59

Is there a way to specify a timeout for the whole execution of HttpClient?

I have tried the following:

httpClient.getParams().setParameter("http.socket.timeout", timeout * 1000);
httpClient.getParams().setParameter("http.connection.timeout", timeout * 1000);
httpClient.getParams().setParameter("http.connection-manager.timeout", new Long(timeout * 1000));
httpClient.getParams().setParameter("http.protocol.head-body-timeout", timeout * 1000);

It actually works fine, except if a remote host sends back data - even at one byte/second - it will continue to read forever! But I want to interrupt the connection in 10 seconds max, whether or not the host responds.

Christophe Roussy
  • 16,299
  • 4
  • 85
  • 85
Rosty Kerei
  • 1,014
  • 1
  • 9
  • 16

7 Answers7

64

For a newer version of httpclient (e.g. http components 4.3 - https://hc.apache.org/httpcomponents-client-4.3.x/index.html):

int CONNECTION_TIMEOUT_MS = timeoutSeconds * 1000; // Timeout in millis.
RequestConfig requestConfig = RequestConfig.custom()
    .setConnectionRequestTimeout(CONNECTION_TIMEOUT_MS)
    .setConnectTimeout(CONNECTION_TIMEOUT_MS)
    .setSocketTimeout(CONNECTION_TIMEOUT_MS)
    .build();

HttpPost httpPost = new HttpPost(URL);
httpPost.setConfig(requestConfig);
mR_fr0g
  • 8,462
  • 7
  • 39
  • 54
RealMan
  • 947
  • 1
  • 7
  • 8
  • 9
    Added the units, I hate it when it just says `setTimeout`, should be `setTimeoutMs`, it is indeed milliseconds. – Christophe Roussy Mar 03 '16 at 09:57
  • 2
    Wouldn't this setup allow a total request lifecycle > CONNECTION_TIMEOUT_MS? You've specified the same timeout at multiple stages of the request. – Todd Freed Mar 28 '16 at 19:47
  • 28
    I don't think this is the correct answer to the original question. The documentation for the connection request timeout states. _Returns the timeout in milliseconds used when requesting a connection from the connection manager_ this is **NOT** the total time executing the requestion just to get the connection from the connection manager. Therefore if the server was returning 1 btye/s , as the OP asks this could easily exceed the connection request timeout. – mR_fr0g May 23 '16 at 12:50
  • 3
    Agree, this is not a solution for the given problem. If request keeps reading packets from the socket the request will never timeout. Question was about hard timeout. – Valchkou Apr 10 '19 at 21:39
39

There is currently no way to set a maximum request duration of that sort: basically you want to say I don't care whether or not any specific request stage times out, but the entire request must not last longer than 15 seconds (for example).

Your best bet would be to run a separate timer, and when it expires fetch the connection manager used by the HttpClient instance and shutdown the connection, which should terminate the link. Let me know if that works for you.

Femi
  • 64,273
  • 8
  • 118
  • 148
11

Timer is evil! Using timer or executor or any other mechanism which creates a thread/runnable object per request is a very bad idea. Please think wisely and don't do it. Otherwise you will quickly run into all kind of memory issues with more or less real environment. Imagine 1000 req/min means 1000 threads or workers / min. Poor GC. The solution I propose require only 1 watchdog thread and will save you resources time and nerves. Basically you do 3 steps.

  1. put request in cache.
  2. remove request from cache when complete.
  3. abort requests which are not complete within your limit.

your cache along with watchdog thread may look like this.

import org.apache.http.client.methods.*;
import java.util.*;
import java.util.concurrent.*;
import java.util.stream.*;

public class RequestCache {

private static final long expireInMillis = 300000;
private static final Map<HttpUriRequest, Long> cache = new ConcurrentHashMap<>();
private static final ScheduledExecutorService exe = Executors.newScheduledThreadPool(1);

static {
    // run clean up every N minutes
    exe.schedule(RequestCache::cleanup, 1, TimeUnit.MINUTES);
}

public static void put(HttpUriRequest request) {
    cache.put(request, System.currentTimeMillis()+expireInMillis);
}

public static void remove(HttpUriRequest request) {
    cache.remove(request);
}

private static void cleanup() {
    long now = System.currentTimeMillis();
    // find expired requests
    List<HttpUriRequest> expired = cache.entrySet().stream()
            .filter(e -> e.getValue() > now)
            .map(Map.Entry::getKey)
            .collect(Collectors.toList());

    // abort requests
    expired.forEach(r -> {
        if (!r.isAborted()) {
            r.abort();
        }
        cache.remove(r);
      });
    }
  }

and the following sudo code how to use cache

import org.apache.http.client.methods.*;

public class RequestSample {

public void processRequest() {
    HttpUriRequest req = null;
    try {
        req = createRequest();

        RequestCache.put(req);

        execute(req);

    } finally {
        RequestCache.remove(req);
    }
  }
}
Valchkou
  • 369
  • 5
  • 8
  • Sorry for late comment, If we not using TIme in each call instead using your method using just 1 schedule thread how do we return call if e.abort() is called. – Sen Mar 29 '22 at 08:32
  • http framework will handle this for you. As connection is aborted it will throw HTTP error. Similar to connection time-out or connection interrupted. – Valchkou Apr 14 '22 at 17:04
11

Works fine, as proposed by Femi. Thanks!

