GridGain 8.9.36 Release Notes

Overview

GridGain 8.9.36 adds graceful node drain over REST and bulk cache destroy in the Java thin client, along with performance, replication, and security fixes.

Deprecation Notice

Kubernetes Operator

Starting with this release, GridGain Operator for Kubernetes is deprecated. It will be replaced by a new Operator in one of the upcoming releases.

New Features

Graceful Node Drain via REST

With this release, the REST API provides a drain command that sets, clears, or reads a node’s drain flag — a Kubernetes pod-lifecycle switch for graceful shutdown. Use it to take a node out of readiness rotation before terminating it, so that traffic drains away cleanly during rolling maintenance and upgrades. The command supports three actions:

start - sets the drain flag.
stop - clears the drain flag, returning the node to normal readiness behavior.
status - returns whether the drain flag is currently set.

http://host:port/ignite?cmd=drain&action=start
http://host:port/ignite?cmd=drain&action=stop
http://host:port/ignite?cmd=drain&action=status

For more information about the REST API, see the Drain command documentation.

Bulk Cache Destroy in the Java Thin Client

With this release, the Java thin client can destroy several caches in a single call with the new IgniteClient.destroyCaches(Collection<String>) method and its asynchronous counterpart IgniteClient.destroyCachesAsync(Collection<String>). This matches the thick-client API and removes the need to destroy caches one at a time.

client.destroyCaches(Arrays.asList("cache1", "cache2", "cache3"));

For more information, see Destroying Caches.

Skipping Cache Store in Java Thin Client

With this release, the Java thin client can bypass a configured cache store with the new ClientCache.withSkipStore() method. Operations performed on the returned cache instance do not read through to the store and do not write through to it.

ClientCache<Integer, String> cacheSkipStore = client.cache("myCache").withSkipStore();

// The cache store is not updated.
cacheSkipStore.put(1, "1");

For more information, see Skipping the Cache Store.

Improvements and Fixed Issues

Community Edition Changes

Issue ID	Category	Description
GG-50286	Platforms & Thin Clients	Java thin: added cache.withSkipStore.
GG-49983	Platforms & Thin Clients	Java thin: improved getAll performance by 2-3x with partition aware per-node split.
GG-49966	Cluster SQL Engine	Fixed inconsistent SQL query results in queries with duplicate group column and index sorting.
GG-49934	Cluster Storage Engine	Fixed a rare bug that could lead to hanging an optimistic transaction on node failure.
GG-49849	Cluster Storage Engine	Added a system property to configure maximum dirty pages ratio.
GG-49825	Platforms & Thin Clients	Java client: added destroyCaches.
GG-49804	Platforms & Thin Clients	Java thin client: fixed connection failure when server node attribute cannot be deserialized.
GG-49460	Diagnostics and Failure Handling	Improved distributed metastorage version logging.

Issue ID

Enterprise Edition Changes

Issue ID Category Description

Issue ID	Category	Description
GG-50207	Licenses	Fixed license CPU/host count inflation for co-located nodes that captured different MAC snapshots at startup; hosts are now grouped by shared MAC address.
GG-49998	Binary Objects	Fixed marshaller class-name filter bypass and added protection against oversized allocations during deserialization.
GG-49796	Cluster Deployment	Added `cmd=drain` REST command (start/stop/status) to gracefully drain a node before shutdown.
GG-49635	Cluster Data Replication	Keep DR traffic receiving components offline when cluster is not in ACTIVE state
GG-49147	Cluster Deployment	Added a new Docker distribution: ultimate-dotnet.
GG-48547	Cluster Data Replication	Added separate DR sender-hub metrics that counts only full-state transfer batches (BatchesSentFst, BatchesAckedFst, BatchesFailedFst and BatchesRejectedFst).

GG-50207

Licenses

Fixed license CPU/host count inflation for co-located nodes that captured different MAC snapshots at startup; hosts are now grouped by shared MAC address.

GG-49998

Binary Objects

Fixed marshaller class-name filter bypass and added protection against oversized allocations during deserialization.

GG-49796

Cluster Deployment

Added cmd=drain REST command (start/stop/status) to gracefully drain a node before shutdown.

GG-49635

Cluster Data Replication

Keep DR traffic receiving components offline when cluster is not in ACTIVE state

GG-49147

Cluster Deployment

Added a new Docker distribution: ultimate-dotnet.

