GridGain 9.1.26 Release Notes

GridGain 9.1.26 tightens access control for administrative operations, changes how read-only SQL queries are routed, and adds cluster-wide and per-statement control over follower reads.

Overview

GridGain 9.1.26 tightens access control for administrative operations, changes how read-only SQL queries are routed, and adds cluster-wide and per-statement control over follower reads.

Major Changes

New Permissions Required for Administrative Operations

Several administrative operations that previously required no dedicated privilege are now protected by role-based access control. The following actions were added:

Action	Required for
WRITE_SECURITY_CONFIG	Modifying the cluster security configuration subtree.
MANAGE_DCR	Managing data center replication.
RENAME_CLUSTER	Renaming the cluster.
ENABLE_METRICS, DISABLE_METRICS, LIST_METRICS	Enabling, disabling, and listing cluster metrics.

When you upgrade a cluster to 9.1.26, the built-in system role receives the new actions automatically, so a user with that role keeps full access to the operations listed above.

Custom roles are not updated. A role that holds WRITE_CLUSTER_CONFIG, for example, can no longer modify the cluster security configuration until you also grant it WRITE_SECURITY_CONFIG, and the first attempt to do so after the upgrade fails with an authorization error.

Review your roles before upgrading and grant the new actions to the roles that need them:

GRANT PRIVILEGES WRITE_SECURITY_CONFIG TO cluster_admin;

See GRANT for the statement syntax, and User Permissions and Roles for the full list of actions.

SQL Follower Reads Disabled by Default

Read-only SQL queries no longer read from non-primary replicas by default. In earlier releases, every read-only SQL query could be served by any replica of a partition. Starting with this release, such queries are mapped to primary replicas only, governed by the new cluster-wide sql.allowFollowerReads property, which defaults to false.

This changes how read-only queries are routed. Read load that was previously spread across all replicas now concentrates on primaries, and a query that a single node could previously serve entirely from its local replicas may now require a network hop. In exchange, these queries no longer wait for a follower replica to catch up.

To restore the previous behavior, enable follower reads cluster-wide:

cluster config update ignite.sql.allowFollowerReads=true

You can also allow or disallow follower reads for an individual statement or JDBC connection - see Per-Statement Control Over Follower Reads and Controlling Follower Reads.

New Features

Per-Statement Control Over Follower Reads

You can now control, for an individual SQL statement, whether the SQL engine may read from non-primary replicas. The new allowFollowerReads statement property overrides the cluster-wide setting for that statement only, and applies to read-only statements.

Enabling follower reads spreads read load across all replicas and lets a query read from a local replica, avoiding a network hop. The trade-off is latency, because reading from a follower replica may introduce delays while the replica catches up. Disabling it forces the statement to always read from primary replicas. Leaving the property unset keeps the cluster-wide setting, which is the default behavior.

The cluster-wide setting itself is new in this release and defaults to disallowing follower reads - see SQL Follower Reads Disabled by Default.

Statement stmt = client.sql().statementBuilder()
    .query("SELECT * FROM ACCOUNTS WHERE STATUS = 'ACTIVE'")
    .allowFollowerReads(false)
    .build();

For the JDBC equivalent, see Controlling Follower Reads. For how follower reads relate to read-only transactions, see Consistency Model.

