Mailing-List: contact dev-help@hc.apache.org; run by ezmlm
Precedence: bulk
Reply-To: "HttpComponents Project" <dev@hc.apache.org>
Date: Wed, 3 Jun 2015 18:45:38 +0000 (UTC)
From: "Alan Silva (JIRA)" <jira@apache.org>
To: dev@hc.apache.org
Message-ID: <JIRA.12835142.1433356523000.3074.1433357138638@Atlassian.JIRA>
In-Reply-To: <JIRA.12835142.1433356523000@Atlassian.JIRA>
References: <JIRA.12835142.1433356523000@Atlassian.JIRA>
 <JIRA.12835142.1433356523649@arcas>
Subject: [jira] [Updated] (HTTPCLIENT-1655) HttpClient 4.4.1 sends RST
 instead of proper FIN ACK sequence when using non-persistance
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit


     [ https://issues.apache.org/jira/browse/HTTPCLIENT-1655?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Alan Silva updated HTTPCLIENT-1655:
-----------------------------------
    Description: 
I'm seeing an interesting behavior with HttpClient 4.4.1 GA (latest release as of the writing of this issue) when trying to make http requests with no keep-alive.

First, let me post my code for the case in which I try to request using keep-alive.

[Java code for keepalive test with httpclient 4.4.1 GA | ^KeepAliveTest.java]

With the code above, everything works great. All my requests are done one the same connection and we get a nice graceful close with a couple FIN ACKs at the end. The capture is here:

[Capture of keepalive test with httpclient 4.4.1 GA | ^POC_keep_alive_WITHOUT_class.pcap]

However, when I try to do the same thing without using keepalive (I set "Connection: close" in the httpclient and make sure I call close on the httpclient as well at the end of each request), I see I the client send a reset (RST) signal to the server at the end of the connection.

The code for the non keepalive test is here:

[Java code for no keepalive test with httpclient 4.4.1 GA | ^NoKeepAliveTest.java]

And its capture is here:

[Capture of no keepalive test with httpclient 4.4.1 GA  |^POC_WITHOUT_keep_alive_WITHOUT_class.pcap]

I don't understand why the test with non persistent connections isn't gracefully closing the connection as it does with the persistent use case by sending a FIN,ACK to the server. It instead chooses to send a RST. Am I doing something wrong?

I tried the same test for no keep-alive with our legacy code using the now deprecated "commons-httpclient-3.1.jar" library. I see a proper connection close there.

[Capture of no keepalive test with commons-httpclient-3.1  |^old_code_mss.pcap]

I've attached as much info as I can think of but if the wire logs or any other info is needed, I'd be more than happy to append it. 

  was:
I'm seeing an interesting behavior with HttpClient 4.4.1 GA (latest release as of the writing of this issue) when trying to make http requests with no keep-alive.

First, let me post my code for the case in which I try to request using keep-alive.

<Placeholder for keepalive code attachment link>

With the code above, everything works great. All my requests are done one the same connection and we get a nice graceful close with a couple FIN ACKs at the end. The capture is here:

[Capture of keepalive test with httpclient 4.4.1 GA | ^POC_keep_alive_WITHOUT_class.pcap]

However, when I try to do the same thing without using keepalive (I set "Connection: close" in the httpclient and make sure I call close on the httpclient as well at the end of each request), I see I the client send a reset (RST) signal to the server at the end of the connection.

The code for the non keepalive test is here:

<Placeholder for no keepalive code attachment link>

And its capture is here:

[Capture of no keepalive test with httpclient 4.4.1 GA  |^POC_WITHOUT_keep_alive_WITHOUT_class.pcap]

I don't understand why the test with non persistent connections isn't gracefully closing the connection as it does with the persistent use case by sending a FIN,ACK to the server. It instead chooses to send a RST. Am I doing something wrong?

I tried the same test for no keep-alive with our legacy code using the now deprecated "commons-httpclient-3.1.jar" library. I see a proper connection close there.

[Capture of no keepalive test with commons-httpclient-3.1  |^old_code_mss.pcap]


> HttpClient 4.4.1 sends RST instead of proper FIN ACK sequence when using non-persistance
> ----------------------------------------------------------------------------------------
>
>                 Key: HTTPCLIENT-1655
>                 URL: https://issues.apache.org/jira/browse/HTTPCLIENT-1655
>             Project: HttpComponents HttpClient
>          Issue Type: Bug
>          Components: HttpClient
>    Affects Versions: 4.4.1
>            Reporter: Alan Silva
>         Attachments: KeepAliveTest.java, NoKeepAliveTest.java, POC_WITHOUT_keep_alive_WITHOUT_class.pcap, POC_keep_alive_WITHOUT_class.pcap, old_code_mss.pcap
>
>
> I'm seeing an interesting behavior with HttpClient 4.4.1 GA (latest release as of the writing of this issue) when trying to make http requests with no keep-alive.
> First, let me post my code for the case in which I try to request using keep-alive.
> [Java code for keepalive test with httpclient 4.4.1 GA | ^KeepAliveTest.java]
> With the code above, everything works great. All my requests are done one the same connection and we get a nice graceful close with a couple FIN ACKs at the end. The capture is here:
> [Capture of keepalive test with httpclient 4.4.1 GA | ^POC_keep_alive_WITHOUT_class.pcap]
> However, when I try to do the same thing without using keepalive (I set "Connection: close" in the httpclient and make sure I call close on the httpclient as well at the end of each request), I see I the client send a reset (RST) signal to the server at the end of the connection.
> The code for the non keepalive test is here:
> [Java code for no keepalive test with httpclient 4.4.1 GA | ^NoKeepAliveTest.java]
> And its capture is here:
> [Capture of no keepalive test with httpclient 4.4.1 GA  |^POC_WITHOUT_keep_alive_WITHOUT_class.pcap]
> I don't understand why the test with non persistent connections isn't gracefully closing the connection as it does with the persistent use case by sending a FIN,ACK to the server. It instead chooses to send a RST. Am I doing something wrong?
> I tried the same test for no keep-alive with our legacy code using the now deprecated "commons-httpclient-3.1.jar" library. I see a proper connection close there.
> [Capture of no keepalive test with commons-httpclient-3.1  |^old_code_mss.pcap]
> I've attached as much info as I can think of but if the wire logs or any other info is needed, I'd be more than happy to append it. 


--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

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