GG-48547

Cluster Data Replication

Added separate DR sender-hub metrics that counts only full-state transfer batches (BatchesSentFst, BatchesAckedFst, BatchesFailedFst and BatchesRejectedFst).

Installation and Upgrade Information

See the Rolling Upgrades page for information about how to perform automated upgrades and for details about version compatibility.

Migrating from GridGain 8.9.17 or Earlier With Java 8

Starting GridGain 8.9.18, GridGain uses Lucene 9 by default. As Lucene 9 requires Java 11, it will be impossible to run GridGain on Java 8 with default configuration.

To continue using Java 8 when migrating to GridGain 8.9.18 or later, replace the {GRIDGAIN_HOME}/libs/ignite-lucene-9 folder with {GRIDGAIN_HOME}/libs/optional/ignite-lucene-8.

Migrating From GridGain 8.9.0

GridGain 8.9.1 introduced a large number of changes in default configuration values. Your setup may be affected if you are using default configuration.

When migrating to GridGain 8.9.36, make sure to check Changed Default Values in GridGain 8.9.1 and Later section for any parameters you need to change.

Migrating From GridGain 8.X

When migrating from GridGain 8.8 to GridGain 8.9, no special actions are required for core functionality migration. By using rolling upgrades, you can update from any GridGain version listed below. You may need to perform minor configuration changes to ensure stability post migration.

If you are using GridGain Enterprise or Ultimate, the COPY command was reworked as described in the [COPY Command Changes] section. The syntax for working with CSV remains the same, but the path is now calculated on the node and the client. This change does not affect GridGain Community edition.

To disable this behavior and use the old copy command, pass the disabledFeatures=SERVER_BULK_LOAD parameter in the JDBC connection command:
```
jdbc:ignite:thin://127.0.0.1/?disabledFeatures=SERVER_BULK_LOAD
```
If you are using one of the optional modules listed in the [Removed Modules] section, they will continue to work, but compatibility and stability on GridGain 8.9 is not guaranteed. Consider using the alternatives:
- Visor CMD and GUI can be replaced by GridGain Control Center.
- The spark module can be replaced by the Spark Ignite Extension.
If you are using default values in your configuration, you may need to set them manually to keep you cluster working the same way as in previous versions. Make sure to check Changed Default Values in GridGain 8.9.1 and Later section for any parameters you need to change.

Older GridGain Versions Compatibility

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

8.9.16, 8.9.16-p1, 8.9.16-p2, 8.9.16-p3, 8.9.17, 8.9.17-p1, 8.9.18, 8.9.19, 8.9.20, 8.9.20-p1, 8.9.21, 8.9.22, 8.9.22-p1, 8.9.22-p2, 8.9.23, 8.9.24, 8.9.25, 8.9.26, 8.9.27, 8.9.28, 8.9.29, 8.9.30, 8.9.31, 8.9.32, 8.9.33, 8.9.34, 8.9.35

Older Thin Clients Compatibility

In most cases, older versions of thin clients should be compatible with newer versions of GridGain. Below is the list of versions of thin clients that are tested for compatibility with the latest version. The compatibility also means that you can safely update from any version on the list to any subsequent version.

Apache Ignite Versions Compatibility

Below is a list of versions that are tested for basic compatibility with the current version. If you are on a version that is not listed, contact GridGain for information on upgrade options.

2.11.1, 2.12.0, 2.13.0, 2.14.0, 2.15.0

Apache Ignite thin clients are not compatible with GridGain. When switching to GridGain, follow the migration instructions to switch to GridGain clients.

Known Limitations

LEFT OUTER JOIN Behavior Change in GridGain 8.9.31

GridGain 8.9.31 fixes an issue in which the WHERE conditions could be incorrectly pushed down into the right branch of LEFT OUTER JOIN operations during query optimization. This optimization could cause queries to return incorrect results by effectively converting LEFT OUTER JOIN operations into INNER JOIN.

Starting with GridGain 8.9.31, the query optimizer correctly prevents WHERE conditions from being pushed down into the right branch of LEFT OUTER JOIN, ensuring that queries return the correct results according to SQL standards.

If your environment expects incorrect results, you set the new DIGNITE_FORCE_PUSHDOWN_CONDITIONS_TO_LEFT_JOIN system property to true on each node in your cluster. Review all affected queries and update them as needed before removing this property.

Java 8 Support in GridGain 8.9.18 and Later

Starting GridGain 8.9.18, GridGain uses Lucene 9 by default. As Lucene 9 requires Java 11, it will be impossible to run GridGain on Java 8 with default configuration.

