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

Configuration and Validation Changes

Configuration changes when migrating from Apache Ignite 2.18 to GridGain 8.10: changed defaults, removed and unsupported properties, and binary field ordering.

This page covers the configuration defaults that change, the properties GridGain 8 removes, its stricter validation, and the binary object field-ordering setting.

Server Configuration Differences

Changed Defaults

This section applies to every migration: review the table and decide each value deliberately, whether you restore the Ignite 2 behavior or keep the GridGain 8 default.

GridGain 8.10 changes a number of configuration defaults relative to Apache Ignite 2.18.

Setting
Ignite 2 default
GridGain 8 default

CacheConfiguration.readFromBackup

true

false

IgniteConfiguration.segmentationPolicy

USE_FAILURE_HANDLER (enum member absent in GridGain 8)

STOP

SqlConfiguration.sqlGlobalMemoryQuota

unlimited (field absent)

"60%" of heap

DataRegionConfiguration.metricsEnabled DataStorageConfiguration.metricsEnabled

false

true

TcpDiscoverySpi IP finder

TcpDiscoveryMulticastIpFinder

TcpDiscoveryVmIpFinder(shared=true)

TcpCommunicationSpi.messageQueueLimit

0 (unlimited)

4096

ClientConnectorConfiguration.socketSendBufferSize / socketReceiveBufferSize

0 (OS default)

262144 (256 KB)

IgniteDataStreamer.allowOverwrite()

false

true (via IGNITE_DATA_STREAMER_ALLOW_OVERWRITE, defaulting to true)

Thin client partition/affinity awareness

partitionAwarenessEnabled = true

affinityAwarenessEnabled = false (the Ignite-named methods remain as deprecated aliases)

Thin client timeouts (ClientConfiguration)

timeout = 0 (infinite)

handshakeTimeout = 10000 ms, requestTimeout = 60000 ms (setTimeout is deprecated)

TransactionConfiguration.txTimeoutOnPartitionMapExchange

0 (no timeout)

60000 (ms)

SqlConfiguration.longQueryWarningTimeout

3000 (ms)

1000 (ms)

IgniteConfiguration.rebalanceBatchesPrefetchCount

3

2

IGNITE_CHECKPOINT_TRIGGER_ARCHIVE_SIZE_PERCENTAGE (system property)

0.25

0.75

Baseline auto-adjust with persistence

single timeout

separate scale-up (5 minutes) and scale-down (never) timeouts

System worker blocked timeout

falls back to failureDetectionTimeout (10 s)

120 s

TcpDiscoverySpi socket SO_LINGER

0

5

IGNITE_QUIET / ignite.sh log verbosity

quiet

verbose (-q makes it quiet)

Sensitive data in logs (IGNITE_SENSITIVE_DATA_LOGGING)

plain values (IGNITE_TO_STRING_INCLUDE_SENSITIVE=true)

hash

SftpConfiguration.strictHostKeyChecking (Ultimate Edition, SFTP snapshot storage; new in 8.10)

not applicable

true: requires a known-hosts file or an explicit false

To preserve Ignite 2 behavior, set each changed default explicitly. This template restores the server-side defaults that differ between Ignite 2 and GridGain 8; the settings that live in cache or client code are listed separately below.

Some Ignite 2 defaults are restored through system properties (JVM -D arguments or environment variables) rather than configuration:

The remaining defaults are not server-side configuration and must be set where they apply:

  • readFromBackup (per cache). Set readFromBackup=true on each CacheConfiguration or cache template. The GridGain 8 default is false.

  • Thin-client affinity awareness (client code). Call ClientConfiguration.setAffinityAwarenessEnabled(true). The GridGain 8 default is false.

  • Thin-client timeouts (client code). Ignite 2 waits indefinitely by default. GridGain 8.10 gives up after 10 s on handshake and 60 s per request, so long-running thin-client operations (large scans, bulk loads) can start failing with timeouts. Set setHandshakeTimeout and setRequestTimeout explicitly (0 is infinite).

  • Discovery (environment-specific). Ignite 2 defaults to multicast discovery. If your cluster relied on it, configure a TcpDiscoveryMulticastIpFinder or supply the static address list; GridGain 8 defaults to TcpDiscoveryVmIpFinder.

Several of these defaults change behavior in ways that pass functional tests and surface only later: under load, on large data, or during a network event. The ones that most often cause problems:

Affinity awareness (clients): Ignite 2 enables it (as partition awareness) and GridGain 8 disables it, so thin clients no longer route each request to the node that owns the key and every operation takes an extra network hop. This passes functional tests but shows up as a latency and throughput regression under load. Re-enable it unless you have a reason not to. Service awareness in the Java thin client (Ignite 2.16.0 and later, also in GridGain 8.10) works only when partition awareness is enabled. With the GridGain 8 default (disabled), the client sends service calls the standard way: to the node it is connected to, which forwards them to a node that runs the service.

