Fixed incorrect locking in partition distribution policies.
• Fixed issues with rle_type when converting a table from AO to AOCO.
• Fixed incorrect handling of empty ranges and NULL values in BRIN indexes.
• Fixed incorrect handling of NULL values when merging BRIN summaries.
• Fixed incorrect TIDs order when building bitmap indexes.
• Fixed possible inconsistency between bitmap LOV table and its index.
• Fixed incorrect behavior of VACUUM in AO tables with indexes.
• Fixed incorrect handling of TOAST values for invisible AppendOptimized tuples during VACUUM.
• Fixed ORCA's invalid processing of nested SubLinks under aggregates.
• Fixed ORCA's invalid processing of nested SubLinks referenced in GROUP BY clauses.
• Fixed ORCA's invalid processing of nested SubLinks with GROUP BY attributes.
• Fixed incorrect predicate pushdown when using casted columns.
• Fixed incorrect join condition loss after pulling up sublinks to join nodes.
• Fixed incorrect hash-key generation for Redistribute Motion in multi-DQA expressions.
• Fixed incorrect plan generation for SEMI JOIN with RANDOM distributed tables.
• Fixed incorrect behavior of gp_stat_bgwriter.
• Fixed incorrect monitoring in pg_stat_slru.
• Fixed incorrect monitoring in gp_stat_progress_dtx_recovery.
• Fixed incorrect monitoring in pg_resgroup_get_status().
• Fixed incorrect monitoring in gp_toolkit.gp_resgroup_config.
• Fixed compilation issues on various platforms.
• Fixed documentation and comment typos.
• Fixed build system and Makefile issues.
• Fixed various memory leaks and resource management issues.
• Fixed various error handling and logging improvements.
• Fixed mismatched types.
• Fixed the ORCA preprocess step for queries with the Select-Project-NaryJoin pattern.
• Fixed the missing discard_output variable in shared scan node functions.
• Fixed the crash caused by running VACUUM AO_AUX_ONLY on an AO-partitioned table.
• Fixed an obvious memory leak in _bitmap_xlog_insert_bitmapwords().
• Fixed a memory leak in the merge join implementation.
• Fixed the issue where the token for user ID xxx did not exist.
• Fixed the issue where plan hints could not derive table descriptors.
• Fixed the issue where inject_fault suspend could not be canceled.
• Fixed fallback in debug builds due to scalars with invalid return types.
• Fixed relptr encoding of the base address.
• Fixed visimap consults for unique checks during UPDATE operations.
• Fixed the issue where external table location URIs containing | caused errors.
• Fixed handling of the time command output containing commas.
• Fixed a small overestimation of the output length of base64 encoding.
• Fixed gp_toolkit.__gp_aocsseg_history crash on non-AO columnar tables.
• Fixed a race condition between termination and resqueue wakeup.
• Fixed a statement leak involving self-deadlocks.
• Fixed the detection of child output columns when the parent is a UNION during join pruning.
• Fixed a query crash when using a negative memory_limit value in resource groups.
• Fixed issues in pgarch new directory-scanning logic.
• Fixed a memory leak in the FTS PROBE process.
• Fixed check_multi_column_list_partition_keys.
• Fixed a memory leak caught via ICW with memory check enabled.
• Fixed query hang and fallback issues involving CTEs on replicated tables.
• Fixed the unrecognized join type error with LASJ Not-In and network types.
• Fixed issues in upgrade_adapt.sql related to queries using WITH OIDS.
• Fixed the double declaration of check_ok() in pg_upgrade.h.
• Fixed logic error with subdirectories generated by pg_upgrade for internal files.
• Fixed a typo in the pg_upgrade file header.
• Fixed the bug where PL/Python functions caused the master process to reset.
• Fixed the Shared Scan hang issue involving initplans.
• Fixed motion toast error.
• Fixed a memory leak related to fsync in AO tables.
• Fixed CDatumSortedSet handling of empty arrays that caused errors in ORCA.
• Fixed ORCA returning incorrect column type modifier information.
• Fixed DbgStr output when printing DP structs in ORCA.
• Fixed the comment on performDtxProtocolPrepare.
• Fixed a memory leak in Dynamic Index, IndexOnly, and BitmapIndex scans during execution.
• Fixed the memory accounting bug when moving MemoryContext under another accounting node.
