MAPTest (and others) failing on Windows

[Andrea Smyth <an...@iona.com>]

I was recently looking into building CXF on HP-UX and came accross a 
problem in relation to non-anonymous replyTo. I found I could work 
around this problem by configuring the http client to use keep-alive 
connections in such a case, and merged a couple of test code changes to 
address the issue http://svn.apache.org/viewvc?view=rev&revision=551872 
<http://svn.apache.org/viewvc?view=rev&revision=551872>
I had never seen this problem on other platforms before, and was under 
the impression it is specific to HP-UX (hence the conservative "fix") 
until today I am encountering similar problems on Windows.
MAPTest fails intermittently in one or more of its test cases, also with 
a SocketException as can be seen in stack trace below.
The same workaround seems to help here also, (e.g. in MAPTest.java use 
the cxf-hpux.xml instead of cxf.xml when on HP-UX OR Windows), but as I 
know for sure that this test was passing reliably until one or two weeks 
ago (can't be more precise here) I think this needs further 
investigation and have created JIRA for this 
https://issues.apache.org/jira/browse/CXF-769
Has anyone else come accross this problem and might have an idea about 
the cause?

Cheers,
Andrea.

----------------------------
IONA Technologies PLC (registered in Ireland)
Registered Number: 171387
Registered Address: The IONA Building, Shelbourne Road, Dublin 4, Ireland

Re: MAPTest (and others) failing on Windows

[Daniel Kulp <dk...@apache.org>]

Hmm....  a BUNCH of changes have gone in the last several weeks that 
could have affected this.

1) For a while, keep-alives weren't actually working at all.   We were 
not reading all the data off the wire and we never called close on the 
stream.    Thus, the connections were never notified that we were done 
with them so new connections had to be made for every request.   If keep 
alives have to be on, I'm surprised they ran during that period.  :-(

2) We fixed (1) by making sure the close method was properly called on 
all the streams.   However, after doing that, I decided to also make 
sure we read off all the data first.   

3) Server side, we now allow Jetty to decide weather to go into chunking 
mode or not.   If the response fits entirely into it's internal NIO 
buffer (think it's 8K), then Jetty won't go into chunking and will just 
set the Content-Length header.    This change MAY be the issue.

Have you tried explicitely turning keep-alives OFF on the other platforms 
(like Linux) and seeing if the problem occurs there?

The only other thing I can suggest is use something like 
wireshark/ethereal/tcpdump to grab the raw network packets and see if 
anything looks fishy.

Dan


On Thursday 05 July 2007 12:54, Andrea Smyth wrote:
> I was recently looking into building CXF on HP-UX and came accross a
> problem in relation to non-anonymous replyTo. I found I could work
> around this problem by configuring the http client to use keep-alive
> connections in such a case, and merged a couple of test code changes
> to address the issue
> http://svn.apache.org/viewvc?view=rev&revision=551872
> <http://svn.apache.org/viewvc?view=rev&revision=551872>
> I had never seen this problem on other platforms before, and was under
> the impression it is specific to HP-UX (hence the conservative "fix")
> until today I am encountering similar problems on Windows.
> MAPTest fails intermittently in one or more of its test cases, also
> with a SocketException as can be seen in stack trace below.
> The same workaround seems to help here also, (e.g. in MAPTest.java use
> the cxf-hpux.xml instead of cxf.xml when on HP-UX OR Windows), but as
> I know for sure that this test was passing reliably until one or two
> weeks ago (can't be more precise here) I think this needs further
> investigation and have created JIRA for this
> https://issues.apache.org/jira/browse/CXF-769
> Has anyone else come accross this problem and might have an idea about
> the cause?
>
> Cheers,
> Andrea.
>
> ----------------------------
> IONA Technologies PLC (registered in Ireland)
> Registered Number: 171387
> Registered Address: The IONA Building, Shelbourne Road, Dublin 4,
> Ireland

-- 
J. Daniel Kulp
Principal Engineer
IONA
P: 781-902-8727    C: 508-380-7194
daniel.kulp@iona.com
http://www.dankulp.com/blog