// with the last element of the heap, toTrickle.
    // Since the last element of the heap is from the bottom level, we
    // optimistically fill index position with elements from lower levels,
    // moving the hole down. In most cases this reduces the number of
    // comparisons with toTrickle, but in some cases we will need to bubble it
    // all the way up again.

Lembra da documentação [Sobre HTTPS](https.md) que diz que apenas um processo pode escutar em uma combinação de porta e endereço IP em um servidor?

Isso ainda é verdade.

Então, para poder ter **vários processos** ao mesmo tempo, tem que haver um **único processo escutando em uma porta** que então transmite a comunicação para cada processo de trabalho de alguma forma.

### Memória por Processo { #memory-per-process }

> Modern versions of Python have support for **"asynchronous code"** using something called **"coroutines"**, with **`async` and `await`** syntax.

That should make more sense now. ✨

All that is what powers FastAPI (through Starlette) and what makes it have such an impressive performance.

## Very Technical Details { #very-technical-details }

/// warning

You can probably skip this.

    // with the last element of the heap, toTrickle.
    // Since the last element of the heap is from the bottom level, we
    // optimistically fill index position with elements from lower levels,
    // moving the hole down. In most cases this reduces the number of
    // comparisons with toTrickle, but in some cases we will need to bubble it
    // all the way up again.

// have been executed. Note that "execution" here refers to kernel execution /
// scheduling of copies, etc. Similar to sync execution, it doesn't guarantee
// that lower level device queues (like GPU streams) have been flushed.
//
// This call may not block for execution of ops enqueued concurrently with this
// call.
TF_CAPI_EXPORT extern void TFE_ExecutorWaitForAllPendingNodes(

  abstract ImmutableSortedSet<E> tailSetImpl(E fromElement, boolean inclusive);

  /**
   * @since 12.0
   */
  @GwtIncompatible // NavigableSet
  @Override
  public @Nullable E lower(E e) {
    return Iterators.<@Nullable E>getNext(headSet(e, false).descendingIterator(), null);
  }

  /**
   * @since 12.0
   */
  @Override
  public @Nullable E floor(E e) {

			} else if nextReg != int(reg) {
				p.errorf("incontiguous register in ARM64 register list: %s", name)
			}
			regCnt++
			nextReg = (nextReg + 1) % 32
		case sys.ARM:
			// Parse the upper and lower bounds.
			lo := p.registerNumber(tok.String())
			hi := lo
			if p.peek() == '-' {
				p.next()
				hi = p.registerNumber(p.next().String())
			}
			if hi < lo {
				lo, hi = hi, lo
			}

  abstract ImmutableSortedSet<E> tailSetImpl(E fromElement, boolean inclusive);

  /**
   * @since 12.0
   */
  @GwtIncompatible // NavigableSet
  @Override
  public @Nullable E lower(E e) {
    return Iterators.<@Nullable E>getNext(headSet(e, false).descendingIterator(), null);
  }

  /**
   * @since 12.0
   */
  @Override
  public @Nullable E floor(E e) {

   * accommodate as many threads as will ever concurrently modify the table. Using a significantly
   * higher value than you need can waste space and time, and a significantly lower value can lead
   * to thread contention. But overestimates and underestimates within an order of magnitude do not
   * usually have much noticeable impact. A value of one permits only one thread to modify the cache

  }

  private TypeToken<? extends T> getSubtypeFromLowerBounds(Class<?> subclass, Type[] lowerBounds) {
    if (lowerBounds.length > 0) {
      @SuppressWarnings("unchecked") // T's lower bound is <? extends T>
      TypeToken<? extends T> bound = (TypeToken<? extends T>) of(lowerBounds[0]);
      // Java supports only one lowerbound anyway.
      return bound.getSubtype(subclass);
    }