• Fixed the ALTER TABLE ALTER COLUMN TYPE issue that reuses an incorrect index.
• Fixed query fallback when a subquery is present within LEAST() or GREATEST().
• Fixed the typo in timestamp.
• Fixed unexpected warnings related to pg_stat_statements node types.
• Fixed the crash involving initplan in MPP.
• Fixed LeftJoinPruning pruning essential LEFT JOINs.
• Fixed the SET command that incorrectly sends DTX protocol commands.
• Fixed the segmentation fault in addOneOption().
• Fixed parallel_retrieve_cursor diffs.
• Fixed gpdiff.pl to ignore information when EXPLAIN ignores costs.
• Fixed the uninitialized-use warning in CTranslatorDXLToPlStmt.cpp.
• Fixed the bug where the LOCALE flag cannot be used with a string pattern.
• Fixed a typo in cdbmutate.c.
• Fixed CColRefSet debug printing.
• Fixed ORCA producing incorrect plans when handling SEMI JOIN with RANDOM distributed tables.
• Fixed orphaned temp tables on the coordinator.
• Fixed the segmentation fault caused by concurrent INSERT ON CONFLICT and DROP TABLE.
• Fixed redundant columns in a multi-stage aggregate plan.
• Fixed the import of ICU collations in pg_import_system_collations().
• Fixed the error: "Cannot add cell to table content: total cell count of XXX exceeded."
• Fixed orphaned temporary namespace catalog entries left on the coordinator.
• Fixed REFRESH MATERIALIZED VIEW on AO tables with indexes.
• Fixed the use of PORTNAME in the gp_toolkit Makefile.
• Fixed pg_stat_activity display for bypassed and unassigned queries.
• Fixed the recursive CTE MergeJoin that involved a motion on WTS.
• Fixed the column width display for partitioned tables.
• Fixed the LDAP crash when ldaptls=1 and ldapscheme is not set.
• Fixed the gpstop pipeline flakiness after the referenced change.
• Fixed the ANALYZE bug in expand_vacuum_rels.
• Fixed the compilation error.
• Fixed the ORCA crash due to improper colref mapping with CTEs.
• Fixed the bug where gpload insert mode was not included in a transaction.
• Fixed the bug where resgroup total wait time was always zero.
• Fixed the gpcheckcat error against pg_description.
• Fixed flakiness caused by waiting for a different number of fault triggers.
• Fixed the bug involving RelabelType in the GROUP BY clause.
• Fixed the planner error with multiple copies of an AlternativeSubPlan.
• Fixed the issue with bitmap indexes.
• Fixed the bug in HashAgg related to selective-column-spilling logic.
• Fixed the bug in disk-based hash aggregation.
• Fixed the pipeline stall issue in LookupTupleHashEntryHash().
• Fixed the use of version in ArgumentParser, which is deprecated.
• Fixed the use of BaseException.message, which has been deprecated since Python 2.6.
• Fixed the case pg_rewind_fail_missing_xlog.
• Fixed the compiler warning for gcc-12.
• Fixed support for the DEFERRABLE keyword on primary and unique keys.
• Fixed the unlocking of pruned partitions in partitioned tables.
• Fixed the crash in ORCA involving skip-level correlated queries.
• Fixed the removal of Assert statements in release builds.
• Fixed the typo in comments: JOIN_SEMI_DEDUP/JOIN_SEMI_DEDUP_REVERSE.
• Fixed the issue where REORGANIZE=TRUE did not redistribute randomly-distributed tables.
• Fixed the core dump caused by concurrent updates on partition tables in DynamicScan.
• Fixed the typo: ANALZE to ANALYZE.
• Fixed the issue where cgroup v1 cpu_quota_us cannot be larger than its parent's value.
• Fixed indentation and trailing whitespace in UDFs in resgroup/resgroup_auxiliary_tools_v1.
• Fixed the name of cpu_hard_quota_limit in resgroup_syntax.sql.
• Fixed multi-row DEFAULT handling in INSERT ... SELECT rules.
• Fixed invalid function references in several comments.
• Fixed the bug where COPY FORM does not throw ERROR: extra data after last expected column.
• Fixed the issue where file .204800 was not being checked in ao_foreach_extent_file.
• Fixed the issue of incorrectly incrementing the command counter.