Timer timer = new Timer();
timer.schedule(new TimerTask() {
    public void run() {
        if(getMethod != null) {
            getMethod.abort();
        }
    }
}, timeout * 1000);
Rosty Kerei
  • 1,014
  • 1
  • 9
  • 16
  • 9
    needs a lock, to avoid NPE. getMethod can become null between the check and the call to abort! – Tony BenBrahim Jun 27 '12 at 05:44
  • @TonyBenBrahim can you show an example of how to add this lock? – Tony Chan Sep 19 '12 at 20:04
  • 2
    @Turbo: synchronized(foo){ ... getMethod=null; } synchronized(foo){ if (getMethod!=null){ getMethod.abort(); } } – Tony BenBrahim Sep 21 '12 at 01:24
  • @TonyBenBrahim thanks for that. I'm new to synchronization. The section of `getMethod=null;` is just an example of the method where the `HttpClient` is being used correct? That is, `getMethod` wouldn't deliberately get set to `null`, but might become `null` when that method exits, therefore it should be synchronized (on the same object) as the `Timer` thread. – Tony Chan Sep 28 '12 at 03:42
  • 4
    This solution is not right for prod. Keep in in mind each timer will spin a thread. 1000/req per minute will spin 1000 thread a min. This will continue to pile up until you service die due to out of memory – Valchkou Apr 10 '19 at 21:50
4

Building off the the other answers, my solution was to use a HttpRequestInterceptor to add the abort runnable to every request. Also i swapped out the Timer for a ScheduledExecutorService.

public class TimeoutInterceptor implements HttpRequestInterceptor {

private int requestTimeout = 1 * DateTimeConstants.MILLIS_PER_MINUTE;

private ScheduledExecutorService executorService = Executors.newSingleThreadScheduledExecutor();

public TimeoutInterceptor() {  }

public TimeoutInterceptor(final int requestTimeout) {
    this.requestTimeout = requestTimeout;
}

@Override
public void process(final HttpRequest request, final HttpContext context) throws HttpException, IOException {
    if (request instanceof AbstractExecutionAwareRequest) {
        final AbstractExecutionAwareRequest abortableRequest = (AbstractExecutionAwareRequest) request;
        setAbort(abortableRequest);
    } else if (request instanceof HttpRequestWrapper) {
        HttpRequestWrapper wrapper = (HttpRequestWrapper) request;
        this.process(wrapper.getOriginal(), context);
    }

}

/**
 * @param abortableRequest
 */
private void setAbort(final AbstractExecutionAwareRequest abortableRequest) {
    final SoftReference<AbstractExecutionAwareRequest> requestRef = new SoftReference<AbstractExecutionAwareRequest>(abortableRequest);

    executorService.schedule(new Runnable() {

        @Override
        public void run() {
            AbstractExecutionAwareRequest actual = requestRef.get();
            if (actual != null && !actual.isAborted()) {
                actual.abort();
            }
        }
    }, requestTimeout, TimeUnit.MILLISECONDS);

}

public void setRequestTimeout(final int requestTimeout) {
    this.requestTimeout = requestTimeout;
}
}
mR_fr0g
  • 8,462
  • 7
  • 39
  • 54
  • 1
    This is working solution but you need create a runnable per request. That quickly turns into a problem on real high volume env. 1000 req per min will result in 1000 runnables per min. Runnables will be queued up. Indeed you will run into various problems causing service failure. – Valchkou Apr 10 '19 at 21:51
2

In HttpClient 4.3 version you can use below example.. let say for 5 seconds

int timeout = 5;
RequestConfig config = RequestConfig.custom()
  .setConnectTimeout(timeout * 1000)
  .setConnectionRequestTimeout(timeout * 1000)
  .setSocketTimeout(timeout * 1000).build();
CloseableHttpClient client = 
  HttpClientBuilder.create().setDefaultRequestConfig(config).build();
HttpGet request = new HttpGet("http://localhost:8080/service"); // GET Request
response = client.execute(request);
MADHAIYAN M
  • 2,028
  • 25
  • 22
-5

That's also work:

HttpClient client = new DefaultHttpClient();
HttpConnectionParams.setConnectionTimeout(client.getParams(), timeout * 1000);
HttpConnectionParams.setSoTimeout(client.getParams(), timeout * 1000);
AlexB
  • 473
  • 2
  • 7
  • 16
  • 3
    actually, that is exactly what he does in the question so that is obviously not working for him(nor me unfortunately either). – Dean Hiller Oct 04 '13 at 18:05