On 7/18/2012 8:49 AM, Jeff Trawick wrote:
> On Thu, Jun 14, 2012 at 1:49 PM, Claudio Caldato (MS OPEN TECH)
> <claud...@microsoft.com> wrote:
>> I have a dev available now so we are ready to start.
>>
>> I have been out of the loop for a while so I’ll start by looking at
>> https://issues.apache.org/bugzilla/show_bug.cgi?id=52476, please point me to
>> any other relevant information. And please add ‘Windows’ to the subject line
>> for any Windows related issue so it will be easier for me to know which
>> messages I have to read.
>>
>> I hope to have good news soon.
>
> Random question from the crowd...
>
> The WinNT MPM doesn't load the address of AcceptEx dynamically via
> WSAIoctl, though that is a documented requirement and shown in the
> AcceptEx sample I found.
>
> Is it possible that this results in flaky behavior depending on
> third-party software installed?
That is entirely possible but unlikely. The present flakey behavior is not
surrounding AcceptEx(), it works for the vast majority of configurations
until incomplete socket providers are added. But I'd be happy to see a patch
to resolve AcceptEx dynamically.
The issues now are twofold;
1. When using AcceptEx() we always attempt to recycle the socket context.
We should make this configurable and avoidable. Many socket providers
have implemented AcceptEx() and then trash the socket making it fail
any attempt at reuse.
2. When using accept() we are doing 'something wrong' compared to our 2.2
code, which causes the timeouts to be other than expected by APR. This
in turn causes the mod_ssl blocking read to fail to block.
