GridGain 9.1.24 Release Notes

GridGain no longer depends on artifacts published on JitPack. The dependency that previously required the jitpack.io repository (see the 9.1.21 release notes) is now served from the GridGain Maven repository.

It is recommended to remove jitpack.io dependency from your pom.xml to keep your configuration clean. If not updated, GridGain will automatically switch to the new source, and leave an unnecessary dependency in configuration.

The cluster init CLI command now accepts two optional flags that make it safe to re-run from scripts and automation. Use --if-needed to skip initialization and exit with code 0 when the cluster is already initialized instead of failing. Add --if-nodes=<N> (which requires --if-needed) to also defer initialization until at least N nodes are visible in the physical topology.

The example below shows how you can initialise a cluster only once and only if 3 or more nodes are visible in the topology:

See CLI tool documentation for more information.

When using the data streamer, you can now control how same-key updates within a single batch are surfaced to continuous queries. The new DataStreamerOptions.sameKeyUpdateMode option takes a DataStreamerSameKeyUpdateMode value: SQUASH (default) conflates multiple updates to the same key within a batch into a single continuous query event for better performance, while PRESERVE splits the batch so that each update produces a separate continuous query event.

See same-key update mode and continuous query events for more information.

SQL GRANT/REVOKE PRIVILEGES …​ ON <object> now accepts a schema-qualified object name. A schema-qualified name such as PUBLIC.MY_TABLE grants the privilege on that single object, while a bare name such as PUBLIC still targets the whole schema. This lets you scope object-level privileges - for example, CREATE_INDEX on a specific table - directly from SQL.

See GRANT for details.

You can use the new ignite.compute.dotnet node configuration section to tune the .NET runtime of the compute executor (sidecar) process. You can toggle Server GC and limit the executor’s memory with an absolute or percentage-based limit.

The example below shows new configuration options:

See .NET Compute Executor for more information.

The .NET client now provides the IIgnite.Cluster API, which exposes the cluster topology. Use Cluster.GetNodesAsync() to retrieve all nodes in the logical topology, and Cluster.LocalNode to get the hosting node from inside a compute job or data streamer receiver (it returns null on the client side). The previous IIgnite.GetClusterNodesAsync() method is now deprecated in favor of Cluster.GetNodesAsync().

See Cluster API for more information.

You can now manually rebalance primary replica leaseholders across cluster nodes. The new distribution balance-leases CLI command forces the placement driver to redistribute leaseholders evenly - one leaseholder per replication group.

See CLI tool documentation for more information.

When performing a rolling upgrade from GridGain 9.1.9 or earlier, it is necessary to first upgrade to 9.1.10 before upgrading to 9.1.11 or a later version. This is caused by improvements in rolling upgrade procedure that make skipping 9.1.10 with a rolling upgrade impossible.

The upgrade to 9.1.10 must be performed for the whole cluster and committed before the next upgrade can be started.

You do not need to perform this intermediary upgrade when upgrading with downtime, as it only affects the rolling upgrade procedure.

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

There are known issues with the performance of SQL read-write transactions in complex read-write scenarios. These issues will be addressed in upcoming releases.