Minor Releases

Axon’s test fixtures perform a "deep equals" operation, using reflection as they go. JDK17, rightfully so, does not allow that for all classes. To solve scenarios where users utilize objects from, for example, java.lang, we have introduced a distinct DeepEqualsMatcher in pull request #2210. This matcher implementation considers the situation that an InaccessibleObjectException might be thrown from Axon’s test fixtures, correctly dealing with the scenario by assuming the assertion failed.

Contributor fabio-couto noticed a predicament within the PooledStreamingEventProcessor (PSEP for short) when they were facing connectivity issues with their RDBMS. In the face of these issues, the PSEP coordinator is incapable of fetching events, resulting in canceled work packages. As part of canceling, the PSEP actively tries to release token claims, which is yet another database operation. This loop of several connectivity issues causes the PSEP to enter a state it could not recover from. Pull request #2225, provided by fabio-couto, solves this predicament.

A fix was introduced to the EventTypeUpcaster to solve issues further down the upcasting chain. Contributor dakr0013 noted that upcaster invoked after an EventTypeUpcaster failed because the expected intermediate event type was adjusted to Object. dakr0013 provided a pull request, which we made some adjustments in PR #2177 to accommodate additional scenarios.

Although Axon Framework doesn’t use the log4j-core dependency directly, we updated it to the most recent version for ease of mind. You can follow these increments in issues #2038, #2040 and #2052.

Contributor jasperfect spotted a predicament with duplicate aggregate creation combined with using caches. Axon didn’t invalidate the cache as it should have, causing unexpected behavior. You can find the issue description here. Additionally, you can find the pull request solving the problem here.

Contributor shubhojitr stated in issue #2051 that the axonserver-connector-java project pulled in a non-secure version of grpc-netty. As this isn’t an issue on Axon Framework itself, we solved the problem under the connector project. As a follow-up, we incremented the framework’s version for the axonserver-connector-java project to 4.5.4, which contains the most recent version of the grpc-bom.

The auto-configuration we introduced for XStream used a suboptimal approach. We assumed searching for the @ComponentScan would suffice but didn’t consider that Spring enabled SpEL operations in the annotation’s properties. This approach thus caused some applications to break on start-up. As such, this approach is replaced entirely by using the outcome of the AutoConfigurationPackages#get(BeanFactory) method. For those interested in the details of the solution, check out this pull request. Kudos to contributor maverick1601 for drafting issue #1963 explaining the predicament.

We introduced an optimization towards updating the TrackingToken. In (distributed) environments where the configuration states several segments per Streaming Processor, there are always threads receiving events that they’re not in charge of due to the configured SequencingPolicy. The old implementation eagerly updated the token in such scenarios, but this didn’t benefit the end-user immediately. Pull request #1999 introduce a wait period for 'event-less-batches', for both the TrackingEventProcessor and PooledStreamingEventProcessor. This adjustment minimizes the number of token updates performed by both processor implementations.

The introduction of Spring Boot version 2.6.0 brought an issue to light within Axon’s Spring usage. The AbstractAnnotationHandlerBeanPostProcessor took FactoryBean instances into account when searching for message handling methods. This approach, however, is not recommended by Spring, which they enforced in their latest release. The result was circular dependency exceptions on start-up whenever somebody used Spring Boot 2.6.0. The fix was simple, though, as we should simply ignore FactoryBean instances. After spotting the issue, we resolved it in this pull request.

Firstly, we updated the XStream version to 1.4.18. This upgrade was a requirement since several CVE’s were noted for XStream version 1.4.17. As a consequence of XStream’s solution imposed through the CVE’s, everybody is required to specify the security context of an XStream instance. This change also has an impact on Axon Framework since the XStreamSerializer is the default serializer. So as of this release, any usages of the default XStreamSerializer will come with warnings, stating it is highly recommended to use an XStream instance for which the security context is set through types or wildcards. When your application uses Spring Boot, Axon will default to selecting the secured types based on your @ComponentScan annotated beans (for example, like the @SpringBootApplication annotation). For those interested in the details of the solution, check out this pull request.

