Semaphore itself does not keep track of a maximum over its lifetime. Implementing a Semphore wrapper around it that keeps track of the maximum can be tricky. Here's a quick draft of such an implementation :
public final class MySemaphore {
private final Semaphore semaphore;
private final AtomicReference<MaxCounter> maxCounter = new AtomicReference<>();
public MySemaphore(int initialAvailable) {
this.semaphore = new Semaphore(initialAvailable);
maxCounter.set(new MaxCounter(initialAvailable, initialAvailable));
}
private static final class MaxCounter {
private final int value;
private final int max;
public MaxCounter(int value, int max) {
this.value = value;
this.max = max;
}
public MaxCounter increment() {
return new MaxCounter(value + 1, Math.max(value + 1, max));
}
public MaxCounter decrement() {
return new MaxCounter(value - 1, max);
}
public int getValue() {
return value;
}
public int getMax() {
return max;
}
}
public void acquire() throws InterruptedException {
semaphore.acquire();
for (;;) {
MaxCounter current = maxCounter.get();
if (maxCounter.compareAndSet(current, current.decrement())) {
return;
}
}
}
public void release() {
for (;;) {
MaxCounter current = maxCounter.get();
if (maxCounter.compareAndSet(current, current.increment())) {
break;
}
}
semaphore.release();
}
public int availablePermits() {
return maxCounter.get().getValue();
}
public int getMaximumEverAvailable() {
return maxCounter.get().getMax();
}
}
The MaxCounter may not be exactly in sync with the internally used semaphore . The internal semaphore may get a release/acquire which is handled from an external perspective as acquire/release. To every client of MySemaphore, though the behavior will be consistent. i.e. availablePermits()
will never return a value that is higher than getMaximumEverAvailable()
disclaimer : code not tested*