ThreadPoolBulkhead with TimeLimiter, seeing TimeoutExceptions

[PureXYZ]

Resilience4j version: 1.7.1

Java version: 8

*Also tried with 2.2.0 & Java 17

I am using ThreadPoolBulkhead with TimeLimiter, but unexpectedly seeing TimeoutExceptions.
It appears that the TimeLimiter starts counting when the function is "submitted", not when it actually starts execution.

My simplified example:

private <T> T execute() {
    try {
      CompletableFuture<T> future = Decorators.ofSupplier(() -> callApi(...))
          .withThreadPoolBulkhead(bulkhead)
          .withTimeLimiter(timeLimiter, scheduler)
          .get().toCompletableFuture();
      return future.get();
    } catch (...) {
      ...
    }
}

Here with configuration settings:

ThreadPoolBulkheadConfig:
maxThreadPoolSize=1
coreThreadPoolSize=1
queueCapacity=100
keepAliveDuration=Duration.ofMillis(20)

TimeLimiterConfig
timeoutDuration=Duration.ofMillis(45000)

callApi(...) in this example takes 5 seconds to respond.
I have confirmed it is not the api timing out, in fact in this experiment, I have just changed it to just do Thread.sleep(5000); to confirm.

In this situation, if execute is called 10 times at the same time (concurrently). Then I get TimeoutException.
My observation is that the bulkhead will let 1 call at a time, 5s each. Then on the 10th call, it will throw TimeoutException.
Even though each execution of callApi(...) only takes 5s.

Is this behaviour expected? My expectation was that the timeout was to apply to each execution of callApi(...)
i.e. TimeoutException is thrown if callApi(...) takes longer than 45s.

[ljluestc]

import io.github.resilience4j.timelimiter.TimeLimiter;

private <T> T execute() {
    try {
        Callable<T> callApiWithTimeout = TimeLimiter.decorateFutureSupplier(
            timeLimiter, 
            () -> callApi(...));
        
        CompletableFuture<T> future = CompletableFuture.supplyAsync(callApiWithTimeout);
        
        return future.get();
    } catch (...) {
        // Handle exceptions
    }
}