User 'nils-christian' noted in issue #1892 that Axon executed Upcaster beans in a Spring environment in the incorrect order. This ordering issue was due to a misconception in deducing the @Order annotation on upcaster beans. We resolved the problem in pull request #1895.

We noticed a TokenStore operation that Axon did not invoke within a transaction. In most scenarios, this worked out, but when using Micronaut, for example, this (correctly) caused an exception. After spotting the issue, we resolved it in this pull request.

One new feature has been introduced in 4.5.3: the PropertySequencingPolicy by contributor nils-christian. This sequencing policy can be configured to look for a common property in the events.

The version of the axonserver-connector-java has been updated to 4.5.2. This update resolves a troublesome issue around permit updates for subscription queries, which exhausted the number of queries an application could have running. For those curious about the solution, pull request 85 addresses this issue.

The WorkerLauncher runnable, used by the TrackingEventProcessor to start its threads, was not considered when you shut down a tracking processor. As a consequence, it could start new segment operations while shutdown already completed "successfully." Pull request 1866 resolves this problem, ensuring a tracking processor shuts down as intended.

Issue 1853 describes an issue where the creation policy always. Exceptions thrown from within a command handler annotated with @CreationPolicy(ALWAYS) weren’t correctly propagated. Pull request 1854 solves this issue.

Added a missing isReplaying flag on the StreamingEventProcessor. Pull request #1821 reintroduces this functionality in this release.

Some enhancements in regards to logging Exceptions and stacktraces when initialization fails. This commit reintroduces this functionality in this release.

Improved Axon Framework (AxonServerEventStore) which will now rethrown Exceptions that has a valid Status.Code. Pull request #1842 reintroduces this functionality in this release.

General improvements on the PooledStreamingEventProcessor made across several Pull Requests.

Some internals have changed concerning command handling exceptions. Within a single JVM, Axon Framework knows whether the exception is transient or not. This piece of information allows the RetryScheduler to retry a non-transient exception since those are retryable. With the move towards distributed environments, the information whether an exception is transient was lost when we moved to the dedicated CommandHandlingException containing a details object. Pull request #1742 reintroduces this functionality in this release.

The new RevisionSnapshotFilter introduced in release 4.5 sneaked in a bug by not validating the aggregate type upon filtering. Pull request #1771 describes and solves the problem by introducing the aggregate type to the RevisionSnapshotFilter.

By enabling the CreationPolicy for the DisruptorCommandBus, a timing issue was introduced with handling events. Contributor "junkdog" marked the problem in issue #1778, after which pull request #1792 solved it.

Contributor "michaelbub" noted in issue #1786 that resetting a StreamingEventProcessor to a point in the future reacted differently when no token was stored yet. This followed from the implementation of the ReplayToken, which wrongfully assumed that if the given 'token at reset' was null, the start position should be null too. However, the start position might be the future, and hence it should be used in favor of null. This issue is addressed under this pull request.

A bug was noted whenever a query handler returned a Future/CompletableFuture in combination with a subscription query, with Axon Server as the infrastructure. In this format, Axon would incorrectly use the scatter-gather query for the initial result of the subscription query. Whenever the returned result was completed, this didn’t cause any issues. However, for a Future/CompletableFuture a TimeoutException would be thrown. The issue was luckily easily mitigated by changing the "number of expected results" from within the QueryRequest to default to 1 instead of zero. As an effect, the point-to-point would be invoked instead of scatter-gather. For reference, the issue can be found here.

Whenever an interface is used as the type of an @AggregateMember annotated field, Axon would throw a NullPointerException. This is far from friendly, and has been changed towards an AxonConfigurationException in pull request #1742.

The Axon Server Connector Java version 4.4.7 has been included in this release as well. As such, it’s fixes (found here) are thus also part of this release.

Contributor "krosenvold" noticed that the SQL to retrieve a stream of events was performed twice in quick concession. The provided solution (in pull request #1689) would resolve this, but the problem was spotted to originate elsewhere. Commit 16b7152 saw an end to this occurrence by making a minor tweak in the EmbeddedEventStore.

