client.close() would be useful like request.close()

This program hangs, specifically on HTTP/2 connections unless you uncomment the lines. I think the second line is the one specific to HTTP/2.

    //client.dispatcher().executorService().shutdown();
    //client.connectionPool().evictAll();

Non-daemon threads

"OkHttp graph.facebook.com" #15 prio=5 os_prio=31 tid=0x00007fbf13d3a000 nid=0x6603 runnable [0x0000700001961000]
"OkHttp Dispatcher" prio=5 tid=0x00007fdda41e5000 nid=0x4a03 waiting on condition [0x0000700000f43000]

I suggest updating the guides, but client.close() seems to make a lot of sense.

package okhttp3;

import java.io.IOException;
import java.util.Arrays;
import java.util.concurrent.CountDownLatch;
import java.util.logging.ConsoleHandler;
import java.util.logging.Level;
import java.util.logging.LogRecord;
import java.util.logging.Logger;
import java.util.logging.SimpleFormatter;
import okhttp3.internal.framed.Http2;

public class TestAsync {
  private static final int CALLS = 1;

  public static void main(String[] args) throws InterruptedException, IOException {
    Logger activeLogger = addFrameLogging();
    final CountDownLatch l = new CountDownLatch(CALLS);

    OkHttpClient client =
        new OkHttpClient.Builder().protocols(Arrays.asList(Protocol.HTTP_2, Protocol.HTTP_1_1))
            .build();

    Request robotsReq = new Request.Builder().url("https://graph.facebook.com/robots.txt").build();

    Callback callback = new Callback() {
      @Override public void onFailure(Call call, IOException e) {
        l.countDown();
        e.printStackTrace();
      }

      @Override public void onResponse(Call call, Response response) throws IOException {
        processResponse(response);
        l.countDown();
      }
    };

    //client.newCall(robotsReq).execute();

    for (int i = 0; i < CALLS; i++) {
      client.newCall(robotsReq).enqueue(callback);
    }

    l.await();

    System.out.println("Finished");

    //client.dispatcher().executorService().shutdown();
    //client.connectionPool().evictAll();
  }

  private static void processResponse(Response response) {
    ResponseBody body = response.body();

    try {
      String bs = body.string();
      System.out.println(bs.substring(Math.max(0, bs.length() - 10)));
    } catch (IOException e) {
      e.printStackTrace();
    } finally {
      body.close();
    }
  }

  public static Logger addFrameLogging() {
    Logger activeLogger = Logger.getLogger(Http2.class.getName() + "$FrameLogger");
    activeLogger.setLevel(Level.FINE);
    ConsoleHandler handler = new ConsoleHandler();
    handler.setLevel(Level.FINE);
    handler.setFormatter(new SimpleFormatter() {
      @Override
      public String format(LogRecord record) {
        return String.format("%s%n", record.getMessage());
      }
    });
    activeLogger.addHandler(handler);
    return activeLogger;
  }
}

[swankjesse]

Adding documentation is a simple, flexible solution.

[karussell]

I’m currently stumbling over a strange, but easy to reproduce issue, where connectionPool.evictAll seems to fix it.

Is it still expected behaviour that the user of OkHttpClient has to call this in certain situation although we do not create a shared connection pool like in #3160? (We are using okhttp 3.8.0 but 3.11.0 does not fix it too)

And do you know why adding the same code as shutdownhook does not help?