> For the complete documentation index, see [llms.txt](https://mariadb.com/docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://mariadb.com/docs/release-notes/gridgain-9/9.1/release-notes_9.1.28.md).

# GridGain 9.1.28 Release Notes

GridGain 9.1.28 hardens distributed map, thin-client metadata, and LDAP role mapping security, fixes the REST transactions API path, and deprecates RocksDB partition storage. Released on 2026-10-20.

**Release date:** 20 Oct 2026

## Overview

GridGain 9.1.28 hardens security around the distributed map, thin-client metadata, and LDAP role mapping, corrects the management REST transactions API path and deprecates the RocksDB partition storage engine.

## Major Changes

### Security Hardening

Several access checks that were previously missing are now enforced. Review them before upgrading, because a client or role that relied on the earlier, looser behavior may begin to be denied:

* Thin-client index and table metadata requests now check privileges, so a client can no longer read index or table metadata it has no privilege to see.
* The distributed map API is now checks user privileges.
* Unmapped LDAP group names are no longer trusted as GridGain roles.
* Renaming the cluster now requires the new `RENAME_CLUSTER` privilege, so a routine authenticated user can no longer rename the cluster.
* Enabling, disabling, or listing cluster-wide metrics now requires the corresponding metrics privilege (`ENABLE_METRICS`, `DISABLE_METRICS`, or `LIST_METRICS`).

### REST Transactions API Path Corrected

The management REST API served its transactions endpoints on a path that differed from the published OpenAPI specification. The endpoints are now served on the path the specification documents. A client that hard-coded the previous path must be updated. See [management REST API](https://mariadb.com/docs/gridgain/gridgain-9/gridgain9-development/accessing-data/rest-api.md).

### Deprecations and Removals

The RocksDB partition storage engine is now deprecated and will be removed in a future release. Migrate profiles that use it to the `aipersist` or `aimem` storage engines. See [RocksDB storage engine](https://mariadb.com/docs/gridgain/gridgain-9/architecture/storage/engines/rocksdb-persistent-storage.md).

## New Features

### .NET Entity Framework Core Provider

The .NET client now ships an Entity Framework Core provider, letting you use GridGain through EF Core.

### Receiver Execution Options in the Java Thin Client

The Java thin client data streamer now accepts `ReceiverExecutionOptions`, giving you control over how a stream receiver runs. See [Data Streamer](https://mariadb.com/docs/gridgain/gridgain-9/gridgain9-development/accessing-data/data-streaming.md).

### LDAP Connect and Read Timeouts

LDAP binds are now bounded by JNDI connect and read timeouts, so a slow or unresponsive directory server can no longer stall authentication indefinitely. See [Authentication](https://mariadb.com/docs/gridgain/gridgain-9/security/authentication-and-authorization/authentication.md).

