Skip to content
For the complete documentation index, see llms.txt. This page is also available as Markdown.

What to Know Before Upgrading

Planning reference for upgrading from GridGain 8.9 to 8.10: choosing a rolling or offline upgrade, upgrade constraints, and the scope and effort of each change.

Read this page before you upgrade a cluster from GridGain 8.9.39 to 8.10. It helps you choose an upgrade path, find the changes that affect you, and estimate the work. When you are ready, follow the Upgrade Procedure.

This guide uses 8.9.39 for the primary starting version, 8.9.x for behavior across the 8.9 line, and 8.9 for a cluster or node that you work with.

Supported Starting Versions

You can upgrade to 8.10 from the following 8.9.x releases.

Show supported releases

8.9.0, 8.9.1, 8.9.5, 8.9.8, 8.9.9, 8.9.12, 8.9.13, 8.9.14, 8.9.14-p2, 8.9.15, 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, 8.9.20-p2, 8.9.21, 8.9.22, 8.9.22-p1, 8.9.22-p2, 8.9.23, 8.9.23-p1, 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, 8.9.33-p1, 8.9.34, 8.9.35, 8.9.36, 8.9.37, 8.9.38, 8.9.39

If your cluster runs a release that is not listed, contact MariaDB support before you upgrade. Behavior can differ in places between releases. See the release notes for the releases between your version and 8.9.39 first.

Choose Your Upgrade Path

  • Rolling upgrade. You upgrade the nodes one at a time while the cluster stays available. It uses Rolling Upgrades, an Enterprise and Ultimate feature, and requires rollingUpdatesEnabled=true on every 8.9 node.

  • Offline upgrade. You stop the cluster and start every node from the 8.10 distribution. It does not need rollingUpdatesEnabled.

The right path depends on your 8.9 cluster:

Your 8.9 cluster
Path

Every node has rollingUpdatesEnabled=true

Rolling upgrade, with no downtime for the upgrade itself. On Enterprise Edition, the backup needs a short stop of the whole cluster. See Step 3.

rollingUpdatesEnabled is not set, and you can schedule one full-cluster restart

Offline upgrade in that window. It is simpler, because you stop the cluster anyway.

rollingUpdatesEnabled is not set, and you want a rolling upgrade

Restart the whole cluster once with the property set, then upgrade. See Step 2.

Upgrade Constraints

  • Java 17 or later is required. Every server node, client application, and application that embeds GridGain must run on it. See Java 17 or Later.

  • Downgrade is not supported. An 8.9 node cannot join the cluster after the rolling upgrade is finished. Back up the 8.9 data first. On Ultimate Edition, take a full snapshot. On Enterprise Edition, which has no snapshots, copy the db folder of every node, and also the folders that storagePath, walPath, and walArchivePath point to if you set them. After the upgrade, restoring that backup is the only way to recover 8.9, and everything written after the backup is lost. See Recovery Path.

  • Caches without backups are unavailable during a rolling upgrade. Partitions on a stopped node cannot be served. See Application Behavior During a Rolling Upgrade.

  • Some integrations have no 8.10 replacement. Mesos deployments stay on 8.9.x. See Removed Modules and Integrations.

Which Pages Apply to You

Upgrade Scope and Effort

The tables list each change, when it applies to you, and the effort it needs. If an item does not apply to your deployment, skip it.

Effort is one of the following:

  • Mechanical. A rename, path change, or similar update with no logic changes.

  • Config-only. Set, remove, or restore a configuration property. No code changes.

  • Code. A contained code change, such as a renamed API call.

  • Redesign. Adopt a different system or API. Expect design work and retesting.

  • Contact Support. No documented replacement exists. Contact MariaDB support before you upgrade.

A range such as Mechanical to Code means that the effort depends on what you use.

For Application Developers

What changes
Applies if
Effort
Details

Java 17 or later

Always: embedding applications, the thin client, and the JDBC driver

Mechanical

Spring 7 and Spring Data module rename

Your application embeds GridGain through Spring

Mechanical to Code

Hibernate 7.4

You use the Hibernate second-level cache

Code

Gridify API replaced by the compute API

You use @Gridify

Code

AWS module replaced by ignite-aws-ext

You use the S3 checkpoint SPI or the S3, ELB, or ALB IP finders

Code

Machine learning module replaced by an extension

You use ignite-ml

Code

Local store replaced by native persistence or CacheStore

You use @CacheLocalStore or CacheFileLocalStore

Redesign

Cassandra store on DataStax driver 4

You use the Cassandra store

Code

Modules with no replacement

You use Mesos. Stay on 8.9.x or move to another system

Redesign

Java thin client timeouts

A thin client operation runs longer than 60 seconds

Config-only

Thin clients in a mixed-version cluster

An 8.10 client uses vector indexes or withPartitionRecover() against an 8.9 server, including during a rolling upgrade

Config-only

.NET and C++ client changes

You use the .NET or C++ client

Mechanical to Code

REST scan queries with a filter class

You send EXECUTE_SCAN_QUERY with className

Config-only

Thin client permission and filter checks

Thin clients call Java services or register continuous query remote filters

Config-only

JMX object metrics are strings

A JMX tool reads ObjectMetric attributes

Code

For Cluster Administrators

What changes
Applies if
Effort
Details

Java 17 or later on every server host

Always

Mechanical

New libs folder and packaging changes

Always

Mechanical

Rolling Upgrades and snapshots on 8.9

You use the rolling upgrade path

Config-only

Per-feature license checks

Always. Compare your license with your configuration

Config-only

Jetty 12

You use a custom jetty.xml for the REST connector

Config-only

Lucene lines

You run ignite-lucene-8, or want to use Lucene 10

Mechanical

ignite-ssh module removed

You start remote nodes with startNodes()

Contact Support

SFTP host key verification

You send snapshots to an SFTP destination

Config-only

Server-side SQL COPY

You use COPY against files or object stores

Config-only

Vector search limits

You use vectors with more than 4096 dimensions, or request more than 10000 neighbors

Config-only

Stale system properties

Your startup scripts set an unsupported Apache Ignite property

Config-only

Control Center agent and OpenCensus

Your 8.9 node loads control-center-agent or ignite-opencensus

Mechanical

This page is: Copyright © 2026 MariaDB. All rights reserved.

Last updated

Was this helpful?