Serf gets stuck on ECONNREFUSED on Windows

[Bert Huijben <be...@qqmail.nl>]

                Hi,

 

On Windows serf doesn't see the difference between a failed connect and a
lingering close. This causes serf to automatically reconnect. See

http://subversion.tigris.org/issues/show_bug.cgi?id=3832 for a few more
details.

 

Reproduction:

svn ls http://127.0.0.1:23456/svn

 

On FreeBSD this gives a proper 'Connection Refused' error, but on Windows
the application just gets in a loop.

 

I found this issue because this sometimes causes the buildbot to get stuck
when the httpd instance crashes on some assertion.

 

Once I start using 1.7.0 for real use this problem will be much worse
because there is no clean way to get out of this error. (Only killing the
application works; which would include losing work for Subclipse and AnkhSVN
users).

 

Either Serf should detect the right error, or Apr should implement a real
poll handler on Windows (available since Windows Vista) which should provide
more error information.

 

                Bert

 

 

Re: Serf gets stuck on ECONNREFUSED on Windows

[Justin Erenkrantz <ju...@erenkrantz.com>]

On Tue, May 24, 2011 at 3:10 PM, Justin Erenkrantz
<ju...@erenkrantz.com> wrote:
>> Either Serf should detect the right error, or Apr should implement a real
>> poll handler on Windows (available since Windows Vista) which should provide
>> more error information.

Well...that's exactly what APR 1.4.x did - but WSAPoll() doesn't seem
to allow for reporting non-blocking connect failures.  There are
plenty of MSDN forum posts about this not working.  I managed to get a
VS2008 build env up on Windows 7.  When I connect to a non-existent
port, nothing ever gets reported back from the apr_pollset_poll()
call.  The connect() call (correctly) returns EINPROGRESS, so serf has
no way of knowing anything is wrong.

I've posted separately to the APR list - this is an APR issue and is a
regression only in APR 1.4.x+ (which added WSAPoll() support) - not a
Serf issue.  Serf is asking for all of the correct flags, but Windows
Vista+ has no way of indicating that the connect() failed via this new
interface.  (As such, this is a new issue - APR 1.3.x and earlier
wouldn't have had this issue as they use select().)

For older Windows platforms using select(), the fix I committed
earlier should still work....but, I'll still try to confirm locally.
-- justin

Re: Serf gets stuck on ECONNREFUSED on Windows

[Justin Erenkrantz <ju...@erenkrantz.com>]

See r1470 of serf - this should stop the infinite loop.

As I don't have a Windows build environment, I'd appreciate if you can
confirm if this resolves the issue.

Thanks for the report!  -- justin

On Thu, Mar 10, 2011 at 9:18 AM, Bert Huijben <be...@qqmail.nl> wrote:
>                Hi,
>
>
>
> On Windows serf doesn't see the difference between a failed connect and a
> lingering close. This causes serf to automatically reconnect. See
>
> http://subversion.tigris.org/issues/show_bug.cgi?id=3832 for a few more
> details.
>
>
>
> Reproduction:
>
> svn ls http://127.0.0.1:23456/svn
>
>
>
> On FreeBSD this gives a proper 'Connection Refused' error, but on Windows
> the application just gets in a loop.
>
>
>
> I found this issue because this sometimes causes the buildbot to get stuck
> when the httpd instance crashes on some assertion.
>
>
>
> Once I start using 1.7.0 for real use this problem will be much worse
> because there is no clean way to get out of this error. (Only killing the
> application works; which would include losing work for Subclipse and AnkhSVN
> users).
>
>
>
> Either Serf should detect the right error, or Apr should implement a real
> poll handler on Windows (available since Windows Vista) which should provide
> more error information.
>
>
>
>                Bert
>
>
>
>
>
>