Spring Integration NOOP Observability Performance Test

Performance test project demonstrating a notable performance improvement in Spring Integration messaging when NOOP observability overhead is eliminated.

Project Overview

This project validates a performance optimization for Spring Integration's MessageProducerSupport.sendMessage() method. Currently, observability infrastructure is invoked even when ObservationRegistry.NOOP is configured, creating unnecessary overhead.

Our fork implementation adds a simple check:

if (this.observationRegistry.isNoop()) {
    this.messagingTemplate.send(getRequiredOutputChannel(), trackMessageIfAny(message));  // Direct send
} else {
    // Existing observability path
    IntegrationObservation.HANDLER.observation(
                this.observationConvention,
                DefaultMessageReceiverObservationConvention.INSTANCE,
                () -> new MessageReceiverContext(message, getComponentName(), "message-producer"),
                this.observationRegistry)
            .observe(() -> this.messagingTemplate.send(getRequiredOutputChannel(), trackMessageIfAny(message)));
}

Performance Results

Latest Results (10-minute test duration with 1-minute warmup):

Configuration	Throughput	Improvement
Standard Spring Integration 6.5.1	6,579,301 msg/s	-
Optimized (observability bypass)	7,793,131 msg/s	+18.4%

Performance Gain: 1,213,830 additional messages per second

Setup

First, you need to locally install the Spring Integration fork with the optimization:

# Clone the fork (as submodule)
git submodule update --init --recursive

# Build and install locally
cd spring-integration-fork
./gradlew publishToMavenLocal

This installs spring-integration-core:6.5.1.TEST-SNAPSHOT to your local Maven repository.

Running the Test

The project includes a single, focused performance test that demonstrates the observability overhead:

# Test standard Spring Integration 6.5.1
mvn test -Dtest=DirectPerformanceTest

# Test optimized version (requires fork setup below)  
mvn test -Dtest=DirectPerformanceTest -Poptimized

Test Configuration: 1-minute warmup + 10-minute (long-ish times for more statistical reliability)