Discovery IP finder: Ignite 2 defaults to multicast discovery; GridGain 8 defaults to the static TcpDiscoveryVmIpFinder, pre-seeded with 127.0.0.1:47500..47600 only. Nodes on the same host still find each other, but nodes on different hosts no longer do. A multi-host cluster that relied on multicast will not form until you supply the real node addresses or configure the multicast IP finder.

readFromBackup: Ignite 2 can serve a read from a local backup copy; GridGain 8 (default false) routes every read to the primary. Results are unaffected, but read load concentrates on primary nodes and latency can rise where you previously benefited from backup locality. Set it per cache to restore the Ignite 2 behavior.

SQL memory quota (sqlGlobalMemoryQuota): Ignite 2 runs SQL with no global memory quota; GridGain 8 caps it at 60% of heap. A query that ran in Ignite 2 can fail with a quota error the first time it processes a large result set, not during ordinary testing. Set it to 0 for unlimited, or tune it. See SQL Changes.

Data streamer overwrite (allowOverwrite): Ignite 2's IgniteDataStreamer does not overwrite existing keys by default; GridGain 8 enables overwrite. A streaming job that assumed existing entries were left in place will now replace them. Set allowOverwrite(false) in client code if you relied on the Ignite 2 behavior.

Transaction timeout on partition map exchange (txTimeoutOnPartitionMapExchange): Ignite 2 does not time out transactions during a partition map exchange (PME) by default (0); GridGain 8 aborts them after 60 seconds. During a long PME, such as a node join or rebalance on a large cluster, transactions that Ignite 2 would have let run can fail in GridGain. Set it to 0 to restore the Ignite 2 behavior, or keep the timeout as a safeguard against PME stalls.

Checkpoint trigger by WAL archive size (IGNITE_CHECKPOINT_TRIGGER_ARCHIVE_SIZE_PERCENTAGE): Ignite 2 triggers a checkpoint when the WAL archive reaches 25% of its maximum size; GridGain 8 waits until 75%. GridGain 8 checkpoints less often, so WAL archives grow larger and crash recovery can take longer. Set the system property to 0.25 to match Ignite.

Segmentation policy: Ignite 2's default USE_FAILURE_HANDLER delegates the response to your configured failure handler; that value does not exist in GridGain 8, whose default is STOP: a segmented node stops itself. If you relied on a custom failure-handler response to network segmentation, choose a GridGain 8 segmentation policy.

Logging and system worker timeouts: GridGain 8 logs more by default (IGNITE_QUIET is false), hashes sensitive values in logs, and waits 120 s instead of 10 s before declaring a system worker blocked. Log volume and log-parsing tools are affected. The worker timeout alone does not make failure detection faster or slower.

Restoring Ignite 2 Discovery

The discovery IP finder is the one changed default with no server-side template entry, because the value you need is environment-specific. Pick one of the following and set it explicitly on TcpDiscoverySpi.

Recommended: static address list. This option is required in the cloud, where multicast is typically unavailable. List your actual server addresses. The port range per host covers multiple nodes and port autoincrement:

To keep the Ignite 2 behavior: multicast finder. Use it only where the network supports multicast (typically bare-metal or on-premises LANs, not cloud). The defaults match Ignite 2: multicast group 228.1.2.4, port 47400.

Removed Configuration Properties

Skip this section if your Ignite 2 deployment used none of the removed features listed below (CDC, snapshot paths, Calcite, multi-disk storage, disk page compression, SQL validation) and set no IGNITE_* system properties (see unsupported system properties).

GridGain 8 removes two kinds of properties, and they fail in opposite ways.

Configuration properties (XML or bean) fail loudly. GridGain 8 does not start until you remove or replace a property that belongs to a removed feature:

  • CDC: DataStorageConfiguration.cdcWalPath and cdcWalDirectoryMaxSize, DataRegionConfiguration.cdcEnabled, and the CdcConfiguration bean

  • SQL: SqlConfiguration.queryEnginesConfiguration and validationEnabled, TransactionConfiguration.txAwareQueriesEnabled, and the Calcite, IndexingQueryEngineConfiguration, and QueryEngineConfiguration beans

  • Storage: walForceArchiveTimeout, writeRecoveryDataOnCheckpoint (and its checkpointRecoveryDataCompression / checkpointRecoveryDataCompressionLevel companions), extraStoragePaths, extraSnapshotPaths, and per-cache storagePaths, indexPath, diskPageCompression, diskPageCompressionLevel

  • Snapshots: IgniteConfiguration.snapshotPath and snapshotThreadPoolSize, and the DumpReaderConfiguration bean

  • Other: IgniteConfiguration.distributedPropertiesDefaultValues, TcpDiscoverySpi.addressFilter, ClientConfiguration.partitionAwarenessMapperFactory, ServiceConfiguration.statisticsEnabled

sqlPlanHistorySize exists in GridGain 8.10 and does not need to be removed.

