I see AbstractAspectJAdvice#calculateArgumentBindings is synchronized, but if first time the Advice method calculates the argument binding, the next time it can return directly without lock, I think it can prevent some lock competition.

such as:

public final void calculateArgumentBindings() {
    // The simple case... nothing to bind.
    if (this.argumentsIntrospected || this.parameterTypes.length == 0) {
        return;
    }
    synchronized(this) {
      if (this.argumentsIntrospected || this.parameterTypes.length == 0) {
        return;
      }
      int numUnboundArgs = this.parameterTypes.length;
      Class<?>[] parameterTypes = this.aspectJAdviceMethod.getParameterTypes();
      if (maybeBindJoinPoint(parameterTypes[0]) || maybeBindProceedingJoinPoint(parameterTypes[0])) {
        numUnboundArgs--;
      }
      else if (maybeBindJoinPointStaticPart(parameterTypes[0])) {
        numUnboundArgs--;
      }

      if (numUnboundArgs > 0) {
      // need to bind arguments by name as returned from the pointcut match
        bindArgumentsByName(numUnboundArgs);
      }

       this.argumentsIntrospected = true;
    }       
}

Comment From: leexgone

We had the same problem！ This led to serious performance issues in our system. I had to modify the source code locally to temporarily fix it. We hope this problem can be fixed as soon as possible.

Comment From: sbrannen

  • superseded by #28783

Comment From: jhoeller

Reopened for a simpler approach as discussed in my closing comment on #28783: We should be able to drop the synchronization completely since ReflectiveAspectJAdvisorFactory always performs an early call to calculateArgumentBindings anyway, covering our standard scenario. I'll resolve this in time for the 5.3.22 release.

Comment From: leexgone

Reopened for a simpler approach as discussed in my closing comment on #28783: We should be able to drop the synchronization completely since ReflectiveAspectJAdvisorFactory always performs an early call to calculateArgumentBindings anyway, covering our standard scenario. I'll resolve this in time for the 5.3.22 release.

That's a great idea! I'm look forward to the new version.