* be a SynchronousQueue.
   */
  private ExecutorService threadPool;

  @Override
  public void setUp() {
    threadPool = newCachedThreadPool();
  }

  @Override
  public void tearDown() throws InterruptedException {
    threadPool.shutdown();
    assertTrue("Some worker didn't finish in time", threadPool.awaitTermination(10, SECONDS));
  }

  private static <T> int drain(

  private ImmutableList<Integer> keys;
  private ExecutorService threadPool;

  @BeforeExperiment
  void setUp() throws Exception {
    multiset = implSupplier.get();
    ImmutableList.Builder<Integer> builder = ImmutableList.builder();
    for (int i = 0; i < size; i++) {
      builder.add(i);
    }
    keys = builder.build();
    threadPool = newFixedThreadPool(threads, new ThreadFactoryBuilder().setDaemon(true).build());

  // TODO(b/113217601): This is needed for EagerContext::runner_ to use a
  // threadpool, so that we avoid the possibility of running the runner_ in the
  // threadpool of GPU event mgr, as that can trigger more callbacks to be
  // scheduled on that same threadpool, causing a deadlock in cases where the
  // caller of event_mgr->ThenExecute() blocks on the completion of the callback

        ForkJoinPool pool1 = new ForkJoinPool(2);
        ForkJoinPool pool2 = new ForkJoinPool(4);

        // Multiple threadPool calls should work
        request.threadPool(pool1);
        request.threadPool(pool2);

        assertNotNull(request);
    }

    @Test
    public void testOnConnectWithNull() {

    ExecutorService threadPool = newFixedThreadPool(nThreads);
    CountDownLatch tasksFinished = new CountDownLatch(nTasks);
    for (int i = 0; i < nTasks; i++) {
      String s = "a" + i;
      @SuppressWarnings("unused") // https://errorprone.info/bugpattern/FutureReturnValueIgnored
      Future<?> possiblyIgnoredError =
          threadPool.submit(
              () -> {

    ExecutorService threadPool = newFixedThreadPool(nThreads);
    CountDownLatch tasksFinished = new CountDownLatch(nTasks);
    for (int i = 0; i < nTasks; i++) {
      String s = "a" + i;
      @SuppressWarnings("unused") // https://errorprone.info/bugpattern/FutureReturnValueIgnored
      Future<?> possiblyIgnoredError =
          threadPool.submit(
              () -> {

@app.get("/items/")
def read_items():
    data = some_blocking_library.fetch_items()
    return data
```

In case of doubt, or by default, use regular `def` functions, those will be run in a threadpool so they don't block the event loop.

The same rules apply to dependencies.

Make sure blocking code is not run inside of `async` functions. The logic will work, but will damage the performance heavily.

 *
 * {@snippet :
 * ListenableFuture<Boolean> adminIsLoggedIn =
 *     FluentFuture.from(usersDatabase.getAdminUser())
 *         .transform(User::getId, directExecutor())
 *         .transform(ActivityService::isLoggedIn, threadPool)
 *         .catching(RpcException.class, e -> false, directExecutor());
 * }
 *
 * <h3>Alternatives</h3>
 *
 * <h4>Frameworks</h4>
 *

這裡是有關 FastAPI 內部技術細節。

如果你有相當多的技術背景（例如協程、執行緒、阻塞等），並且對 FastAPI 如何處理 `async def` 與常規 `def` 感到好奇，請繼續閱讀。

///

### 路徑操作函式 { #path-operation-functions }

當你使用 `def` 而不是 `async def` 宣告*路徑操作函式*時，該函式會在外部的執行緒池（threadpool）中執行，然後等待結果，而不是直接呼叫（因為這樣會阻塞伺服器）。

///

### Funções de operação de rota { #path-operation-functions }

Quando você declara uma *função de operação de rota* com `def` normal ao invés de `async def`, ela é rodada em uma threadpool externa que é então aguardada, ao invés de ser chamada diretamente (já que ela bloquearia o servidor).