## Improvements and Fixed Issues

| Issue ID | Category                   | Description                                                                                                                                                                 |
| -------- | -------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| GG-52265 | Cluster SQL Engine         | Fixed an issue where the MERGE operator failed under some conditions.                                                                                                       |
| GG-51913 | General                    | The management REST transactions API is now served on the path published in the OpenAPI specification.                                                                      |
| GG-51820 | Cluster Storage Engine     | Segstore log records are now validated before they are interpreted.                                                                                                         |
| GG-51795 | Cluster Storage Engine     | Fixed page replacement writing a page of a destroyed partition.                                                                                                             |
| GG-51772 | Cluster Storage Engine     | Raft log storages opened by a start after node stop has begun are now closed.                                                                                               |
| GG-51758 | General                    | Absent logical topology members are now probed before they are removed.                                                                                                     |
| GG-51757 | General                    | A restarted node's previous launch is now removed from the logical topology.                                                                                                |
| GG-51756 | General                    | Logical topology leaps are now propagated to the network topology service.                                                                                                  |
| GG-51738 | Cluster Storage Engine     | The write-intents list head is now kept when unlinking a detached row version.                                                                                              |
| GG-51737 | General                    | Bounded the recovery queue and the sends waiting for a channel.                                                                                                             |
| GG-51721 | Cluster Storage Engine     | The RocksDB partition storage engine is now deprecated.                                                                                                                     |
| GG-51675 | Cluster Continuous Queries | Pending continuous query requests now fail when the target node leaves.                                                                                                     |
| GG-51598 | Migration Tools            | Fixed the migration-tools CLI crashing instead of printing its version.                                                                                                     |
| GG-51595 | General                    | Fixed the wrong MBean attribute type for UuidGauge metrics.                                                                                                                 |
| GG-51565 | General                    | Fixed Closeable views #close delegation.                                                                                                                                    |
| GG-51516 | General                    | Fixed a race in the log storage size limiter.                                                                                                                               |
| GG-51505 | Cluster Continuous Queries | Continuous Query: Fixed poll delay on new demand.                                                                                                                           |
| GG-51493 | Cluster SQL Engine         | Fixed statistics refresh and plan invalidation reliability.                                                                                                                 |
| GG-51492 | Migration Tools            | The migration-tools adapter no longer ships the AI3 API stub JAR on runtime classpaths.                                                                                     |
| GG-51491 | Platforms & Clients        | The client channel now stays alive when a server error class cannot be linked.                                                                                              |
| GG-51439 | Cluster SQL Engine         | Continuations of key-value plans now execute on the completing thread when possible.                                                                                        |
| GG-51436 | Platforms & Clients        | C++ client: A send operation now reports failure on EAGAIN while the packet stays queued.                                                                                   |
| GG-51425 | Cluster SQL Engine         | Fused a trimming projection into the required columns for primary-key lookups.                                                                                              |
| GG-51422 | Platforms & Clients        | Java thin client: Added support for ReceiverExecutionOptions in the data streamer.                                                                                          |
| GG-51418 | General                    | Resources acquired by a bootstrap compute job are now released.                                                                                                             |
| GG-51416 | Migration Tools            | The migration-tools adapter is now deployed as three deployment units instead of one per JAR.                                                                               |
| GG-51391 | Cluster Security           | The authentication thread pool is now sized from authentication.threadPoolSize.                                                                                             |
| GG-51390 | Cluster Security           | LDAP binds now honor JNDI connect and read timeouts.                                                                                                                        |
| GG-51388 | Cluster Storage Engine     | Partition storages left in the rebalance state are now cleared.                                                                                                             |
| GG-51386 | General                    | Reduced write latency when a partition's primary replica and Raft leader are on different nodes.                                                                            |
| GG-51377 | Platforms & Clients        | C++: Fixed a race in the Windows network code.                                                                                                                              |
| GG-51284 | General                    | A failing cloud license probe no longer stops node start.                                                                                                                   |
| GG-51189 | Migration Tools            | Updated the GridGain 8 dependency to 8.9.37.                                                                                                                                |
| GG-51135 | Platforms & Clients        | Fixed corner cases in direct client transaction error processing.                                                                                                           |
| GG-51131 | General                    | Tables are now enlisted into an existing partition enlistment under the enlist lock.                                                                                        |
| GG-50977 | Platforms & Clients        | C++: Transaction commit is now sent in the last message of the transaction.                                                                                                 |