Improvements and Fixed Issues

Issue ID	Category	Description
GG-50536	Cluster SQL Engine	Fixed a query with a range predicate under OR returning incomplete results.
GG-50257	Cluster SQL Engine	Improved performance of certain UPDATE queries.
GG-50225	Cluster Storage Engine	Fixed partition rebalance failing to transfer partitions that contain deleted rows (tombstones).
GG-50190	General	Fixed incorrect auto-commit when closing a read-write scan cursor of an explicit transaction.
GG-50080	Cluster Continuous Queries	Java client: fixed a further case of continuous query event loss when the underlying table is dropped.
GG-50076	Cluster Continuous Queries	.NET: Fixed a continuous query losing events when the underlying table is dropped.
GG-50068	Cluster SQL Engine	Fixed a memory leak when a query timeout is set.
GG-50033	Platforms & Clients	C++: Fixed a possible crash during client shutdown on Windows.
GG-49967	Cluster SQL Engine	Reduced allocations when returning single-row result sets.
GG-49952	Cluster SQL Engine	Removed primitive boxing when sorting rows in SQL.
GG-49944	Cluster Storage Engine	Reduced heap allocations in the MVCC garbage collector.
GG-49865	Cluster SQL Engine	Fixed excessive CPU usage caused by the query execution root node busy-looping while waiting for its source.
GG-49835	Cluster SQL Engine	Improved performance of SQL queries that use index scans inside read-write transactions.
GG-49689	Cluster Continuous Queries	Fixed a continuous query losing events when the underlying table is dropped.
GG-49688	Cluster SQL Engine	The optimizer now prefers a primary key index lookup over a secondary index scan when possible.
GG-49619	Cluster Security	Renaming the cluster now requires the new RENAME_CLUSTER action.
GG-49617	Cluster Security	Deployment units now carry an integrity manifest that is validated on deployment, so a unit whose contents were altered after packaging is rejected.
GG-49614	Cluster Security	Enabling, disabling, or listing cluster-wide metrics now requires the corresponding metrics action (ENABLE_METRICS, DISABLE_METRICS, or LIST_METRICS).
GG-49613	Cluster Security	Writing the cluster security configuration (ignite.security) through the REST configuration endpoint now requires the new WRITE_SECURITY_CONFIG action in addition to WRITE_CLUSTER_CONFIG.
GG-49609	Cluster Security	Hardened code deployment against path-traversal entries in deployment unit archives.
GG-49582	Cluster Security	Managing data center replication now requires the new MANAGE_DCR action.
GG-49569	General	Colocated nodes that share a network interface are now counted as a single host for license CPU and host limits. Set the GG_LICENSE_HOST_MAC_INTERSECTION system property to false to restore the previous counting.
GG-49413	Cluster SQL Engine	Reduced memory pressure during SQL query execution.
GG-49379	Cluster SQL Engine	SQL reads from non-primary replicas are now disabled by default. Added the allowFollowerReads property, cluster-wide and per-statement, to re-enable them.
GG-49302	Cluster Storage Engine	Added an API to advance the low watermark past orphaned locks.
GG-49086	Cluster Data Replication	Added an experimental Microsoft SQL Server CDC source that reads native SQL Server change data capture, gated behind the --experimental flag.
GG-49085	Cluster Data Replication	Added a GridGain 9 CDC sink that writes change events into a local GridGain 9 table. Use the targetTable parameter to remap the destination; replicating a table to itself is rejected.
GG-48647	Platforms & Clients	.NET: Added DataStreamerOptions.SameKeyUpdateMode with Squash and Preserve modes.
GG-48432	CLI Tool	Added the --tx-timeout option to the CLI sql command, which overrides the cluster-wide read-write transaction timeout for the current session.
GG-48408	Cluster Rolling Upgrade	Unless a rolling upgrade is in progress, all cluster nodes must now run the same version. During a rolling upgrade, only the base and target versions are accepted.
GG-48391	Cluster Data Snapshots and Recovery	Fixed an out-of-memory error during snapshot restore.
GG-47963	Platforms & Clients	C++: Added a configurable retry policy for client operations, bringing the C++ client in line with the Java and .NET clients.
GG-47911	Platforms & Clients	C++: Optimized table schema retrieval.
GG-47814	Migration Tools	Migration Tools adapter can now read java.time.Instant values serialized in the extra column.
GG-46020	Distributed Computing	.NET: Added JobExecutorType.WasmEmbedded and WasmJobDescriptor to execute Wasm jobs.

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.

9.1.20, 9.1.21, 9.1.22, 9.1.23, 9.1.24, 9.1.25

When updating from older versions, we recommend updating to version 9.1.20 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: http://support.gridgain.com/.