We recommend switching to Java 11 or later before updating to GridGain 8.9.18 or later.

To continue using Java 8, replace the {GRIDGAIN_HOME}/libs/ignite-lucene-9 folder with {GRIDGAIN_HOME}/libs/optional/ignite-lucene-8.

Unsupported Features When Using Java 8 in GridGain 8.9.17

In GridGain 8.9.17, full text search is not supported when GridGain is running on Java 8. The new vector search feature requires using Java 11.

Additionally, GridGain 8.9.16 or earlier automatically created full text indexes for caches with string values, which may cause issues for working with the cache while using Java 8. To avoid this behavior, set the sql.disableCreateLuceneIndexForStringValueType property to true. If such index is already built for a cache with persistence, contact our support team for assistance with removing it.

We recommend switching to Java 11 or later before updating to GridGain 8.9.17 if you plan to use these features.

Changed Permissions for Control Center Agent in GridGain 8.9.4 and Later

GridGain 8.9.4 introduced stricter permissions checking for Control Center Agent. When running secured clusters, some actions that previously were available without permissions, will now require additional permissions.

Before updating, make sure to provide the required permissions, otherwise some actions may become unavailable.

New TRACING_CONFIGURATION_UPDATE Permission in GridGain 8.9.2 and Later

GridGain 8.9.2 introduced the TRACING_CONFIGURATION_UPDATE permission. If you were using security on a cluster before updating to this version, make sure that you provide the permissions to control utilities or users who need to update cluster permissions. Otherwise, they will not be able to update

Changed Default Values in GridGain 8.9.1 and Later

If you are updating from GridGain 8.8.X or 8.9.0, a large number of default values have been changed.

These changes may affect the stability or performance of your cluster if you are using default values.

We recommend checking the list below to make sure the changes do not have an adverse effect, and setting the value manually if necessary.

Service processor now use event-driven implementation by default. You can keep the old behavior by setting the IGNITE_EVENT_DRIVEN_SERVICE_PROCESSOR_ENABLE property to false. Make sure that all nodes in the cluster are set to the same value. See Services documentation for more information.
Data region metrics are now enabled by default. This may have a minor (within 2%) adverse effect on performance. You can keep the old behavior by setting the DataRegionConfiguration.metricsEnabled value to false.
Data storage metrics are now enabled by default. This may have a minor (within 2%) adverse effect on performance. You can keep the old behavior by setting the DataStorageConfiguration.metricsEnabled value to false.
TcpCommunicationSpi communication protocol now has a limit of 4096 messages for the outgoing messages queue by default.
Atomic operations are now not allowed in transactions by default. You can keep the old behavior by setting the IGNITE_ALLOW_ATOMIC_OPS_IN_TX value to true.
Logs are now in verbose mode by default. You can use the -q command line argument in the ignite.sh script to keep current log behavior.
SQL queries are now loaded lazily by default. This reduces memory consumption on medium and large queries and potentially improves garbage collection performance. You can keep the old behavior by setting the SqlFieldsQuery.setLazy(false).
Cache entries are now read from primary partitions by default, even if an entry is available on the node in a backup partition. This may have a minor performance impact, but significantly increases cluster stability. You can keep the old behavior by setting the readFromBackup property to true.
Partition map exchange transactions now time out after 1 minute instead of 0 (infinite) by default. You can keep the old timeout by setting the TX_TIMEOUT_ON_PARTITION_MAP_EXCHANGE setting to 0.
Data streamer now overwrites entries by default. You can keep the old behavior by setting the stmr.allowOverwrite property to false.
TCP discovery now uses static IP finder TcpDiscoveryVmIpFinder by default. To keep the old behavior, set the IP finder to use multicast IP finder.
Checkpointing process now starts upon reaching 75% of minWalArchiveSize instead of 25%. You can keep the old behavior by setting the IGNITE_CHECKPOINT_TRIGGER_ARCHIVE_SIZE_PERCENTAGE system variable to 0.25.

© 2026 GridGain Systems, Inc. All Rights Reserved. Privacy Policy | Legal Notices. GridGain® is a registered trademark of GridGain Systems, Inc.
Apache, Apache Ignite, the Apache feather and the Apache Ignite logo are either registered trademarks or trademarks of The Apache Software Foundation.

Last updated on Jul 30, 2026