| GG-50940 | Cluster Continuous Queries | Removed the deprecated ContinuousQueryTransactionWatermark.transaction() method.                                                                                            |
| GG-50921 | Platforms & Clients        | Fixed a direct-mapped thin-client transaction leaking a lock when aborted before its enlistment reached the coordinator.                                                    |
| GG-50918 | General                    | Catalog compaction now responds to the coordinator even when an unexpected exception occurs.                                                                                |
| GG-50902 | Cluster Data Replication   | The data replication connector now stores multiple entries within a single transaction.                                                                                     |
| GG-50861 | Platforms & Clients        | Thin client: Fixed heartbeatInterval validation.                                                                                                                            |
| GG-50772 | Platforms & Clients        | C++ extension: Catalog-snapshot schemas are now published into the client cache.                                                                                            |
| GG-50629 | Cluster Storage Engine     | The Raft meta storage of evicted partitions is now destroyed when their zone is dropped.                                                                                    |
| GG-50488 | Cluster Data Replication   | The DR connector now logs a warning if metadata has not arrived within 30 seconds.                                                                                          |
| GG-50381 | Cluster Security           | Unmapped LDAP group names are no longer trusted as GridGain roles.                                                                                                          |
| GG-50377 | Cluster Security           | The distributed map API is now secured with privilege checks.                                                                                                               |
| GG-50376 | Cluster Security           | Thin-client index and table metadata requests now check privileges.                                                                                                         |
| GG-50375 | Platforms & Clients        | Added JDBC error codes for transaction serialization failures.                                                                                                              |
| GG-50358 | General                    | The Catalog DSL now reads a zone name the way CREATE ZONE does.                                                                                                             |
| GG-50250 | Cluster Storage Engine     | Evicted partition replicas are now destroyed on recovery.                                                                                                                   |
| GG-50208 | Cluster Data Replication   | Transaction cleanup is now moved off the saturated pool, preventing a full-state-transfer stall on large tables.                                                            |
| GG-50070 | General                    | Abandoned resources for destroyed zones are now handled in the catalog.                                                                                                     |
| GG-49982 | General                    | Fixed a rare case where a primary replica change during transaction cleanup could leave duplicate rows for the same primary key.                                            |
| GG-49937 | Cluster SQL Engine         | An implicit SQL transaction is now finalized when plan validation fails.                                                                                                    |
| GG-49878 | Platforms & Clients        | Java client: Fixed compatible schema-change retry within a transaction.                                                                                                     |
| GG-49726 | Cluster Security           | LDAP credential rejections are now classified as expected authentication failures.                                                                                          |
| GG-49675 | General                    | Reduced the zone-creation safe-time stall under fsync.                                                                                                                      |
| GG-49621 | General                    | Compute: Job, task, and receiver classes are now resolved without initializing them before the type check.                                                                  |
| GG-49619 | Cluster Security           | Renaming the cluster now requires the new RENAME\_CLUSTER privilege.                                                                                                        |
| GG-49614 | Cluster Security           | Enabling, disabling, or listing cluster-wide metrics now requires the corresponding metrics privilege.                                                                      |
| GG-49163 | Platforms & Clients        | C++: The client now returns the same Unknown Error Code as other clients.                                                                                                   |
| GG-48995 | General                    | Fixed a possible ignoring of a write intent by a late-committed transaction.                                                                                                |
| GG-48531 | General                    | MetaStorage unavailability is no longer treated as a critical failure.                                                                                                      |
| GG-48180 | Cluster Storage Engine     | The GC/vacuum pipeline now resolves a null column gracefully instead of failing with a critical error when processing orphaned write intents after a schema version change. |
| GG-47996 | Cluster Continuous Queries | Fixed a continuous query event batch being reused asynchronously in the near-cache subscriber.                                                                              |
| GG-44292 | Platforms & Clients        | C++: Added a source-code distribution package for building on non-glibc or non-x86-64 Linux.                                                                                |
| GG-42272 | General                    | Improved how nodes choose their advertised addresses on machines with more than one network interface.                                                                      |

## Upgrade Information

You can upgrade to current GridGain version from previous releases. Below is a list of versions that are compatible with the current version. Compatibility with other versions is not guaranteed. If you are on a version that is not listed, contact GridGain for information on upgrade options.

`9.1.21`, `9.1.22`, `9.1.23`, `9.1.24`, `9.1.25`, `9.1.26`, `9.1.27`

When updating from older versions, we recommend updating to version `9.1.21` first, before performing an update to current version.

## Known Limitations

### Data Restoration After Data Rebalance

Currently, data rebalance may cause partition distribution to change and cause issues with snapshots and data recovery. In particular:

* It is currently not possible to restore a `LOCAL` snapshot if data rebalance happened after snapshot creation. This will be addressed in one of the upcoming releases.
* It is currently not possible to perform point-in-time recovery if data rebalance happened after table creation. This will be addressed in one of the upcoming releases.

## We Value Your Feedback

Your comments and suggestions are always welcome. You can reach us here: <https://mariadb.com/support/>.

<sub>*This page is: Copyright © 2026 MariaDB. All rights reserved.*</sub>

{% @marketo/form formId="4316" %}
