Mailing-List: contact httpclient-users-help@hc.apache.org; run by ezmlm
Precedence: bulk
Reply-To: "HttpClient User Discussion" <httpclient-users@hc.apache.org>
Message-ID: <1453887521.6398.4.camel@apache.org>
Subject: Re: Close Hang
From: Oleg Kalnichevski <olegk@apache.org>
To: HttpClient User Discussion <httpclient-users@hc.apache.org>
Date: Wed, 27 Jan 2016 10:38:41 +0100
In-Reply-To: <56A7C962.2030207@gmail.com>
References: <56A2B5FA.4090706@gmail.com>
	 <1453735568.18234.0.camel@apache.org> <56A7A687.4080108@gmail.com>
	 <52ECC9F7-C072-400C-B0EE-362EC9869C57@ok2consulting.com>
	 <56A7C962.2030207@gmail.com>
Content-Type: text/plain; charset="UTF-8"
Mime-Version: 1.0
Content-Transfer-Encoding: 7bit

On Tue, 2016-01-26 at 13:30 -0600, Mark Johnson wrote:
> Hi Oleg,
> 
> Below is a program that reproduces the hang.  It might take a couple of 
> runs to make this happen - it depends on what thread the callback is 
> generated on.  If it's on the IO Dispatcher thread, we can close the 
> connection on that thread also and there's no problem.  If the callback 
> comes on the pool-thread the program hangs.  The stack trace of the 
> pool-thread when this happens looks like this:
> 
> Thread [pool-1-thread] (Suspended)    <--  I paused it in the debugger 
> but this is where it stays
>    owns: SessionRequestImpl (id=44)
>    waiting for: Thread (id=21)
>    Object.wait(long)
>     Thread.join(long)
>     Thread.join()
>     InternalHttpAsyncClient(ClosableHttpAsyncClientBase).close()
>     Testprog$Callback.failed(Exception)
> etc.

...

> @Override
> 
> public void cancelled() {
> 
> Thread t = Thread.currentThread();
> 
> System.out.println(t.getName() + ": " + "Request cancelled");
> 
> try {
> 
> System.out.println(t.getName() + ": " + "Closing client connection");
> 
> _httpclient.close();
> 
> } catch (Exception ex) {
> 
> System.out.println(t.getName() + ": " + ex.getMessage());
> 
> }
> 
> _latch.countDown();
> 
> }
> 

Good Lord! You are shutting down the client from a callback. Is there a
good reason for that?

Regardless I'll see if there is a way to avoid a deadlock here.

Thanks for the reproducer.

Oleg


---------------------------------------------------------------------
To unsubscribe, e-mail: httpclient-users-unsubscribe@hc.apache.org
For additional commands, e-mail: httpclient-users-help@hc.apache.org