On Mon, 2010-07-19 at 16:04 +0200, DELHOSTE Fabrice wrote:
> Thanks Oleg and Asankha.
>
> So, as you known, in httpclientasync, I changed the dispatcher to SSL and
> after the first request (or a while), it gets stuck not publishing the next
> request.
>
> After analyzing and comparing HTTP vs HTTPS logs, actually, I may have found
> my mistake. Changing to SSL dispatcher seems not enough; I must also set the
> SSL session to the OP_WRITE event in addition to the regular session (in
> HttpExchangeImpl.requestCompleted):
>
> private synchronized void requestCompleted(final ManagedIOSession session) {
> this.managedSession = session;
> IOSession iosession = session.getSession();
> iosession.setAttribute(InternalRequestExecutionHandler.HTTP_EXCHANGE,
> this);
> iosession.setEvent(SelectionKey.OP_WRITE);
> SSLIOSession sslsession = (SSLIOSession)
> iosession.getAttribute("SSL_SESSION");
> if (sslsession != null) {
> sslsession.setEvent(SelectionKey.OP_WRITE);
> }
> }
>
> I guess that comes probably from the way SSLIOSession decorates the iosession.
>
Fabrice,
The protocol code should always be using the SSL i/o session instead of
the underlying plain i/o session, when available. You may want to
consider tweaking BasicManagedIOSession class to return the instance of
the SSL i/o session in case of a secure route.
> Am I right? Is it the correct way to do it?
>
> FYI, to achieve SSL as well as HTTP in httpclientasync, I actually made
> InternalClientEventDispatch a composite of DefaultClientIOEventDispatch and
> SSLClientIOEventDispatch, forwarding to the right one depending on the scheme
> of the target host.
>
Please consider contributing it back to the project.
Cheers
Oleg
> Cheers,
> Fabrice
>
> -----Original Message-----
> From: Asankha Perera [mailto:[email protected]] On Behalf Of Asankha
> C. Perera
> Sent: vendredi 16 juillet 2010 21:57
> To: HttpComponents Project
> Subject: Re: HttpCore NIO & SSL behind proxy
>
> Hi Fabrice
> > Two things you should try:
> > (1) reproducing the problem with a test case. HttpCore has a number of
> > test cases for non-blocking SSL and you could use those as a starting
> > point.
> > (2) reproducing the problem with I/O session event logging on. This
> > should help find out whether I/O interests are set correctly (write
> > events suspended by mistake). Logging is simply indispensable when
> > troubleshooting concurrency issues.
> >
> You could also try the "-Djavax.net.debug=all" system property, which
> may give you a hint if the cause is SSL specific
>
> cheers
> asankha
>
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