• Fixed the coordinator crash in MPPnoticeReceiver.
• Fixed the dangling pointer in ExecDynamicIndexScan().
• Fixed the ORCA bug that incorrectly removed required redistribution motion when using GROUP BY over gp_segment_id.
• Fixed header handling in url_curl.c.
• Fixed ao_filehandler to support new attnum to filenum mapping changes.
• Fixed pg_aocsseg to work with attnum to filenum mapping.
• Fixed a comment in pg_dump.
• Fixed the ORCA build break.
• Fixed the gpconfig SSH retry undefined parameter issue.
• Fixed the stale gp_default_storage_options comment.
• Fixed the bug: unrecognized node type: 147.
• Fixed spelling errors identified by lintian.
• Fixed the bypass catalog unit test.
• Fixed erroneous Valgrind markings in AllocSetRealloc.
• Fixed the legacy bug in the DatabaseFrozenIds lock.
• Fixed the mirror checkpointer error on the ALTER DATABASE query.
• Fixed the bug: get_ao_compression_ratio() failed on root partitioned tables with AO children.
• Fixed the issue where InterruptHoldoffCount was not being reset.
• Fixed gpexpand failure caused by an event trigger.
• Fixed missing redistribute for CTAS or INSERT INTO on randomly distributed tables when using ORCA.
• Fixed the double free of remapper->typmodmap in TeardownUDPIFCInterconnect().
• Fixed the bug in the upstream-merged COMMIT AND CHAIN feature.
• Fixed inconsistency between gp_fastsequence row and index after a crash.
• Fixed the typo allocatd to allocated.
• Fixed the error: unrecognized node type: 145 in transformExpr.
• Fixed build error caused by unused variable.
• Fixed the issue where the distribution key was missing when creating a stage table.
• Fixed the regex for etc/environment.d.
• Fixed the string comparison warning.
• Fixed obsolete references to SnapshotNow in comments.
• Fixed pull-up error when the target list contains a RelabelType node.
• Fixed the issue where index DDL operations were recorded in QEs' pg_last_stat_operation.
• Fixed two compiler warnings.
• Fixed the wrong value of maxAttrNum in TupleSplitState.
• Fixed the bug of incorrect index position in target list in ExecTupleSplit.
• Fixed the format error of the library name on Mac M1.
• Fixed the pg_resgroup_get_status_kv() function.
• Fixed interconnect bugs in ic_proxy_ibuf_push().
• Fixed memory leaks in auto_explain.
• Fixed ic_proxy compilation when HOST_NAME_MAX is unavailable.
• Fixed duplicate filters caused by reversed operator argument order.
• Fixed pg_rewind when the log file is a symbolic link.
• Fixed and enabled 64-bit bitmapset and updated visimap.
• Fixed the hang caused by multi-DQA with filters in the planner.
• Fixed the bogus ORCA plan that incorrectly joins a CTE and a REPLICATED table.
• Fixed the error in ATSETAM when applied to ao_column with a dropped column.
• Fixed the LWLockHeldByMe assert failure in SharedSnapshotDump.
• Fixed the KeepLogSeg() unit test.
• Fixed the race condition when invalidating obsolete replication slots.
• Fixed the uninitialized value in segno calculation.
• Fixed issues in the invalidation logic for obsolete replication slots.
• Fixed checkpoint signalling.
• Fixed memory overrun when querying pg_stat_slru.
• Fixed the bug where ORCA fails to decorrelate subqueries ordered by outer references.
• Fixed unused variable compile warnings.
• Fixed the bug where NestLoop join fails to materialize the inner child in some cases.
• Fixed COPY execution via FDW on coordinator as executor.
• Fixed inFunction usage for auto_stats in CTAS.
• Fixed a compiler warning.
• Fixed the syntax error with CREATE MATERIALIZED VIEW.
• Fixed the issue preventing temporary table creation LIKE existing tables with comments.
• Fixed and rewrote IndexOpProperties API.
• Removed redundant Get/SetStaticPruneResult usage.
• Fixed EPQ handling for DML operations.
• Fixed gpcheckperf failure when using -V with -f option.
• Fixed possible mirror startup failure triggered by FTS promotion.
• Fixed the parallel retrieve cursor issue when selecting transient record types.