As rightfully noticed by user "pepperbob", there was a type discrepancy when reading events through a tracking token. An event would always become a DomainEventMessage when read through the EventStorageEngine, whereas it might originally have been a regular EventMessage. The problem has been fixed in commit c61a95b. Furthermore, the entire description of the issue can be found here.

Through the use of the AxonServerQueryBus, a cancelled subscription query was wrongfully completed normally where it should complete exceptionally. This problem is marked and resolved under pull request #1695.

Contributor "Rafaesp" noted that a registered CommandHandlerInterceptor in the AggregateTestFixture could be invoked more often than desired. This only occurred if the fixture’s givenCommands(…​) method was invoked, but nonetheless this behaviour was incorrect. The issue is marked under #1665 and resolved in pull request #1666.

In 4.4.4, a fix was introduced which ensured a ChildEntity (read, the Aggregate Members) was no longer duplicated in an aggregate hierarchy. This fix had the troublesome side effect that aggregate member command handlers weren’t registered on every level of the aggregate hierarchy anymore. The resolution to this problem can be found in pull request #1674.

Using the subscription query in a distributed environment had a possible troublesome side effect. If a consumer of updates was closed for whatever reason, it could also close the producing side. This is obviously undesired, as no single consumer should influence if the producer should still dispatch updates to other consumers. The problem was marked under issue #1680 and resolved in this commit.

Right before we aimed to release 4.4.6, contributor "haraldk" provided a thorough issue description when using the SequenceEventStorageEngine. He noted that if snapshots were used for an aggregate, there was a window of opportunity that the 'active' EventStorageEngine in the sequencing engine did not return any events. This followed from the sequence number logic, which wrongfully defaulted to position "0", even though the starting sequence number is per definition higher if a snapshot has been found. The clarifying issue can be found here, with its resolution present in pull request #1683.

When creating a TrackingToken at a certain position through StreamableMessageSource#createTokenAt(Instant), a tail token was wrongfully returned if the provided timestamp exceeded the timestamp of the last event. Instead, the token closests to the provided timestamp should be returned, was equals the head token. This discrepancy between documentation and implementation was marked by mbreevoort and resolved in pull request #1607.

A certain path within the AxonServerEventStore allowed for event retrieval without correctly deserializing the MetaData of the events. If someone tried to access the MetaData, a CannotConvertBetweenTypesException was being thrown. This problem, among others, was remedied in pull request #1612, by ensuring the correct Serializer taking gRPC message types into account is consistently used.

There was a bug which made it so that an @ResetHandler annotated method without any parameters was included for validation if a component could handle a specific type of event. This exact validation is used to filter out events from the event stream to optimize the entire stream. The optimization was thus mitigated by the simple fact of introducing a default @ResetHandler. The problem was marked by @kad-hesseg (for which thanks) and resolved in pull request #1597.

A new SnapshotTriggerDefinition called AggregateLoadTimeSnapShotTriggerDefinition has been introduced, which uses the load time of an aggregate to trigger a snapshot creation.

When using an aggregate class hierarchy, @AggregateMember annotated fields present on the root would be duplicated for every class in the hierarchy which included message handling functions. This problem was traced back to the AnnotatedAggregateMetaModelFactory.AnnotatedAggregateModel which looped over an inconsistent set of classes to find these members. The issue was marked by @kad-malota and resolved in pull request #1595.

An optimization in the snapshotting process was introduced in pull request #1510. This PR ensures no unnecessary snapshots are staged in the AbstractSnapshotter by validating none have been scheduled yet. This fix will resolve potential high I.O. when snapshots are being recreated for aggregates which have a high number of events.

The assignment rules used by the EventProcessingConfigurer weren’t always taken into account as desired. This inconsistency compared to regular assignment through the @ProcessingGroup annotation has been resolved in this pull request.

Heartbeat messages between Axon Server and an Axon Framework application were already configurable, but only from the server’s side. Properties have been introduced to also enables this from the clients end, as specified further in this pull request. Enabling heartbeat messages will ensure the connection is preemptively closed if no response has been received in the configured time frame.