An unknown XML or bean property fails startup with a message naming the offending property, for example Configuration property CacheConfiguration.<name> does not exist. Use it to find and remove the leftover Ignite-only property in your configuration.

System properties (IGNITE_*, set as JVM -D arguments or environment variables) behave the opposite way. An unrecognized one does not block startup, so a setting you relied on can stop taking effect with no failure. Audit your -D flags and environment against the unsupported system properties below.

Unsupported System Properties

Many Ignite 2 IGNITE_* system properties have no effect in GridGain. If you set one, whether as a JVM -D argument or an environment variable, GridGain 8 ignores it and logs a warning at node startup:

Remove the following properties from your startup configuration when migrating:

Property
Reason it is unsupported
GridGain 8 alternative

IGNITE_CALCITE_EXEC_IN_BUFFER_SIZE IGNITE_CALCITE_EXEC_IO_BATCH_CNT IGNITE_CALCITE_EXEC_IO_BATCH_SIZE IGNITE_CALCITE_EXEC_MODIFY_BATCH_SIZE IGNITE_CALCITE_REL_JSON_PRETTY_PRINT

The GridGain SQL engine is H2-based; Calcite is not available.

IGNITE_SNAPSHOT_SEQUENTIAL_WRITE

The GridGain 8 snapshot subsystem differs from Ignite 2.

IGNITE_PERF_STAT_BUFFER_SIZE IGNITE_PERF_STAT_CACHED_STRINGS_THRESHOLD IGNITE_PERF_STAT_FILE_MAX_SIZE IGNITE_PERF_STAT_FLUSH_SIZE

The Performance Statistics subsystem is not implemented in GridGain.

IGNITE_ALLOW_MIXED_CACHE_GROUPS

Mixed cache groups are not available in GridGain.

IGNITE_ENABLE_OBJECT_INPUT_FILTER_AUTOCONFIGURATION

Object input filter auto-configuration is not available in GridGain.

IGNITE_USE_BINARY_ARRAYS

Typed binary arrays are not available in GridGain.

Confirm it was not enabled on the Ignite 2 source before migrating data. An explicitly enabled binary-array format changes the on-disk and on-wire layout.

IGNITE_SQL_FORCE_LAZY_RESULT_SET

Not available in the GridGain H2-based SQL engine.

Enable lazy execution per query with SqlFieldsQuery.setLazy(true) or the JDBC thin lazy connection parameter. See Lazy SQL Queries.

Two other Ignite 2 properties are not on the warning list but are absent from GridGain 8, so they are ignored with no log line: IGNITE_DATA_CENTER_ID (use the data center ID of GridGain Data Center Replication) and IGNITE_LOCAL_STORE_KEEPS_PRIMARY_ONLY (the local-store feature was removed).

Stricter Validation

GridGain 8 rejects things Ignite 2 allowed: duplicate SQL schema or index names, invalid cache names, and nodes that disagree on the binary field-sort setting (see Binary Object Field Ordering).

Binary Object Field Ordering

The IGNITE_BINARY_SORT_OBJECT_FIELDS system property controls the order in which a binary object's fields are laid out on disk and on the wire. The value must be the same on every node and every client, and it must match the value that was set when the persisted data was written.

The default is false on both Ignite 2 and GridGain 8.x. If your Ignite 2 cluster never set this property, the migration needs no action; skip the rest of this section.

1. Check the Source Value

Search for IGNITE_BINARY_SORT_OBJECT_FIELDS in the JVM arguments and environment of the running Ignite 2 nodes. If it is not set, the source uses the default. Otherwise, record the value.

2. Apply the Value in GridGain

Set -DIGNITE_BINARY_SORT_OBJECT_FIELDS=<source-value> as a JVM system property on every GridGain 8 server node and every client. Roll it out everywhere at once. Any node or client with a different value is rejected at startup.

3. Verify After Start

After the first GridGain 8 node starts, run two checks:

  • Look at the logs. A topology-validation or handshake error mentioning IGNITE_BINARY_SORT_OBJECT_FIELDS means a server or client is running with a different value.

  • Smoke-test a known key. Run a key-based get against a row you know exists. If it returns nothing while SELECT over the same row succeeds, GridGain 8 is using a different value than what is on disk.

For background, see Working with Binary Objects.

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

Last updated

Was this helpful?