FunctionInvocationWrapper.wrapped flag never resets when a function invocation throws

Bug

In SimpleFunctionRegistry.FunctionInvocationWrapper.apply(), the FunctionInvocationWrapper.wrapped flag is set to true before delegating to doApply() (to prevent re-entrant wrapping). It is reset to false afterwards — but not in a finally block:

Object result = this.doApply(input, targetFunction);
targetFunction.wrapped = false; // ← skipped if doApply throws
return result;

When a function invocation throws, wrapped stays true forever. Every subsequent message finds wrapped = true and bypasses the ObservationFunctionAroundWrapper entirely.

This means ObservationFunctionAroundWrapper — and therefore the spring_cloud_function_seconds Micrometer timer — only records the very first message invocation which leads to an error. Every subsequent message bypasses the observation entirely.

Symptom

spring_cloud_function_seconds_count with an error is permanently stuck at 1 regardless of how many more messages result in an error. Confirmed with OTLP cumulative temporality (default), no pod restarts.

Expected behaviour

spring_cloud_function_seconds_count increments on every function invocation which leads to an error.

Actual behaviour

spring_cloud_function_seconds_count is always 1 after the first invocation which led to an error — further invocations are not observed.

Versions

• spring-cloud-function-context:4.3.0
• Also confirmed present in main branch (same lines 622–623 in SimpleFunctionRegistry.java)

Suggested fix

Reset this.wrapped = false after the wrapper call returns, e.g. in a finally block:

    this.wrapped = true;
    try {
        result = functionAroundWrapper.apply(input, this);
    }
    finally {
        this.wrapped = false;
    }
}

Commencez dans SimpleFunctionRegistry.java, au niveau de SimpleFunctionRegistry.FunctionInvocationWrapper.apply(), en particulier le reset enveloppant autour de doApply(). Suivez le chemin de l’erreur et l’interaction avec ObservationFunctionAroundWrapper. Le travail est considéré comme terminé lorsque les invocations échouées répétées continuent d’être observées et que spring_cloud_function_seconds_count est incrémenté pour chaque erreur.