• Fixed the resource management DDL warning: unrecognized node type when log_statement='ddl'
• Fixed the resgroup init error when many cores are present in cpuset.cpus.
• Fixed resqueue malfunction when using JDBC extended protocol.
• Fixed the missing LOCKING CLAUSE on foreign tables when ORCA is enabled.
• Fixed the test_consume_xids behavior where it consumes one more transaction ID than expected.
• Fixed the ONLY keyword handling for multiple tables in GRANT/REVOKE statements.
• Fixed the regression test to ignore memory usage values in JSON format EXPLAIN output.
• Fixed relcache lookup in ORCA when selecting from sequences.
• Fixed missing WAL files required by pg_rewind.
• Fixed the gp_dqa test to explicitly ANALYZE tables.
• Fixed the crash of AggNode in the executor caused by an ORCA plan.
• Fixed the resource group cpuset test case.
• Fixed the compiler warning caused by gpfdist with compressed external tables.
• Fixed link issues on macOS and Windows.
• Fixed failure when DynamicSeqScan contains a SubPlan.
• Fixed the error: cache lookup failed for type 0.
• Fixed the multi-level correlated subquery bug.
• Fixed checkpoint WAL replay failure.
• Fixed the check for BufFileRead() in ExecHashJoinGetSavedTuple().
• Fixed the test extension to allow executing SQL code inside a Portal.
• Fixed resgroup view test cases.
• Fixed incorrect DISTKEY assignment when copying partitions on segments.
• Fixed ic-proxy mis-disconnecting addresses after reloading the config file.
• Fixed the gpcheckcat check on partition distribution policies.
• Fixed colid remapping in disjunctive constraints.
• Fixed the Makefile by removing the tablespace-step target from all.
• Fixed CBitSet intersection logic in ORCA.
• Fixed the query preprocessor for nested Select-Project-NaryJoin patterns.
• Fixed incorrect unlock mode in DefineRelation.
• Fixed the upgrade process for external tables with dropped columns.
• Fixed the formatting issue in SECURITY.md.
• Fixed gp_gettmid to return the correct startup timestamp.
• Fixed the gpload regression test failure when the OS user is not gpadmin.
• Fixed the compiler warning in appendonlyblockdirectory.c.
• Fixed missing reloptions in partition roots created using Cloudberry syntax.
• Fixed the crash when calling get_ao_compression_ratio on HEAP tables.
• Fixed incorrect sortOp and eqOp values generated by IsCorrelatedEqualityOpExpr.
• Fixed the dependency bug involving minirepro and materialized views.
• Fixed recursion handling in ALTER TABLE ... ENABLE/DISABLE TRIGGER.
• Fixed SPE plans to display Partitions selected: 1 (out of 5).
• Fixed incorrect hash-key generation for Redistribute Motion when creating paths for multi-DQA expressions.
• Removed gp_enable_sort_distinct and noduplicates optimizations.
• Fixed gpinitsystem Behave tests that use environment variables.
• Fixed false alarms in gpcheckcat for pg_default_acl.
• Fixed gpinitsystem failure with custom locale settings.
• Fixed a panic in the greenplum_fdw test.
• Fixed the failure in bitmap index null-array condition.
• Fixed the compilation warning in gram.y.
• Fixed multiple issues related to DistributedTransaction handling.
• Fixed compile-time warnings in pg_basebackup code.
• Fixed gplogfilter to correctly generate CSV output.
• Fixed the assert in the OpExecutor node.
• Fixed improper copying of group statistics in ORCA.
• Fixed error reporting after ioctl() call in pg_upgrade --clone mode.
• Fixed replay of CREATE DATABASE records on standby.
• Fixed a minor memory leak in pg_dump.
• Fixed parallel restore of foreign keys to partitioned tables.
• Fixed the issue where the pg_appendonly entry was not removed during AO-to-HEAP table conversion.
• Fixed assertion failure and segmentation fault in the backup code.
• Fixed fallback behavior for non-default collations.
• Fixed the subtransaction test for Python 3.10.
• Fixed Windows client compilation of libpgcommon.
• Fixed compiler warnings introduced by the Dynamic Scan commit.
• Fixed the issue where CREATE OR REPLACE TRANSFORM failed.
• Fixed compiler warnings for non-assert builds.
• Fixed lock assertions in dshash.c.
• Fixed \watch interaction with libedit on C.