Fault Tolerance 6.7.0

Today, we announce the release of SmallRye Fault Tolerance 6.7.0. This release contains several big improvements and refactorings.

Big Internal Refactoring

The internal implementations of all fault tolerance strategies up until now had 2 copies: one synchronous and one asynchronous (based on CompletionStage). This changes now: all strategies are unified on an asynchronous implementation (based on an internal async type), with a few adjustments for sychronous invocations to support thread interrupts.

This has been done especially for the next feature, which is…​

New Programmatic API

The first version of the programmatic API (FaultTolerance, @ApplyFaultTolerance) was deprecated and scheduled for removal in SmallRye Fault Tolerance 7.0. The second version (Guard, TypedGuard, @ApplyGuard) is fairly similar, but has some important differences too.

Comparison of FaultTolerance, Guard and TypedGuard:

FaultTolerance TypedGuard Guard

typed

typed

not typed

allows fallback

allows fallback

does not allow fallback

allows casting, but only if there is no fallback

does not allow casting

no casting is necessary, target type has to be specified for each call

does not allow using synchronous FaultTolerance to guard asynchonous actions et vice versa

does not allow casting

allows guarding any type

created using create() or createAsync()

only create(), but it needs the target type (which may be sync or async)

only create(); the call() method needs the target type (which may be sync or async)

Comparison of @ApplyFaultTolerance and @ApplyGuard:

@ApplyFaultTolerance @ApplyGuard

the value() points to a bean of type FaultTolerance

the value() points to a bean of type Guard or TypedGuard

ignores all other fault tolerance annotations

supports @Fallback and thread offload (@Asynchronous, @AsynchronousNonBlocking)

does not support Kotlin suspend functions

supports Kotlin suspend functions

For more information about migrating to the new programmatic API, see the migration guides:

Further, the MutinyFaultTolerance API has no replacement. The Guard and TypedGuard APIs support Uni out of the box, if the Mutiny support library is present.

New Configuration Properties

Support for new configuration properties was added. The existing, MicroProfile Fault Tolerance-defined configuration still works, but the new configuration is properly scoped and looks more natural:

Existing config property New config property

<classname>/<methodname>/<annotation>/<member>

smallrye.faulttolerance."<classname>/<methodname>".<annotation>.<member>

<classname>/<annotation>/<member>

smallrye.faulttolerance."<classname>".<annotation>.<member>

<annotation>/<member>

smallrye.faulttolerance.global.<annotation>.<member>

MP_Fault_Tolerance_NonFallback_Enabled

smallrye.faulttolerance.enabled

MP_Fault_Tolerance_Metrics_Enabled

smallrye.faulttolerance.metrics.enabled

All the <annotation> and <member> parts are changed from camel case (BeforeRetry, methodName) to kebab case (before-retry, method-name). Two annotation members are special cased to improve consistency:

  • Retry/durationUnit moves to retry.max-duration-unit

  • Retry/jitterDelayUnit moves to retry.jitter-unit

These new configuration properties exist especially for Quarkus, where they will look Quarkus-native: instead of the smallrye.faulttolerance prefix, they will start with quarkus.fault-tolerance.

Conclusion

Nothing should change externally, this release should be fully backward compatible. But please do pay extra attention.

And as usual, if you have any ideas for improvements, please file an issue!

SmallRye is open source under Apache Software License 2.0.

This website was built with Jekyll is hosted on Github Pages and is completely open source. If you want to make it better, fork the website and show us what you’ve got.

Navigation
Contribute
Get Help
External Links
CC by 3.0 | Red Hat Privacy Policy a Red Hat sponsored project