A persistent loop of 500 ms was spotted during event consumption from Axon Server. Credits go to Damir Murat who has spotted the issue. With his help the issue was found quickly and eventually resolved in pull request #1484.

A serialization issue was found when working with the ConfigToken and de-/serialize it through the JacksonSerializer. This problem was uncovered in issue #1482 and resolved in pull request #1485.

The introduction of the AxonServer Connector for Java to simplify the framework’s integration with Axon Server introduced some configuration issues. For example, the AxonServerConfiguration#isForceReadFromLeader wasn’t used when opening an event stream (resolved in PR #1488).

Furthermore, properties like the max-message-size, gRPC keep alive settings and processorNotificationRate weren’t used when forming a connection with Axon Server. This issue was covered by pull request #1487.

The TrackingEventProcessor#mergeSegment(int) method was invoked with the high segment number of the pair to merge,

A small connectivity adjustment which was performed in the AxonServerConnectionManager for bug release 4.3.4 has been reverted.

Whilst adjusting the JdbcEventStorageEngine in #1187 to allow more flexibility to configure the used statements, we accidentally dropped support for adjusting how the store wrote timestamps.

Snapshots were incorrectly created in the same phase as the publication of events.

When using the SequenceEventStorageEngine to merge an active and historic event stream there was a discrepancy when the active stream didn’t contain any events and the historic stream did.

When using the QueryGateway, it was not possible to provide a QueryMessage as the query field since the queryName would be derived from the class name of the provided query.

There was a window of opportunity where the Snapshotter would publish the last event in its stream twice.

The bi-directional stream created by the Axon Server Connector wasn’t always closed correctly; specifically in error cases.

Through the new Create-or-Update

The graceful shutdown process introduced in 4.3 had a couple of minor problems.

The AggregateCreationPolicy#ALWAYS did not behave as expected, resulting in faulty behaviour when used.

In a distributed setup, the DisruptorCommandBus was not always correctly identified as being the local segment.

As described in #1274,

An issue was solved where the JdbcAutoConfiguration unintentionally depended on a JPA specific class.

A one-to-many Upcaster instance tied to Axon Server would only use the first event result and ignore the rest.

The axon-legacy module’s GapAwareTrackingToken did not implement the TrackingToken interface.

The builders of the ExponentialBackOffIntervalRetryScheduler and IntervalRetryScheduler previously

A dependency on XStream was enforced undesirably through the Builder pattern introduced in 4.0.

Due to a hierarchy issue in the Spring Boot auto configuration, the JdbcTokenStore was not always used as expected.

The ordering of message handling functions was incorrect according to the documentation.

Query Dispatch Interceptors were not called correctly when a subscription query was performed when Axon Server was used as the QueryBus.

When Axon Server was (auto) configured without being able to connect to an actual instance, processing instructions were incorrectly dispatched regardless.

The Spring Boot auto configuration did not allow the exclusion of the axon-server-connector dependency due to a direct dependency on classes.

The JpaEventStorageEngine was not wrapping the appendEvents operation in a transaction.

Deserialization failures were accidentally swallowed by the command and query gateway (marked under #967).

Resolved an issue where custom exception in a Command Handling constructor caused NullPointerExceptions.

The SimpleQueryBus reported exceptions on the initial result incorrectly upon performing a subscription query.

Resolved issue where the the "download Axon Server" message was shown upon a reconnect of an application to a Axon Server node.

Large global index gaps between events caused issues when querying the event stream (described here).

Fixed inconsistency in the GlobalSequenceTrackingToken#covers(TrackingToken) method.

A timeout was thrown instead of a exception by Axon Server when a duplicate aggregate id was created, which is resolved in #903.

Command or Query handling exceptions were not properly serialized through Axon Server (resolved in #904).

Resolved QueryUpdateEmitter configuration for the Axon Server connector set up (see issue here).

For migration purposes legacy TrackingTokens should have been added, which is resolved here.

Event Processing was stopped after a reconnection with Axon Server. Resolve the problem in issue #883.