// level:
  //
  //             ACV(l) = ceil( SCL * ACS(l) / ( sum[priority levels k] ACS(k) ) )
  //
  // bigger numbers of ACS mean more reserved concurrent requests (at the
  // expense of every other PL).
  // This field has a default value of 30.
  // +optional
  optional int32 assuredConcurrencyShares = 1;

  // `limitResponse` indicates what to do with requests that can not be executed right now

  //
  // NominalCL(i)  = ceil( ServerCL * NCS(i) / sum_ncs )
  // sum_ncs = sum[limited priority level k] NCS(k)
  //
  // Bigger numbers mean a larger nominal concurrency limit,
  // at the expense of every other Limited priority level.
  // This field has a default value of 30.
  // +optional
  optional int32 nominalConcurrencyShares = 1;

For large responses, returning a `Response` directly is much faster than returning a dictionary.

If your PR includes any changes to the Gradle Public API, it will cause the binary compatibility check to fail.
The binary compatibility check runs as a part of the broader sanity check.
The latter runs on every PR and is a prerequisite for merging.

If you run the sanity check locally with the `./gradlew sanityCheck`, you can see the binary compatibility error in the output.
It looks like the following:

```

To see how to achieve this parallelism in production see the section about [Deployment](deployment/index.md){.internal-link target=_blank}.

## `async` and `await`

    assertThat(server.receivedPingCount()).isEqualTo(0)
    assertThat(server.receivedPongCount()).isEqualTo(0)
    closeWebSockets(webSocket, server)
  }

  /**
   * Configure the websocket to send pings every 500 ms. Artificially prevent the server from
   * responding to pings. The client should give up when attempting to send its 2nd ping, at about
   * 1000 ms.
   */
  @Test
  fun unacknowledgedPingFailsConnection() {

  @ParameterizedTest
  @ArgumentsSource(ProtocolParamProvider::class)
  fun pingsTransmitted(
    protocol: Protocol,
    mockWebServer: MockWebServer,
  ) {
    setUp(protocol, mockWebServer)
    // Ping every 500 ms, starting at 500 ms.
    client =
      client.newBuilder()
        .pingInterval(Duration.ofMillis(500))
        .build()

    // Delay the response to give 1 ping enough time to be sent and replied to.

    * Although it's probably not very problematic here with the way `SQLAlchemy` works.
    * But if you added more code to the middleware that had a lot of <abbr title="input and output">I/O</abbr> waiting, it could then be problematic.
* A middleware is run for *every* request.
    * So, a connection will be created for every request.

If you have an API that does a comparable amount of computations every time and you have a lot of clients, then the **CPU utilization** will probably *also be stable* (instead of constantly going up and down quickly).

### Examples of Replication Tools and Strategies

    Settings().apply {
      // Flow control was designed more for servers, or proxies than edge clients. If we are a client,
      // set the flow control window to 16MiB.  This avoids thrashing window updates every 64KiB, yet
      // small enough to avoid blowing up the heap.
      if (builder.client) {
        set(Settings.INITIAL_WINDOW_SIZE, OKHTTP_CLIENT_WINDOW_SIZE)
      }
    }

  /**